rickub
bots-garden/mini-mepublic Fork
⑂ fead47a
Branches
feature/agents-file feature/ask-agent-hooks feature/ask-cli feature/declared-tools feature/display-called-hooks feature/fetch-build-in feature/genkit-v1.13.1 feature/github-mcp-in-sbx feature/go-1.27.3 feature/hooks feature/improve-init feature/interactive-prompt feature/last-question feature/llmman-provider feature/log-errors feature/lsp-builtin feature/mcp-gateway-session feature/mcp-support feature/memory-search feature/mini-mes feature/mm-init feature/open-terminal feature/prompt-flag feature/response-request feature/resume-acp feature/sbx-template-kit feature/search-in-doc feature/selectors feature/skill-discovery feature/structured-output feature/system-one feature/talk-to-another-mm feature/tasks-builtin feature/try-again feature/unload-model feature/wasm-custom-tools feature/yolo feature/zed-skill-selector fix/generation-timeout fix/plan-mode fix/too-long-completion fix/tool-order fix/word-wrap improve-mem-usage main thinking
Tags
v0.0.0 v0.0.1 v0.0.2 v0.0.3 v0.0.4 v0.1.0 v0.1.1 v0.1.2 v0.1.3 v0.1.4 v0.1.5 v0.1.6 v0.1.7 v0.1.8 v0.1.9 v0.2.0 v0.2.1 v0.2.2 v0.2.3 v0.2.4 v0.2.5 v0.2.6 v0.3.0 v0.4.0 v0.5.0 v0.6.0 v0.7.0 v0.8.0 v0.9.0 v1.0.0 View all branches → View all tags →
Commits
⬇ Clone ▾
git clone https://git.rickub.com/bots-garden/mini-me.git
git clone ssh://git@rickub.com/bots-garden/mini-me.git

Host key fingerprint (ed25519): SHA256:iycHnxEyq0Q7uyVpB7JlznP0G7JrTPXLYRcAU5CSLhc — verify it before your first connect.

Download ZIP

🛟 Updated.

k33g committed 2026-09-21T23:30:14+02:00 Browse files
fead47a parent: c26f9e8
added .dockerignore +19 -0
new file mode 100644
@@ -0,0 +1,19 @@
1+# Build context of docker-sbx/template/Dockerfile (run from the repository
2+# root). Everything `go build .` does not need stays out, so the context is a
3+# few megabytes rather than the sandboxes/ checkout and the release binaries.
4+.git
5+sandboxes
6+release
7+tmp
8+kits
9+ebook
10+blog
11+mm
12+.mm
13+**/.mm
14+.memory
15+.quality
16+.qlty
17+notes.txt
18+*.env
19+*.tar
new file mode 100644
@@ -0,0 +1,19 @@
1+# Build context of docker-sbx/template/Dockerfile (run from the repository
2+# root). Everything `go build .` does not need stays out, so the context is a
3+# few megabytes rather than the sandboxes/ checkout and the release binaries.
4+.git
5+sandboxes
6+release
7+tmp
8+kits
9+ebook
10+blog
11+mm
12+.mm
13+**/.mm
14+.memory
15+.quality
16+.qlty
17+notes.txt
18+*.env
19+*.tar
modified .github/workflows/release.yml +64 -0
@@ -97,3 +97,67 @@ jobs:
9797 release/${{ github.ref_name }}/SHA256SUMS
9898 release/${{ github.ref_name }}/README.md
9999 fail_on_unmatched_files: true
100+
101+ # The Docker Sandbox template and its two kits, on Docker Hub, at the same
102+ # version as the binaries: k33g/mm-sandbox:0.0.2 (the image, linux/amd64 +
103+ # linux/arm64), k33g/mm-tui:0.0.2 and k33g/mm-acp:0.0.2 (the kits, OCI
104+ # artifacts). The scripts are the ones a laptop runs — docker-sbx/build.sh
105+ # with PUSH=1, then docker-sbx/push-kits.sh — so a local publish and a CI
106+ # publish are the same pipeline. Runs after `release` so nothing is pushed
107+ # to the Hub when the tests failed.
108+ #
109+ # Credentials: two repository secrets, DOCKER_HANDLE (the Docker Hub
110+ # account, which is also the image owner) and DOCKER_PAT (an access token
111+ # with read/write scope), exposed to the job as environment variables.
112+ docker-sandbox:
113+ name: publish the Docker Sandbox template and kits ${{ github.ref_name }}
114+ runs-on: ubuntu-latest
115+ needs: release
116+ env:
117+ DOCKER_HANDLE: ${{ secrets.DOCKER_HANDLE }}
118+ DOCKER_PAT: ${{ secrets.DOCKER_PAT }}
119+ # The sbx release whose CLI publishes the kits (docker/sbx-releases).
120+ SBX_VERSION: v0.45.0
121+ steps:
122+ - name: Checkout
123+ uses: actions/checkout@v4
124+
125+ - name: Check the Docker Hub credentials are set
126+ run: |
127+ set -euo pipefail
128+ if [ -z "${DOCKER_HANDLE}" ] || [ -z "${DOCKER_PAT}" ]; then
129+ echo "❌ DOCKER_HANDLE and DOCKER_PAT must be set as repository secrets"
130+ exit 1
131+ fi
132+
133+ - name: Set up QEMU
134+ # The final image stage runs `mm --version` for linux/arm64 on an
135+ # amd64 runner; the Go stage itself cross-compiles without emulation.
136+ uses: docker/setup-qemu-action@v3
137+
138+ - name: Set up Docker Buildx
139+ uses: docker/setup-buildx-action@v3
140+
141+ - name: Log in to Docker Hub
142+ run: printf '%s' "${DOCKER_PAT}" | docker login --username "${DOCKER_HANDLE}" --password-stdin
143+
144+ - name: Build and push the template image
145+ run: PUSH=1 IMAGE_OWNER="${DOCKER_HANDLE}" bash ./docker-sbx/build.sh "${GITHUB_REF_NAME}"
146+
147+ - name: Install the sbx CLI
148+ # `sbx kit push` talks to the registry directly and reads the Docker
149+ # credential store; no sandbox daemon is started here.
150+ run: |
151+ set -euo pipefail
152+ mkdir -p "${RUNNER_TEMP}/sbx"
153+ curl -fsSL -o "${RUNNER_TEMP}/sbx.tgz" \
154+ "https://github.com/docker/sbx-releases/releases/download/${SBX_VERSION}/DockerSandboxes-linux-amd64.tar.gz"
155+ tar -xzf "${RUNNER_TEMP}/sbx.tgz" -C "${RUNNER_TEMP}/sbx" --strip-components=1 docker-sbx/sbx
156+ "${RUNNER_TEMP}/sbx/sbx" version
157+
158+ - name: Publish the kits
159+ run: IMAGE_OWNER="${DOCKER_HANDLE}" SBX="${RUNNER_TEMP}/sbx/sbx" bash ./docker-sbx/push-kits.sh "${GITHUB_REF_NAME}"
160+
161+ - name: Log out of Docker Hub
162+ if: always()
163+ run: docker logout || true
@@ -97,3 +97,67 @@ jobs:
97 release/${{ github.ref_name }}/SHA256SUMS97 release/${{ github.ref_name }}/SHA256SUMS
98 release/${{ github.ref_name }}/README.md98 release/${{ github.ref_name }}/README.md
99 fail_on_unmatched_files: true99 fail_on_unmatched_files: true
100+
101+ # The Docker Sandbox template and its two kits, on Docker Hub, at the same
102+ # version as the binaries: k33g/mm-sandbox:0.0.2 (the image, linux/amd64 +
103+ # linux/arm64), k33g/mm-tui:0.0.2 and k33g/mm-acp:0.0.2 (the kits, OCI
104+ # artifacts). The scripts are the ones a laptop runs — docker-sbx/build.sh
105+ # with PUSH=1, then docker-sbx/push-kits.sh — so a local publish and a CI
106+ # publish are the same pipeline. Runs after `release` so nothing is pushed
107+ # to the Hub when the tests failed.
108+ #
109+ # Credentials: two repository secrets, DOCKER_HANDLE (the Docker Hub
110+ # account, which is also the image owner) and DOCKER_PAT (an access token
111+ # with read/write scope), exposed to the job as environment variables.
112+ docker-sandbox:
113+ name: publish the Docker Sandbox template and kits ${{ github.ref_name }}
114+ runs-on: ubuntu-latest
115+ needs: release
116+ env:
117+ DOCKER_HANDLE: ${{ secrets.DOCKER_HANDLE }}
118+ DOCKER_PAT: ${{ secrets.DOCKER_PAT }}
119+ # The sbx release whose CLI publishes the kits (docker/sbx-releases).
120+ SBX_VERSION: v0.45.0
121+ steps:
122+ - name: Checkout
123+ uses: actions/checkout@v4
124+
125+ - name: Check the Docker Hub credentials are set
126+ run: |
127+ set -euo pipefail
128+ if [ -z "${DOCKER_HANDLE}" ] || [ -z "${DOCKER_PAT}" ]; then
129+ echo "❌ DOCKER_HANDLE and DOCKER_PAT must be set as repository secrets"
130+ exit 1
131+ fi
132+
133+ - name: Set up QEMU
134+ # The final image stage runs `mm --version` for linux/arm64 on an
135+ # amd64 runner; the Go stage itself cross-compiles without emulation.
136+ uses: docker/setup-qemu-action@v3
137+
138+ - name: Set up Docker Buildx
139+ uses: docker/setup-buildx-action@v3
140+
141+ - name: Log in to Docker Hub
142+ run: printf '%s' "${DOCKER_PAT}" | docker login --username "${DOCKER_HANDLE}" --password-stdin
143+
144+ - name: Build and push the template image
145+ run: PUSH=1 IMAGE_OWNER="${DOCKER_HANDLE}" bash ./docker-sbx/build.sh "${GITHUB_REF_NAME}"
146+
147+ - name: Install the sbx CLI
148+ # `sbx kit push` talks to the registry directly and reads the Docker
149+ # credential store; no sandbox daemon is started here.
150+ run: |
151+ set -euo pipefail
152+ mkdir -p "${RUNNER_TEMP}/sbx"
153+ curl -fsSL -o "${RUNNER_TEMP}/sbx.tgz" \
154+ "https://github.com/docker/sbx-releases/releases/download/${SBX_VERSION}/DockerSandboxes-linux-amd64.tar.gz"
155+ tar -xzf "${RUNNER_TEMP}/sbx.tgz" -C "${RUNNER_TEMP}/sbx" --strip-components=1 docker-sbx/sbx
156+ "${RUNNER_TEMP}/sbx/sbx" version
157+
158+ - name: Publish the kits
159+ run: IMAGE_OWNER="${DOCKER_HANDLE}" SBX="${RUNNER_TEMP}/sbx/sbx" bash ./docker-sbx/push-kits.sh "${GITHUB_REF_NAME}"
160+
161+ - name: Log out of Docker Hub
162+ if: always()
163+ run: docker logout || true
added docker-sbx/README.md +197 -0
new file mode 100644
@@ -0,0 +1,197 @@
1+# mm in a Docker Sandbox
2+
3+Everything needed to run `mm` inside a [Docker Sandbox](https://docs.docker.com/ai/sandboxes/): a **template** (the `shell` sandbox image with `mm` installed) and two **kits** that start it — one for the window mode in your terminal, one for an editor that speaks the Agent Client Protocol.
4+
5+The full guide, with the reasoning, is in [`docs/en/how-to/run-in-a-docker-sandbox.md`](../docs/en/how-to/run-in-a-docker-sandbox.md) ([français](../docs/fr/how-to/run-in-a-docker-sandbox.md)). This page is the short version.
6+
7+```
8+docker-sbx/
9+├── build.sh builds the template image, k33g/mm-sandbox:<version> (PUSH=1: multi-arch, pushed)
10+├── sync-kits.sh points the kits at one image: owner + version
11+├── push-kits.sh publishes the kits to Docker Hub as OCI artifacts (sbx kit push)
12+├── template/
13+│ ├── Dockerfile docker/sandbox-templates:shell + mm (built from this checkout)
14+│ ├── mm-tui entry point of the mm-tui kit: `mm -tui "$@"`
15+│ └── mm-acp-info entry point of the mm-acp kit: prints the editor wiring, opens a shell
16+└── kits/
17+ ├── mm-tui/spec.yaml kind: sandbox — `sbx run` opens mm's window on the workspace
18+ └── mm-acp/spec.yaml kind: sandbox — `sbx create` once, the editor runs `sbx exec -i … mm -acp`
19+```
20+
21+## 0. Use the published kits (no build)
22+
23+Every release publishes the image and the two kits to Docker Hub at mm's version:
24+
25+| On Docker Hub | What it is |
26+|---|---|
27+| `k33g/mm-sandbox:0.0.2` | The template image. You never name it yourself: each kit pulls it. |
28+| `k33g/mm-tui:0.0.2` | The window-mode kit, for `sbx run`. |
29+| `k33g/mm-acp:0.0.2` | The editor kit, for `sbx create` + `sbx exec -i … mm -acp`. |
30+
31+Nothing to build, no `--pull never`, and no setting to change: Docker Hub is an allowed kit source out of the box (`sbx settings get kit.allowedSources` shows `docker.io/`), and a reference with a `/` but no registry host is read as `docker.io/…`. Only Docker Desktop with the `sbx` command, and a model server on your machine, are needed.
32+
33+**Look before you run.** `sbx kit inspect` reads the kit from the Hub without creating anything: its image, its entry point, its arguments and their defaults, the ports it allows.
34+
35+```bash
36+sbx kit inspect k33g/mm-tui:0.0.2
37+```
38+
39+**Window mode.** From the project you want mm to work on:
40+
41+```bash
42+cd /path/to/your/project
43+sbx run k33g/mm-tui:0.0.2 . -- agent.dmr.yaml
44+```
45+
46+The first run pulls the kit and the image, creates the sandbox `mm-tui-<directory>`, mounts the current directory as its workspace and opens `mm -tui agent.dmr.yaml` in it. What follows `--` is mm's own command line, so `-- -config other.yaml` works too. Without anything after `--`, mm reads the kit's `config` argument, then `./agent.yaml` in the workspace:
47+
48+```bash
49+sbx run k33g/mm-tui:0.0.2 . --kit-arg config=agent.dmr.yaml
50+```
51+
52+Later, `sbx run --name mm-tui-<directory>` re-attaches to that sandbox; `sbx ls` lists them.
53+
54+**From an editor.** Create the sandbox once, from the project the editor will open, with the configuration file baked in:
55+
56+```bash
57+cd /path/to/your/project
58+sbx create --name mm-acp k33g/mm-acp:0.0.2 . --kit-arg config=agent.dmr.yaml
59+```
60+
61+then register it in the editor — in Zed's `settings.json`:
62+
63+```json
64+{
65+ "agent_servers": {
66+ "mini-me (sandbox)": {
67+ "type": "custom",
68+ "command": "sbx",
69+ "args": ["exec", "-i", "mm-acp", "mm", "-acp"]
70+ }
71+ }
72+}
73+```
74+
75+The editor starts `mm -acp` inside the sandbox for every conversation; `sbx exec` starts the sandbox first when it is stopped, and runs mm in the workspace with the baked-in `AGENT_CONFIG`. One sandbox per project, each with its own `--name` and its own editor entry. Section 3 has the details (another config for one session, where the logs are, why `sbx exec -i` and not `sbx run`).
76+
77+**Kit arguments.** Both kits take the same two, passed with `--kit-arg name=value` (repeatable) to `sbx run` or `sbx create`:
78+
79+| `--kit-arg` | Default | Effect |
80+|---|---|---|
81+| `config=<path>` | empty | The file mm reads (its `AGENT_CONFIG`): relative to the workspace, or absolute inside the sandbox. Empty: `./agent.yaml`, then the built-in defaults. |
82+| `tag=<version>` | the kit's own version (`0.0.2`) | Which `k33g/mm-sandbox:<version>` image to use. Normally left alone: a kit and its image share a version. |
83+
84+**Other versions.** Replace `0.0.2` with the release you want, in the kit reference only — the kit brings the matching image. A kit can also be pinned by digest, `k33g/mm-tui@sha256:…`, the digest being what `sbx kit inspect` or the Hub page shows.
85+
86+**If sbx refuses the source.** A message like `its source is not in your allowlist` means `kit.allowedSources` was narrowed on this machine; `sbx settings set kit.allowedSources '["docker.io/"]'` restores the default, or add `docker.io/k33g/` to your list.
87+
88+Sections 1–3 are the same flow with a template and kits built locally, for a work-in-progress mm.
89+
90+## 1. Build the template
91+
92+From the repository root:
93+
94+```bash
95+./docker-sbx/build.sh # TAG from release.env: v0.0.2 → k33g/mm-sandbox:0.0.2
96+./docker-sbx/build.sh v0.0.3 # → k33g/mm-sandbox:0.0.3
97+```
98+
99+The image tag is mm's tag without the `v`, the same rule as the release assets, and the script rewrites the kits' default `tag` argument to the version it built, so the kits always name the image just built. The image lands in the local Docker image store: every `sbx run` / `sbx create` below carries `--pull never` until you `docker push k33g/mm-sandbox:0.0.2`. If `sbx` runs on another machine, `EXPORT=1 ./docker-sbx/build.sh` also writes `tmp/mm-sandbox-<version>.tar`; load it there with `sbx template load`.
100+
101+## 2. Window mode in the terminal — kit `mm-tui`
102+
103+From the project you want `mm` to work on:
104+
105+```bash
106+sbx run --pull never ./docker-sbx/kits/mm-tui . -- agent.dmr.yaml
107+```
108+
109+What follows `--` reaches `mm` as its own arguments, so this runs `mm -tui agent.dmr.yaml` in the workspace. Any `mm` flag passes through the same way (`-- -config other.yaml`). Two kit arguments exist as well:
110+
111+| `--kit-arg` | Default | Effect |
112+|---|---|---|
113+| `config=<path>` | empty | The file `mm` reads when nothing follows `--` (its `AGENT_CONFIG`). Relative to the workspace, or absolute. Empty: `./agent.yaml`, then the built-in defaults. |
114+| `tag=<version>` | the version last built (`0.0.2`) | Which `k33g/mm-sandbox:<version>` image to use. |
115+
116+`sbx run --name <name>` re-attaches to a sandbox that already exists. Sessions, when enabled in the YAML, are written under the workspace's `.mm/sessions`, which is your host directory.
117+
118+## 3. From an editor (ACP) — kit `mm-acp`
119+
120+Create the sandbox once, without attaching, **from the project the editor will open**. `.` is the workspace; sbx makes it absolute. A path that does not exist on your machine makes sbx ask "The selected workspace does not exist. Would you like to create it?" — answer `N` and fix the path. The kit path must be explicit: `./`, `../` or absolute.
121+
122+```bash
123+cd /path/to/your/project
124+sbx create --pull never --name mm-acp ./docker-sbx/kits/mm-acp . \
125+ --kit-arg config=agent.dmr.yaml
126+
127+# once the image is pushed (docker push k33g/mm-sandbox:0.0.2), --pull never is not needed:
128+sbx create --name mm-acp ./docker-sbx/kits/mm-acp . \
129+ --kit-arg config=agent.dmr.yaml
130+```
131+
132+Then register it in the editor. In Zed's `settings.json`:
133+
134+```json
135+{
136+ "agent_servers": {
137+ "📦 mini-me (dmr sbx)": {
138+ "type": "custom",
139+ "command": "sbx",
140+ "args": ["exec", "-i", "mm-acp", "mm", "-acp"],
141+ },
142+ }
143+}
144+```
145+
146+Or:
147+
148+```bash
149+sbx create --pull never --name mm-acp-2 ./docker-sbx/kits/mm-acp .
150+```
151+
152+And:
153+
154+```json
155+{
156+ "agent_servers": {
157+ "😄 mini-me (dmr sbx)": {
158+ "type": "custom",
159+ "command": "sbx",
160+ "args": ["exec", "-i", "mm-acp-2", "mm", "-acp"],
161+ "env": { "AGENT_CONFIG": "/Users/k33g/kDrive/Rickub/bots-garden/mini-me/demo/agent.dmr.yaml" }
162+ },
163+ }
164+}
165+```
166+
167+
168+
169+The editor starts `mm -acp` inside the sandbox for every conversation; `sbx exec` starts the sandbox first when it is stopped, and runs `mm` in the workspace with the `AGENT_CONFIG` baked in by `--kit-arg config`. Everything the model does — commands, file edits — happens inside the sandbox, on the mounted project.
170+
171+Why `sbx exec -i` and not `sbx run`: an ACP client speaks JSON-RPC on the agent's raw stdin/stdout, and `sbx run` always allocates a terminal. Attaching to this sandbox with `sbx run --name mm-acp` is still useful: its entry point prints the wiring above with the sandbox's real name, then opens a shell.
172+
173+- One sandbox per project: the workspace is fixed at creation. Give each one its own `--name` and its own entry in the editor.
174+- Another configuration file for one session: `"args": ["exec", "-i", "-e", "AGENT_CONFIG=agent.lsp.yaml", "mm-acp", "mm", "-acp"]`.
175+- `sbx` not found by the editor: use the absolute path printed by `which sbx` as `"command"`.
176+- Logs: in Zed, `dev: open acp logs` shows `mm`'s banner, warnings and `[acp] …` trail.
177+
178+## 4. Publishing a version
179+
180+The Release workflow (`.github/workflows/release.yml`, job `docker-sandbox`) does this on every `v*` tag, after the binaries: it reads two repository secrets, `DOCKER_HANDLE` (the Docker Hub account, also the image owner) and `DOCKER_PAT` (an access token), logs in, then runs the same two scripts you can run from a laptop:
181+
182+```bash
183+docker login -u k33g # with an access token
184+PUSH=1 ./docker-sbx/build.sh v0.0.2 # buildx, linux/amd64 + linux/arm64, pushed as k33g/mm-sandbox:0.0.2
185+./docker-sbx/push-kits.sh v0.0.2 # sbx kit validate + push → k33g/mm-tui:0.0.2, k33g/mm-acp:0.0.2
186+```
187+
188+`push-kits.sh` needs the `sbx` CLI (the workflow downloads the Linux tarball from `docker/sbx-releases`); `sbx kit push` talks to the registry directly and authenticates from the Docker credential store, no sandbox daemon involved. Both scripts call `sync-kits.sh` first, so the published kits always name the image just pushed. `IMAGE_OWNER=<handle>` publishes under another account.
189+
190+## The model server stays on the host
191+
192+Inside a sandbox `localhost` is the sandbox itself. Docker Model Runner and llama-server keep running on your machine and are reached through `host.docker.internal`, which the sandbox proxy maps to your host's `localhost`:
193+
194+- `agent.dmr.yaml` already has `fallback: http://host.docker.internal:12434/engines/v1` — nothing to change.
195+- `agent.llamacpp.yaml` has no fallback: pass `--env AGENT_BASE_URL=http://host.docker.internal:8080/v1` to `sbx run` / `sbx create`, or edit its `baseUrl`.
196+
197+Both kits allow ports 12434 (Docker Model Runner) and 8080 (llama-server) on the host. Another port: `sbx policy allow network host.docker.internal:<port>`.
new file mode 100644
@@ -0,0 +1,197 @@
1+# mm in a Docker Sandbox
2+
3+Everything needed to run `mm` inside a [Docker Sandbox](https://docs.docker.com/ai/sandboxes/): a **template** (the `shell` sandbox image with `mm` installed) and two **kits** that start it — one for the window mode in your terminal, one for an editor that speaks the Agent Client Protocol.
4+
5+The full guide, with the reasoning, is in [`docs/en/how-to/run-in-a-docker-sandbox.md`](../docs/en/how-to/run-in-a-docker-sandbox.md) ([français](../docs/fr/how-to/run-in-a-docker-sandbox.md)). This page is the short version.
6+
7+```
8+docker-sbx/
9+├── build.sh builds the template image, k33g/mm-sandbox:<version> (PUSH=1: multi-arch, pushed)
10+├── sync-kits.sh points the kits at one image: owner + version
11+├── push-kits.sh publishes the kits to Docker Hub as OCI artifacts (sbx kit push)
12+├── template/
13+│ ├── Dockerfile docker/sandbox-templates:shell + mm (built from this checkout)
14+│ ├── mm-tui entry point of the mm-tui kit: `mm -tui "$@"`
15+│ └── mm-acp-info entry point of the mm-acp kit: prints the editor wiring, opens a shell
16+└── kits/
17+ ├── mm-tui/spec.yaml kind: sandbox — `sbx run` opens mm's window on the workspace
18+ └── mm-acp/spec.yaml kind: sandbox — `sbx create` once, the editor runs `sbx exec -i … mm -acp`
19+```
20+
21+## 0. Use the published kits (no build)
22+
23+Every release publishes the image and the two kits to Docker Hub at mm's version:
24+
25+| On Docker Hub | What it is |
26+|---|---|
27+| `k33g/mm-sandbox:0.0.2` | The template image. You never name it yourself: each kit pulls it. |
28+| `k33g/mm-tui:0.0.2` | The window-mode kit, for `sbx run`. |
29+| `k33g/mm-acp:0.0.2` | The editor kit, for `sbx create` + `sbx exec -i … mm -acp`. |
30+
31+Nothing to build, no `--pull never`, and no setting to change: Docker Hub is an allowed kit source out of the box (`sbx settings get kit.allowedSources` shows `docker.io/`), and a reference with a `/` but no registry host is read as `docker.io/…`. Only Docker Desktop with the `sbx` command, and a model server on your machine, are needed.
32+
33+**Look before you run.** `sbx kit inspect` reads the kit from the Hub without creating anything: its image, its entry point, its arguments and their defaults, the ports it allows.
34+
35+```bash
36+sbx kit inspect k33g/mm-tui:0.0.2
37+```
38+
39+**Window mode.** From the project you want mm to work on:
40+
41+```bash
42+cd /path/to/your/project
43+sbx run k33g/mm-tui:0.0.2 . -- agent.dmr.yaml
44+```
45+
46+The first run pulls the kit and the image, creates the sandbox `mm-tui-<directory>`, mounts the current directory as its workspace and opens `mm -tui agent.dmr.yaml` in it. What follows `--` is mm's own command line, so `-- -config other.yaml` works too. Without anything after `--`, mm reads the kit's `config` argument, then `./agent.yaml` in the workspace:
47+
48+```bash
49+sbx run k33g/mm-tui:0.0.2 . --kit-arg config=agent.dmr.yaml
50+```
51+
52+Later, `sbx run --name mm-tui-<directory>` re-attaches to that sandbox; `sbx ls` lists them.
53+
54+**From an editor.** Create the sandbox once, from the project the editor will open, with the configuration file baked in:
55+
56+```bash
57+cd /path/to/your/project
58+sbx create --name mm-acp k33g/mm-acp:0.0.2 . --kit-arg config=agent.dmr.yaml
59+```
60+
61+then register it in the editor — in Zed's `settings.json`:
62+
63+```json
64+{
65+ "agent_servers": {
66+ "mini-me (sandbox)": {
67+ "type": "custom",
68+ "command": "sbx",
69+ "args": ["exec", "-i", "mm-acp", "mm", "-acp"]
70+ }
71+ }
72+}
73+```
74+
75+The editor starts `mm -acp` inside the sandbox for every conversation; `sbx exec` starts the sandbox first when it is stopped, and runs mm in the workspace with the baked-in `AGENT_CONFIG`. One sandbox per project, each with its own `--name` and its own editor entry. Section 3 has the details (another config for one session, where the logs are, why `sbx exec -i` and not `sbx run`).
76+
77+**Kit arguments.** Both kits take the same two, passed with `--kit-arg name=value` (repeatable) to `sbx run` or `sbx create`:
78+
79+| `--kit-arg` | Default | Effect |
80+|---|---|---|
81+| `config=<path>` | empty | The file mm reads (its `AGENT_CONFIG`): relative to the workspace, or absolute inside the sandbox. Empty: `./agent.yaml`, then the built-in defaults. |
82+| `tag=<version>` | the kit's own version (`0.0.2`) | Which `k33g/mm-sandbox:<version>` image to use. Normally left alone: a kit and its image share a version. |
83+
84+**Other versions.** Replace `0.0.2` with the release you want, in the kit reference only — the kit brings the matching image. A kit can also be pinned by digest, `k33g/mm-tui@sha256:…`, the digest being what `sbx kit inspect` or the Hub page shows.
85+
86+**If sbx refuses the source.** A message like `its source is not in your allowlist` means `kit.allowedSources` was narrowed on this machine; `sbx settings set kit.allowedSources '["docker.io/"]'` restores the default, or add `docker.io/k33g/` to your list.
87+
88+Sections 1–3 are the same flow with a template and kits built locally, for a work-in-progress mm.
89+
90+## 1. Build the template
91+
92+From the repository root:
93+
94+```bash
95+./docker-sbx/build.sh # TAG from release.env: v0.0.2 → k33g/mm-sandbox:0.0.2
96+./docker-sbx/build.sh v0.0.3 # → k33g/mm-sandbox:0.0.3
97+```
98+
99+The image tag is mm's tag without the `v`, the same rule as the release assets, and the script rewrites the kits' default `tag` argument to the version it built, so the kits always name the image just built. The image lands in the local Docker image store: every `sbx run` / `sbx create` below carries `--pull never` until you `docker push k33g/mm-sandbox:0.0.2`. If `sbx` runs on another machine, `EXPORT=1 ./docker-sbx/build.sh` also writes `tmp/mm-sandbox-<version>.tar`; load it there with `sbx template load`.
100+
101+## 2. Window mode in the terminal — kit `mm-tui`
102+
103+From the project you want `mm` to work on:
104+
105+```bash
106+sbx run --pull never ./docker-sbx/kits/mm-tui . -- agent.dmr.yaml
107+```
108+
109+What follows `--` reaches `mm` as its own arguments, so this runs `mm -tui agent.dmr.yaml` in the workspace. Any `mm` flag passes through the same way (`-- -config other.yaml`). Two kit arguments exist as well:
110+
111+| `--kit-arg` | Default | Effect |
112+|---|---|---|
113+| `config=<path>` | empty | The file `mm` reads when nothing follows `--` (its `AGENT_CONFIG`). Relative to the workspace, or absolute. Empty: `./agent.yaml`, then the built-in defaults. |
114+| `tag=<version>` | the version last built (`0.0.2`) | Which `k33g/mm-sandbox:<version>` image to use. |
115+
116+`sbx run --name <name>` re-attaches to a sandbox that already exists. Sessions, when enabled in the YAML, are written under the workspace's `.mm/sessions`, which is your host directory.
117+
118+## 3. From an editor (ACP) — kit `mm-acp`
119+
120+Create the sandbox once, without attaching, **from the project the editor will open**. `.` is the workspace; sbx makes it absolute. A path that does not exist on your machine makes sbx ask "The selected workspace does not exist. Would you like to create it?" — answer `N` and fix the path. The kit path must be explicit: `./`, `../` or absolute.
121+
122+```bash
123+cd /path/to/your/project
124+sbx create --pull never --name mm-acp ./docker-sbx/kits/mm-acp . \
125+ --kit-arg config=agent.dmr.yaml
126+
127+# once the image is pushed (docker push k33g/mm-sandbox:0.0.2), --pull never is not needed:
128+sbx create --name mm-acp ./docker-sbx/kits/mm-acp . \
129+ --kit-arg config=agent.dmr.yaml
130+```
131+
132+Then register it in the editor. In Zed's `settings.json`:
133+
134+```json
135+{
136+ "agent_servers": {
137+ "📦 mini-me (dmr sbx)": {
138+ "type": "custom",
139+ "command": "sbx",
140+ "args": ["exec", "-i", "mm-acp", "mm", "-acp"],
141+ },
142+ }
143+}
144+```
145+
146+Or:
147+
148+```bash
149+sbx create --pull never --name mm-acp-2 ./docker-sbx/kits/mm-acp .
150+```
151+
152+And:
153+
154+```json
155+{
156+ "agent_servers": {
157+ "😄 mini-me (dmr sbx)": {
158+ "type": "custom",
159+ "command": "sbx",
160+ "args": ["exec", "-i", "mm-acp-2", "mm", "-acp"],
161+ "env": { "AGENT_CONFIG": "/Users/k33g/kDrive/Rickub/bots-garden/mini-me/demo/agent.dmr.yaml" }
162+ },
163+ }
164+}
165+```
166+
167+
168+
169+The editor starts `mm -acp` inside the sandbox for every conversation; `sbx exec` starts the sandbox first when it is stopped, and runs `mm` in the workspace with the `AGENT_CONFIG` baked in by `--kit-arg config`. Everything the model does — commands, file edits — happens inside the sandbox, on the mounted project.
170+
171+Why `sbx exec -i` and not `sbx run`: an ACP client speaks JSON-RPC on the agent's raw stdin/stdout, and `sbx run` always allocates a terminal. Attaching to this sandbox with `sbx run --name mm-acp` is still useful: its entry point prints the wiring above with the sandbox's real name, then opens a shell.
172+
173+- One sandbox per project: the workspace is fixed at creation. Give each one its own `--name` and its own entry in the editor.
174+- Another configuration file for one session: `"args": ["exec", "-i", "-e", "AGENT_CONFIG=agent.lsp.yaml", "mm-acp", "mm", "-acp"]`.
175+- `sbx` not found by the editor: use the absolute path printed by `which sbx` as `"command"`.
176+- Logs: in Zed, `dev: open acp logs` shows `mm`'s banner, warnings and `[acp] …` trail.
177+
178+## 4. Publishing a version
179+
180+The Release workflow (`.github/workflows/release.yml`, job `docker-sandbox`) does this on every `v*` tag, after the binaries: it reads two repository secrets, `DOCKER_HANDLE` (the Docker Hub account, also the image owner) and `DOCKER_PAT` (an access token), logs in, then runs the same two scripts you can run from a laptop:
181+
182+```bash
183+docker login -u k33g # with an access token
184+PUSH=1 ./docker-sbx/build.sh v0.0.2 # buildx, linux/amd64 + linux/arm64, pushed as k33g/mm-sandbox:0.0.2
185+./docker-sbx/push-kits.sh v0.0.2 # sbx kit validate + push → k33g/mm-tui:0.0.2, k33g/mm-acp:0.0.2
186+```
187+
188+`push-kits.sh` needs the `sbx` CLI (the workflow downloads the Linux tarball from `docker/sbx-releases`); `sbx kit push` talks to the registry directly and authenticates from the Docker credential store, no sandbox daemon involved. Both scripts call `sync-kits.sh` first, so the published kits always name the image just pushed. `IMAGE_OWNER=<handle>` publishes under another account.
189+
190+## The model server stays on the host
191+
192+Inside a sandbox `localhost` is the sandbox itself. Docker Model Runner and llama-server keep running on your machine and are reached through `host.docker.internal`, which the sandbox proxy maps to your host's `localhost`:
193+
194+- `agent.dmr.yaml` already has `fallback: http://host.docker.internal:12434/engines/v1` — nothing to change.
195+- `agent.llamacpp.yaml` has no fallback: pass `--env AGENT_BASE_URL=http://host.docker.internal:8080/v1` to `sbx run` / `sbx create`, or edit its `baseUrl`.
196+
197+Both kits allow ports 12434 (Docker Model Runner) and 8080 (llama-server) on the host. Another port: `sbx policy allow network host.docker.internal:<port>`.
added docker-sbx/build.sh +105 -0
new file mode 100755
@@ -0,0 +1,105 @@
1+#!/usr/bin/env bash
2+# Build the mm sandbox template image, <owner>/mm-sandbox:<version>, from this
3+# checkout. The image tag follows mm's own tag, minus the "v": release.env
4+# says TAG="v0.0.2", the image is k33g/mm-sandbox:0.0.2 and `mm --version`
5+# inside it reports v0.0.2 — the same rule as the release assets.
6+#
7+# Usage:
8+# ./docker-sbx/build.sh # TAG from release.env, image for this machine
9+# ./docker-sbx/build.sh v0.0.3 # override the tag for this run (what CI does)
10+# PUSH=1 ./docker-sbx/build.sh # multi-arch (linux/amd64 + linux/arm64) built
11+# # with buildx and PUSHED to the registry —
12+# # needs `docker login`; this is the CI path
13+# IMAGE_OWNER=someone ./docker-sbx/build.sh # another image owner (default k33g)
14+# PLATFORMS=linux/amd64 PUSH=1 … # narrower multi-arch set
15+# SHELL_TEMPLATE=docker/sandbox-templates:shell-docker ./docker-sbx/build.sh
16+# # base it on the Docker-in-Docker shell
17+# EXPORT=1 ./docker-sbx/build.sh # also `docker save` to tmp/, for
18+# # `sbx template load` on another host
19+#
20+# Whatever the mode, the two kits under docker-sbx/kits/ are then pointed at
21+# the image just built (docker-sbx/sync-kits.sh), so kits and image move
22+# together. Publishing the kits themselves is docker-sbx/push-kits.sh.
23+set -euo pipefail
24+
25+cd "$(dirname "$0")/.."
26+
27+# release.env carries TAG ("v0.0.2"). It is git-ignored, so a checkout without
28+# it names the tag on the command line. A tag given on the command line wins.
29+if [ -f release.env ]; then
30+ # shellcheck source=/dev/null
31+ source release.env
32+fi
33+TAG="${1:-${TAG:-}}"
34+if ! [[ "${TAG}" =~ ^v[0-9]+\.[0-9]+\.[0-9]+(-[0-9A-Za-z.-]+)?$ ]]; then
35+ echo "❌ TAG must look like v1.2.3 or v1.2.3-rc.1, got '${TAG}' (check release.env or pass it: ./docker-sbx/build.sh v1.2.3)"
36+ exit 1
37+fi
38+
39+# Not OWNER: release.env uses that name for the git owner (bots-garden), and
40+# the image owner is a different thing — the Docker Hub account, k33g.
41+IMAGE_OWNER="${IMAGE_OWNER:-k33g}"
42+VERSION="${TAG#v}"
43+IMAGE="${IMAGE_OWNER}/mm-sandbox:${VERSION}"
44+PLATFORMS="${PLATFORMS:-linux/amd64,linux/arm64}"
45+
46+build_args=(--build-arg "MM_VERSION=${TAG}")
47+if [ -n "${SHELL_TEMPLATE:-}" ]; then
48+ build_args+=(--build-arg "SHELL_TEMPLATE=${SHELL_TEMPLATE}")
49+fi
50+
51+if [ -n "${PUSH:-}" ]; then
52+ # Multi-arch needs buildx's container driver: the classic builder can
53+ # neither assemble a manifest list nor build for a foreign platform. The
54+ # Go stage cross-compiles on the build platform; only the final stage's
55+ # `RUN mm --version` runs under emulation (QEMU, set up by the caller).
56+ echo "🐳 Building and pushing ${IMAGE} for ${PLATFORMS} (mm ${TAG}, base ${SHELL_TEMPLATE:-docker/sandbox-templates:shell})"
57+ if ! docker buildx inspect mm-sandbox-builder >/dev/null 2>&1; then
58+ docker buildx create --name mm-sandbox-builder --driver docker-container --bootstrap >/dev/null
59+ fi
60+ docker buildx build --builder mm-sandbox-builder \
61+ --platform "${PLATFORMS}" \
62+ -f docker-sbx/template/Dockerfile "${build_args[@]}" \
63+ -t "${IMAGE}" --push .
64+ # The proof that ships: the registry holds one manifest per platform.
65+ docker buildx imagetools inspect "${IMAGE}" | grep -E '^Name:|Platform:' | sed 's/^/ /'
66+else
67+ echo "🐳 Building ${IMAGE} (mm ${TAG}, base ${SHELL_TEMPLATE:-docker/sandbox-templates:shell})"
68+ docker build -f docker-sbx/template/Dockerfile "${build_args[@]}" -t "${IMAGE}" .
69+
70+ # The one proof that what was built starts: --version needs no config and
71+ # no model server. It also confirms the stamp reached the binary.
72+ reported="$(docker run --rm "${IMAGE}" mm --version)"
73+ if ! grep -q -- "${TAG}" <<<"${reported}"; then
74+ echo "❌ the image does not report ${TAG}: ${reported}"
75+ exit 1
76+ fi
77+ echo "✅ ${reported}"
78+fi
79+
80+./docker-sbx/sync-kits.sh "${IMAGE_OWNER}" "${VERSION}"
81+
82+if [ -n "${EXPORT:-}" ]; then
83+ mkdir -p tmp
84+ docker save -o "tmp/mm-sandbox-${VERSION}.tar" "${IMAGE}"
85+ echo "📦 tmp/mm-sandbox-${VERSION}.tar — load it where sbx runs with: sbx template load tmp/mm-sandbox-${VERSION}.tar"
86+fi
87+
88+if [ -n "${PUSH:-}" ]; then
89+ cat <<NEXT
90+
91+Image on the registry. Publish the kits next: ./docker-sbx/push-kits.sh ${TAG}
92+NEXT
93+else
94+ cat <<NEXT
95+
96+Next, from the project you want mm to work on:
97+
98+ window mode sbx run --pull never ./docker-sbx/kits/mm-tui . -- agent.dmr.yaml
99+ editor (ACP) sbx create --pull never --name mm-acp ./docker-sbx/kits/mm-acp . --kit-arg config=agent.dmr.yaml
100+ then register sbx exec -i mm-acp mm -acp in the editor (docker-sbx/README.md)
101+
102+ --pull never because ${IMAGE} exists only in the local image store;
103+ after \`docker push ${IMAGE}\` (or PUSH=1) it is no longer needed.
104+NEXT
105+fi
new file mode 100755
@@ -0,0 +1,105 @@
1+#!/usr/bin/env bash
2+# Build the mm sandbox template image, <owner>/mm-sandbox:<version>, from this
3+# checkout. The image tag follows mm's own tag, minus the "v": release.env
4+# says TAG="v0.0.2", the image is k33g/mm-sandbox:0.0.2 and `mm --version`
5+# inside it reports v0.0.2 — the same rule as the release assets.
6+#
7+# Usage:
8+# ./docker-sbx/build.sh # TAG from release.env, image for this machine
9+# ./docker-sbx/build.sh v0.0.3 # override the tag for this run (what CI does)
10+# PUSH=1 ./docker-sbx/build.sh # multi-arch (linux/amd64 + linux/arm64) built
11+# # with buildx and PUSHED to the registry —
12+# # needs `docker login`; this is the CI path
13+# IMAGE_OWNER=someone ./docker-sbx/build.sh # another image owner (default k33g)
14+# PLATFORMS=linux/amd64 PUSH=1 … # narrower multi-arch set
15+# SHELL_TEMPLATE=docker/sandbox-templates:shell-docker ./docker-sbx/build.sh
16+# # base it on the Docker-in-Docker shell
17+# EXPORT=1 ./docker-sbx/build.sh # also `docker save` to tmp/, for
18+# # `sbx template load` on another host
19+#
20+# Whatever the mode, the two kits under docker-sbx/kits/ are then pointed at
21+# the image just built (docker-sbx/sync-kits.sh), so kits and image move
22+# together. Publishing the kits themselves is docker-sbx/push-kits.sh.
23+set -euo pipefail
24+
25+cd "$(dirname "$0")/.."
26+
27+# release.env carries TAG ("v0.0.2"). It is git-ignored, so a checkout without
28+# it names the tag on the command line. A tag given on the command line wins.
29+if [ -f release.env ]; then
30+ # shellcheck source=/dev/null
31+ source release.env
32+fi
33+TAG="${1:-${TAG:-}}"
34+if ! [[ "${TAG}" =~ ^v[0-9]+\.[0-9]+\.[0-9]+(-[0-9A-Za-z.-]+)?$ ]]; then
35+ echo "❌ TAG must look like v1.2.3 or v1.2.3-rc.1, got '${TAG}' (check release.env or pass it: ./docker-sbx/build.sh v1.2.3)"
36+ exit 1
37+fi
38+
39+# Not OWNER: release.env uses that name for the git owner (bots-garden), and
40+# the image owner is a different thing — the Docker Hub account, k33g.
41+IMAGE_OWNER="${IMAGE_OWNER:-k33g}"
42+VERSION="${TAG#v}"
43+IMAGE="${IMAGE_OWNER}/mm-sandbox:${VERSION}"
44+PLATFORMS="${PLATFORMS:-linux/amd64,linux/arm64}"
45+
46+build_args=(--build-arg "MM_VERSION=${TAG}")
47+if [ -n "${SHELL_TEMPLATE:-}" ]; then
48+ build_args+=(--build-arg "SHELL_TEMPLATE=${SHELL_TEMPLATE}")
49+fi
50+
51+if [ -n "${PUSH:-}" ]; then
52+ # Multi-arch needs buildx's container driver: the classic builder can
53+ # neither assemble a manifest list nor build for a foreign platform. The
54+ # Go stage cross-compiles on the build platform; only the final stage's
55+ # `RUN mm --version` runs under emulation (QEMU, set up by the caller).
56+ echo "🐳 Building and pushing ${IMAGE} for ${PLATFORMS} (mm ${TAG}, base ${SHELL_TEMPLATE:-docker/sandbox-templates:shell})"
57+ if ! docker buildx inspect mm-sandbox-builder >/dev/null 2>&1; then
58+ docker buildx create --name mm-sandbox-builder --driver docker-container --bootstrap >/dev/null
59+ fi
60+ docker buildx build --builder mm-sandbox-builder \
61+ --platform "${PLATFORMS}" \
62+ -f docker-sbx/template/Dockerfile "${build_args[@]}" \
63+ -t "${IMAGE}" --push .
64+ # The proof that ships: the registry holds one manifest per platform.
65+ docker buildx imagetools inspect "${IMAGE}" | grep -E '^Name:|Platform:' | sed 's/^/ /'
66+else
67+ echo "🐳 Building ${IMAGE} (mm ${TAG}, base ${SHELL_TEMPLATE:-docker/sandbox-templates:shell})"
68+ docker build -f docker-sbx/template/Dockerfile "${build_args[@]}" -t "${IMAGE}" .
69+
70+ # The one proof that what was built starts: --version needs no config and
71+ # no model server. It also confirms the stamp reached the binary.
72+ reported="$(docker run --rm "${IMAGE}" mm --version)"
73+ if ! grep -q -- "${TAG}" <<<"${reported}"; then
74+ echo "❌ the image does not report ${TAG}: ${reported}"
75+ exit 1
76+ fi
77+ echo "✅ ${reported}"
78+fi
79+
80+./docker-sbx/sync-kits.sh "${IMAGE_OWNER}" "${VERSION}"
81+
82+if [ -n "${EXPORT:-}" ]; then
83+ mkdir -p tmp
84+ docker save -o "tmp/mm-sandbox-${VERSION}.tar" "${IMAGE}"
85+ echo "📦 tmp/mm-sandbox-${VERSION}.tar — load it where sbx runs with: sbx template load tmp/mm-sandbox-${VERSION}.tar"
86+fi
87+
88+if [ -n "${PUSH:-}" ]; then
89+ cat <<NEXT
90+
91+Image on the registry. Publish the kits next: ./docker-sbx/push-kits.sh ${TAG}
92+NEXT
93+else
94+ cat <<NEXT
95+
96+Next, from the project you want mm to work on:
97+
98+ window mode sbx run --pull never ./docker-sbx/kits/mm-tui . -- agent.dmr.yaml
99+ editor (ACP) sbx create --pull never --name mm-acp ./docker-sbx/kits/mm-acp . --kit-arg config=agent.dmr.yaml
100+ then register sbx exec -i mm-acp mm -acp in the editor (docker-sbx/README.md)
101+
102+ --pull never because ${IMAGE} exists only in the local image store;
103+ after \`docker push ${IMAGE}\` (or PUSH=1) it is no longer needed.
104+NEXT
105+fi
added docker-sbx/push-kits.sh +59 -0
new file mode 100755
@@ -0,0 +1,59 @@
1+#!/usr/bin/env bash
2+# Publish the two kits to Docker Hub as OCI kit artifacts:
3+# <owner>/mm-tui:<version> and <owner>/mm-acp:<version>
4+#
5+# Usage:
6+# ./docker-sbx/push-kits.sh # TAG from release.env
7+# ./docker-sbx/push-kits.sh v0.0.3 # override the tag (what CI does)
8+# IMAGE_OWNER=someone … # another owner (default k33g); the kits
9+# # are re-pointed at someone/mm-sandbox too
10+# REGISTRY=localhost:5000 … # another registry (default docker.io)
11+# SBX=/path/to/sbx … # the sbx binary, if not on the PATH
12+#
13+# Needs a login: `docker login -u <handle>` with a Docker Hub access token.
14+# `sbx kit push` talks to the registry directly — no sandbox daemon involved —
15+# and falls back to the Docker credential store for authentication.
16+#
17+# Push the image first (PUSH=1 ./docker-sbx/build.sh): a kit names
18+# <owner>/mm-sandbox:<version>, and sbx pulls that image when the kit is used.
19+set -euo pipefail
20+
21+cd "$(dirname "$0")/.."
22+
23+if [ -f release.env ]; then
24+ # shellcheck source=/dev/null
25+ source release.env
26+fi
27+TAG="${1:-${TAG:-}}"
28+if ! [[ "${TAG}" =~ ^v[0-9]+\.[0-9]+\.[0-9]+(-[0-9A-Za-z.-]+)?$ ]]; then
29+ echo "❌ TAG must look like v1.2.3 or v1.2.3-rc.1, got '${TAG}' (check release.env or pass it: ./docker-sbx/push-kits.sh v1.2.3)"
30+ exit 1
31+fi
32+
33+IMAGE_OWNER="${IMAGE_OWNER:-k33g}"
34+REGISTRY="${REGISTRY:-docker.io}"
35+VERSION="${TAG#v}"
36+SBX="${SBX:-sbx}"
37+
38+if ! command -v "${SBX}" >/dev/null 2>&1; then
39+ echo "❌ sbx not found (${SBX}). Install Docker Sandboxes, or SBX=/path/to/sbx"
40+ exit 1
41+fi
42+
43+./docker-sbx/sync-kits.sh "${IMAGE_OWNER}" "${VERSION}"
44+
45+for dir in docker-sbx/kits/*/; do
46+ kit="$(basename "${dir}")"
47+ ref="${REGISTRY}/${IMAGE_OWNER}/${kit}:${VERSION}"
48+ "${SBX}" kit validate "${dir}"
49+ "${SBX}" kit push "${dir}" "${ref}"
50+done
51+
52+cat <<NEXT
53+
54+Published. From any machine with Docker Sandboxes (docker.io is an allowed kit source by default):
55+
56+ window mode sbx run ${IMAGE_OWNER}/mm-tui:${VERSION} . -- agent.dmr.yaml
57+ editor (ACP) sbx create --name mm-acp ${IMAGE_OWNER}/mm-acp:${VERSION} . --kit-arg config=agent.dmr.yaml
58+ then register sbx exec -i mm-acp mm -acp in the editor
59+NEXT
new file mode 100755
@@ -0,0 +1,59 @@
1+#!/usr/bin/env bash
2+# Publish the two kits to Docker Hub as OCI kit artifacts:
3+# <owner>/mm-tui:<version> and <owner>/mm-acp:<version>
4+#
5+# Usage:
6+# ./docker-sbx/push-kits.sh # TAG from release.env
7+# ./docker-sbx/push-kits.sh v0.0.3 # override the tag (what CI does)
8+# IMAGE_OWNER=someone … # another owner (default k33g); the kits
9+# # are re-pointed at someone/mm-sandbox too
10+# REGISTRY=localhost:5000 … # another registry (default docker.io)
11+# SBX=/path/to/sbx … # the sbx binary, if not on the PATH
12+#
13+# Needs a login: `docker login -u <handle>` with a Docker Hub access token.
14+# `sbx kit push` talks to the registry directly — no sandbox daemon involved —
15+# and falls back to the Docker credential store for authentication.
16+#
17+# Push the image first (PUSH=1 ./docker-sbx/build.sh): a kit names
18+# <owner>/mm-sandbox:<version>, and sbx pulls that image when the kit is used.
19+set -euo pipefail
20+
21+cd "$(dirname "$0")/.."
22+
23+if [ -f release.env ]; then
24+ # shellcheck source=/dev/null
25+ source release.env
26+fi
27+TAG="${1:-${TAG:-}}"
28+if ! [[ "${TAG}" =~ ^v[0-9]+\.[0-9]+\.[0-9]+(-[0-9A-Za-z.-]+)?$ ]]; then
29+ echo "❌ TAG must look like v1.2.3 or v1.2.3-rc.1, got '${TAG}' (check release.env or pass it: ./docker-sbx/push-kits.sh v1.2.3)"
30+ exit 1
31+fi
32+
33+IMAGE_OWNER="${IMAGE_OWNER:-k33g}"
34+REGISTRY="${REGISTRY:-docker.io}"
35+VERSION="${TAG#v}"
36+SBX="${SBX:-sbx}"
37+
38+if ! command -v "${SBX}" >/dev/null 2>&1; then
39+ echo "❌ sbx not found (${SBX}). Install Docker Sandboxes, or SBX=/path/to/sbx"
40+ exit 1
41+fi
42+
43+./docker-sbx/sync-kits.sh "${IMAGE_OWNER}" "${VERSION}"
44+
45+for dir in docker-sbx/kits/*/; do
46+ kit="$(basename "${dir}")"
47+ ref="${REGISTRY}/${IMAGE_OWNER}/${kit}:${VERSION}"
48+ "${SBX}" kit validate "${dir}"
49+ "${SBX}" kit push "${dir}" "${ref}"
50+done
51+
52+cat <<NEXT
53+
54+Published. From any machine with Docker Sandboxes (docker.io is an allowed kit source by default):
55+
56+ window mode sbx run ${IMAGE_OWNER}/mm-tui:${VERSION} . -- agent.dmr.yaml
57+ editor (ACP) sbx create --name mm-acp ${IMAGE_OWNER}/mm-acp:${VERSION} . --kit-arg config=agent.dmr.yaml
58+ then register sbx exec -i mm-acp mm -acp in the editor
59+NEXT
added docker-sbx/sync-kits.sh +37 -0
new file mode 100755
@@ -0,0 +1,37 @@
1+#!/usr/bin/env bash
2+# Point the two kits at one image: <owner>/mm-sandbox:<version>.
3+#
4+# Usage: ./docker-sbx/sync-kits.sh OWNER VERSION (e.g. k33g 0.0.2)
5+#
6+# Rewrites, in docker-sbx/kits/*/spec.yaml, only these three values:
7+# version: "<version>" the kit's own version
8+# args.tag.default: "<version>" which image tag the kit uses
9+# sandbox.image: "<owner>/mm-sandbox:…" which owner the image lives under
10+# Everything else — comments included — is left byte-for-byte alone, so a
11+# kit that already matches is not changed at all. build.sh and push-kits.sh
12+# call this; CI relies on it so the published kits name the image it pushed.
13+set -euo pipefail
14+
15+if [ $# -ne 2 ]; then
16+ echo "usage: $0 OWNER VERSION" >&2
17+ exit 2
18+fi
19+OWNER="$1"
20+VERSION="$2"
21+
22+cd "$(dirname "$0")/.."
23+
24+for kit in docker-sbx/kits/*/spec.yaml; do
25+ if ! grep -q '^ tag:$' "${kit}" || ! grep -q '^version: ' "${kit}" || ! grep -q '^ image: "' "${kit}"; then
26+ echo "❌ ${kit}: expected a top-level 'version:', an 'args.tag' and a quoted 'sandbox.image' — not updated" >&2
27+ exit 1
28+ fi
29+ # sed -i.bak works on both GNU and BSD sed; the backup is removed at once.
30+ sed -i.bak -E \
31+ -e "s|^(version: )\"[^\"]*\"|\1\"${VERSION}\"|" \
32+ -e "/^ tag:\$/,/^ default:/ s|^( default: )\"[^\"]*\"|\1\"${VERSION}\"|" \
33+ -e "s|^( image: \")[^/\"]+(/mm-sandbox:.*)|\1${OWNER}\2|" \
34+ "${kit}"
35+ rm -f "${kit}.bak"
36+ echo "✅ ${kit} → version \"${VERSION}\", tag default \"${VERSION}\", image ${OWNER}/mm-sandbox:\${{ kit.args.tag }}"
37+done
new file mode 100755
@@ -0,0 +1,37 @@
1+#!/usr/bin/env bash
2+# Point the two kits at one image: <owner>/mm-sandbox:<version>.
3+#
4+# Usage: ./docker-sbx/sync-kits.sh OWNER VERSION (e.g. k33g 0.0.2)
5+#
6+# Rewrites, in docker-sbx/kits/*/spec.yaml, only these three values:
7+# version: "<version>" the kit's own version
8+# args.tag.default: "<version>" which image tag the kit uses
9+# sandbox.image: "<owner>/mm-sandbox:…" which owner the image lives under
10+# Everything else — comments included — is left byte-for-byte alone, so a
11+# kit that already matches is not changed at all. build.sh and push-kits.sh
12+# call this; CI relies on it so the published kits name the image it pushed.
13+set -euo pipefail
14+
15+if [ $# -ne 2 ]; then
16+ echo "usage: $0 OWNER VERSION" >&2
17+ exit 2
18+fi
19+OWNER="$1"
20+VERSION="$2"
21+
22+cd "$(dirname "$0")/.."
23+
24+for kit in docker-sbx/kits/*/spec.yaml; do
25+ if ! grep -q '^ tag:$' "${kit}" || ! grep -q '^version: ' "${kit}" || ! grep -q '^ image: "' "${kit}"; then
26+ echo "❌ ${kit}: expected a top-level 'version:', an 'args.tag' and a quoted 'sandbox.image' — not updated" >&2
27+ exit 1
28+ fi
29+ # sed -i.bak works on both GNU and BSD sed; the backup is removed at once.
30+ sed -i.bak -E \
31+ -e "s|^(version: )\"[^\"]*\"|\1\"${VERSION}\"|" \
32+ -e "/^ tag:\$/,/^ default:/ s|^( default: )\"[^\"]*\"|\1\"${VERSION}\"|" \
33+ -e "s|^( image: \")[^/\"]+(/mm-sandbox:.*)|\1${OWNER}\2|" \
34+ "${kit}"
35+ rm -f "${kit}.bak"
36+ echo "✅ ${kit} → version \"${VERSION}\", tag default \"${VERSION}\", image ${OWNER}/mm-sandbox:\${{ kit.args.tag }}"
37+done
added docker-sbx/template/Dockerfile +53 -0
new file mode 100644
@@ -0,0 +1,53 @@
1+# syntax=docker/dockerfile:1
2+#
3+# The mm sandbox template: Docker Sandboxes' `shell` image with mm installed.
4+#
5+# Build it from the REPOSITORY ROOT (the context is the mm sources), or run
6+# docker-sbx/build.sh, which does exactly that and stamps mm's tag — the
7+# image tag is that tag without the "v", k33g/mm-sandbox:0.0.2 for v0.0.2:
8+#
9+# docker build -f docker-sbx/template/Dockerfile \
10+# --build-arg MM_VERSION=v0.0.2 -t k33g/mm-sandbox:0.0.2 .
11+#
12+# The two kits under docker-sbx/kits/ name this image.
13+# ENTRYPOINT and CMD are deliberately not set: the sandbox runtime keeps the
14+# shell template's tini entrypoint and replaces CMD with its own, and the kit's
15+# `sandbox.entrypoint` is what starts mm when you attach.
16+
17+ARG GO_VERSION=1.26
18+ARG SHELL_TEMPLATE=docker/sandbox-templates:shell
19+
20+# ---- build stage: cross-compile mm for the target platform -----------------
21+FROM --platform=$BUILDPLATFORM golang:${GO_VERSION}-bookworm AS build
22+ARG TARGETOS
23+ARG TARGETARCH
24+# What `mm --version` reports: the tag, as 02-build-releases.sh stamps it.
25+ARG MM_VERSION=dev
26+
27+WORKDIR /src
28+COPY go.mod go.sum ./
29+RUN --mount=type=cache,target=/go/pkg/mod go mod download
30+COPY . .
31+# Same flags as 02-build-releases.sh: no C, no local paths, no symbols, the
32+# version stamped into internal/version.
33+RUN --mount=type=cache,target=/go/pkg/mod \
34+ --mount=type=cache,target=/root/.cache/go-build \
35+ CGO_ENABLED=0 GOOS=${TARGETOS} GOARCH=${TARGETARCH} \
36+ go build -trimpath -ldflags "-s -w -X mm/internal/version.stamped=${MM_VERSION}" -o /out/mm .
37+
38+# ---- template: the shell sandbox plus mm -----------------------------------
39+FROM ${SHELL_TEMPLATE}
40+
41+COPY --from=build --chmod=0755 /out/mm /usr/local/bin/mm
42+# The two kit entry points. mm-tui runs `mm -tui "$@"`; mm-acp-info prints how
43+# an editor reaches this sandbox and opens a shell (see docker-sbx/README.md).
44+COPY --chmod=0755 docker-sbx/template/mm-tui docker-sbx/template/mm-acp-info /usr/local/bin/
45+
46+# tcell (the window mode) needs a terminal type and a UTF-8 locale; the kits
47+# set them too, this is the fallback for a plain `docker run`.
48+ENV TERM=xterm-256color \
49+ COLORTERM=truecolor \
50+ LANG=C.UTF-8
51+
52+# Prove the binary runs on this platform before the image is tagged.
53+RUN mm --version && mm-tui --version
new file mode 100644
@@ -0,0 +1,53 @@
1+# syntax=docker/dockerfile:1
2+#
3+# The mm sandbox template: Docker Sandboxes' `shell` image with mm installed.
4+#
5+# Build it from the REPOSITORY ROOT (the context is the mm sources), or run
6+# docker-sbx/build.sh, which does exactly that and stamps mm's tag — the
7+# image tag is that tag without the "v", k33g/mm-sandbox:0.0.2 for v0.0.2:
8+#
9+# docker build -f docker-sbx/template/Dockerfile \
10+# --build-arg MM_VERSION=v0.0.2 -t k33g/mm-sandbox:0.0.2 .
11+#
12+# The two kits under docker-sbx/kits/ name this image.
13+# ENTRYPOINT and CMD are deliberately not set: the sandbox runtime keeps the
14+# shell template's tini entrypoint and replaces CMD with its own, and the kit's
15+# `sandbox.entrypoint` is what starts mm when you attach.
16+
17+ARG GO_VERSION=1.26
18+ARG SHELL_TEMPLATE=docker/sandbox-templates:shell
19+
20+# ---- build stage: cross-compile mm for the target platform -----------------
21+FROM --platform=$BUILDPLATFORM golang:${GO_VERSION}-bookworm AS build
22+ARG TARGETOS
23+ARG TARGETARCH
24+# What `mm --version` reports: the tag, as 02-build-releases.sh stamps it.
25+ARG MM_VERSION=dev
26+
27+WORKDIR /src
28+COPY go.mod go.sum ./
29+RUN --mount=type=cache,target=/go/pkg/mod go mod download
30+COPY . .
31+# Same flags as 02-build-releases.sh: no C, no local paths, no symbols, the
32+# version stamped into internal/version.
33+RUN --mount=type=cache,target=/go/pkg/mod \
34+ --mount=type=cache,target=/root/.cache/go-build \
35+ CGO_ENABLED=0 GOOS=${TARGETOS} GOARCH=${TARGETARCH} \
36+ go build -trimpath -ldflags "-s -w -X mm/internal/version.stamped=${MM_VERSION}" -o /out/mm .
37+
38+# ---- template: the shell sandbox plus mm -----------------------------------
39+FROM ${SHELL_TEMPLATE}
40+
41+COPY --from=build --chmod=0755 /out/mm /usr/local/bin/mm
42+# The two kit entry points. mm-tui runs `mm -tui "$@"`; mm-acp-info prints how
43+# an editor reaches this sandbox and opens a shell (see docker-sbx/README.md).
44+COPY --chmod=0755 docker-sbx/template/mm-tui docker-sbx/template/mm-acp-info /usr/local/bin/
45+
46+# tcell (the window mode) needs a terminal type and a UTF-8 locale; the kits
47+# set them too, this is the fallback for a plain `docker run`.
48+ENV TERM=xterm-256color \
49+ COLORTERM=truecolor \
50+ LANG=C.UTF-8
51+
52+# Prove the binary runs on this platform before the image is tagged.
53+RUN mm --version && mm-tui --version
added docker-sbx/template/mm-acp-info +52 -0
new file mode 100755
@@ -0,0 +1,52 @@
1+#!/bin/bash
2+# Entry point of the mm-acp sandbox kit (docker-sbx/kits/mm-acp).
3+#
4+# An editor never runs this script. It reaches mm with
5+#
6+# sbx exec -i <sandbox> mm -acp
7+#
8+# because an ACP client needs raw stdin/stdout and `sbx run` always allocates
9+# a terminal. So attaching to this sandbox with `sbx run` prints the wiring
10+# for the editor, with this sandbox's real name and paths, then opens a shell.
11+set -u
12+
13+name="${SANDBOX_NAME:-$(hostname)}"
14+workspace="${WORKSPACE_DIR:-$PWD}"
15+config="${AGENT_CONFIG:-}"
16+if [ -z "$config" ]; then
17+ config_note="none set — mm reads ./agent.yaml in the workspace, then its built-in defaults"
18+elif [ "${config#/}" = "$config" ]; then
19+ config_note="$config (relative to the workspace: ${workspace%/}/$config)"
20+else
21+ config_note="$config"
22+fi
23+
24+cat <<INFO
25+
26+ mm $(mm --version 2>/dev/null | awk '{print $2}') in sandbox "$name"
27+ workspace : $workspace
28+ config : $config_note
29+
30+ This sandbox is meant for an editor (Agent Client Protocol). Register it in
31+ Zed's settings.json — the editor starts mm inside the sandbox on every
32+ conversation, and sbx starts the sandbox first when it is stopped:
33+
34+ {
35+ "agent_servers": {
36+ "mini-me (sandbox)": {
37+ "type": "custom",
38+ "command": "sbx",
39+ "args": ["exec", "-i", "$name", "mm", "-acp"]
40+ }
41+ }
42+ }
43+
44+ Use the absolute path of sbx as "command" if the editor cannot find it
45+ (\`which sbx\` on the host). To change the configuration file for one
46+ session, add "-e", "AGENT_CONFIG=/path/in/workspace/agent.yaml" before
47+ "$name" in "args". Full guide: docs/en/how-to/run-in-a-docker-sandbox.md
48+
49+ You are now in a shell inside the sandbox. \`mm agent.yaml\` works here too.
50+
51+INFO
52+exec bash -l
new file mode 100755
@@ -0,0 +1,52 @@
1+#!/bin/bash
2+# Entry point of the mm-acp sandbox kit (docker-sbx/kits/mm-acp).
3+#
4+# An editor never runs this script. It reaches mm with
5+#
6+# sbx exec -i <sandbox> mm -acp
7+#
8+# because an ACP client needs raw stdin/stdout and `sbx run` always allocates
9+# a terminal. So attaching to this sandbox with `sbx run` prints the wiring
10+# for the editor, with this sandbox's real name and paths, then opens a shell.
11+set -u
12+
13+name="${SANDBOX_NAME:-$(hostname)}"
14+workspace="${WORKSPACE_DIR:-$PWD}"
15+config="${AGENT_CONFIG:-}"
16+if [ -z "$config" ]; then
17+ config_note="none set — mm reads ./agent.yaml in the workspace, then its built-in defaults"
18+elif [ "${config#/}" = "$config" ]; then
19+ config_note="$config (relative to the workspace: ${workspace%/}/$config)"
20+else
21+ config_note="$config"
22+fi
23+
24+cat <<INFO
25+
26+ mm $(mm --version 2>/dev/null | awk '{print $2}') in sandbox "$name"
27+ workspace : $workspace
28+ config : $config_note
29+
30+ This sandbox is meant for an editor (Agent Client Protocol). Register it in
31+ Zed's settings.json — the editor starts mm inside the sandbox on every
32+ conversation, and sbx starts the sandbox first when it is stopped:
33+
34+ {
35+ "agent_servers": {
36+ "mini-me (sandbox)": {
37+ "type": "custom",
38+ "command": "sbx",
39+ "args": ["exec", "-i", "$name", "mm", "-acp"]
40+ }
41+ }
42+ }
43+
44+ Use the absolute path of sbx as "command" if the editor cannot find it
45+ (\`which sbx\` on the host). To change the configuration file for one
46+ session, add "-e", "AGENT_CONFIG=/path/in/workspace/agent.yaml" before
47+ "$name" in "args". Full guide: docs/en/how-to/run-in-a-docker-sandbox.md
48+
49+ You are now in a shell inside the sandbox. \`mm agent.yaml\` works here too.
50+
51+INFO
52+exec bash -l
added docker-sbx/template/mm-tui +11 -0
new file mode 100755
@@ -0,0 +1,11 @@
1+#!/bin/bash
2+# Entry point of the mm-tui sandbox kit (docker-sbx/kits/mm-tui).
3+#
4+# `sbx run` forwards whatever follows "--" as this script's arguments, so
5+#
6+# sbx run --pull never ./docker-sbx/kits/mm-tui . -- agent.dmr.yaml
7+#
8+# runs `mm -tui agent.dmr.yaml` in the workspace. Without an argument mm reads
9+# $AGENT_CONFIG (the kit's `config` argument), then ./agent.yaml, then its
10+# built-in defaults. Flags pass through too: `-- -config other.yaml`.
11+exec mm -tui "$@"
new file mode 100755
@@ -0,0 +1,11 @@
1+#!/bin/bash
2+# Entry point of the mm-tui sandbox kit (docker-sbx/kits/mm-tui).
3+#
4+# `sbx run` forwards whatever follows "--" as this script's arguments, so
5+#
6+# sbx run --pull never ./docker-sbx/kits/mm-tui . -- agent.dmr.yaml
7+#
8+# runs `mm -tui agent.dmr.yaml` in the workspace. Without an argument mm reads
9+# $AGENT_CONFIG (the kit's `config` argument), then ./agent.yaml, then its
10+# built-in defaults. Flags pass through too: `-- -config other.yaml`.
11+exec mm -tui "$@"
modified docs/en/README.md +1 -0
@@ -12,6 +12,7 @@ This documentation is organised along the four [Diátaxis](https://diataxis.fr)
1212
1313 - [Run mini-me on llama.cpp instead of Docker Model Runner](how-to/run-with-llama-cpp.md)
1414 - [Use mini-me from an editor (ACP)](how-to/use-from-an-editor.md)
15+- [Run mini-me in a Docker Sandbox](how-to/run-in-a-docker-sandbox.md) — the template and the two kits under `docker-sbx/`, window mode and editor
1516 - [Add a skill](how-to/add-a-skill.md)
1617 - [Measure whether the model picks the right skill](how-to/measure-skill-routing.md) — the routing eval: run it, read its table, use it as a before-and-after
1718 - [Keep a long session inside the context window](how-to/manage-the-context-window.md)
@@ -12,6 +12,7 @@ This documentation is organised along the four [Diátaxis](https://diataxis.fr)
12 12
13 - [Run mini-me on llama.cpp instead of Docker Model Runner](how-to/run-with-llama-cpp.md)13 - [Run mini-me on llama.cpp instead of Docker Model Runner](how-to/run-with-llama-cpp.md)
14 - [Use mini-me from an editor (ACP)](how-to/use-from-an-editor.md)14 - [Use mini-me from an editor (ACP)](how-to/use-from-an-editor.md)
15+- [Run mini-me in a Docker Sandbox](how-to/run-in-a-docker-sandbox.md) — the template and the two kits under `docker-sbx/`, window mode and editor
15 - [Add a skill](how-to/add-a-skill.md)16 - [Add a skill](how-to/add-a-skill.md)
16 - [Measure whether the model picks the right skill](how-to/measure-skill-routing.md) — the routing eval: run it, read its table, use it as a before-and-after17 - [Measure whether the model picks the right skill](how-to/measure-skill-routing.md) — the routing eval: run it, read its table, use it as a before-and-after
17 - [Keep a long session inside the context window](how-to/manage-the-context-window.md)18 - [Keep a long session inside the context window](how-to/manage-the-context-window.md)
added docs/en/how-to/run-in-a-docker-sandbox.md +138 -0
new file mode 100644
@@ -0,0 +1,138 @@
1+# How to run mini-me in a Docker Sandbox
2+
3+This guide shows how to run `mm` inside a [Docker Sandbox](https://docs.docker.com/ai/sandboxes/), so that every command the model runs and every file it edits happens in a container, on a mounted copy of your project. It covers the two ways of using it: the window mode in your terminal, and from an editor through the Agent Client Protocol. It assumes Docker Desktop with the `sbx` command, and a model server (Docker Model Runner or llama-server) running on your machine.
4+
5+Everything lives under [`docker-sbx/`](../../../docker-sbx/README.md) in the repository:
6+
7+| Piece | What it is |
8+|---|---|
9+| `template/Dockerfile` | The **template**: Docker Sandboxes' `shell` image with `mm` built from this checkout, plus the two entry points below. |
10+| `kits/mm-tui/spec.yaml` | A **sandbox kit**: `sbx run` on it opens `mm`'s window mode on the workspace. |
11+| `kits/mm-acp/spec.yaml` | A **sandbox kit** for an editor: created once with `sbx create`, the editor then runs `mm -acp` inside it. |
12+| `build.sh` | Builds the template image, `k33g/mm-sandbox:<version>`; `PUSH=1` builds it for linux/amd64 and linux/arm64 and pushes it. |
13+| `sync-kits.sh`, `push-kits.sh` | Point the kits at one image, and publish them to Docker Hub as OCI kit artifacts. |
14+
15+## Use the published kits
16+
17+Every release publishes the image and both kits to Docker Hub at mm's version — `k33g/mm-sandbox:0.0.2`, `k33g/mm-tui:0.0.2`, `k33g/mm-acp:0.0.2` — so the shortest path needs no build at all. Docker Hub is an allowed kit source by default (`sbx settings get kit.allowedSources` shows `docker.io/`), and a reference with a `/` but no registry host is read as `docker.io/…`:
18+
19+```bash
20+cd /path/to/your/project
21+sbx run k33g/mm-tui:0.0.2 . -- agent.dmr.yaml
22+```
23+
24+for the window mode, and for an editor:
25+
26+```bash
27+sbx create --name mm-acp k33g/mm-acp:0.0.2 . --kit-arg config=agent.dmr.yaml
28+```
29+
30+then the editor registration of step 3 below. No `--pull never`: the kit names `k33g/mm-sandbox:0.0.2` and sbx pulls it. `sbx kit inspect k33g/mm-tui:0.0.2` shows what a kit declares before you run it. The steps that follow build the same thing locally, for a work-in-progress mm.
31+
32+## Steps
33+
34+### 1. Build the template
35+
36+From the repository root:
37+
38+```bash
39+./docker-sbx/build.sh
40+```
41+
42+This reads mm's tag from `release.env` (`v0.0.2`) and builds `k33g/mm-sandbox:0.0.2`: the image tag is the version without the `v`, like the release assets, and `mm` inside is stamped with the tag — `mm --version` is run in the image as proof. The script then rewrites the kits' default `tag` argument to `0.0.2`, so the kits and the image move together. `./docker-sbx/build.sh v0.0.3` builds another version; an earlier image stays reachable with `--kit-arg tag=0.0.1`.
43+
44+The image is in your local Docker image store: every `sbx run` or `sbx create` on these kits takes `--pull never` until the image is pushed (`docker push k33g/mm-sandbox:0.0.2`), or `sbx` tries to pull it and fails. If `sbx` runs on another machine, `EXPORT=1 ./docker-sbx/build.sh` also writes `tmp/mm-sandbox-0.0.2.tar`, to load there with `sbx template load`.
45+
46+### 2. Window mode in the terminal (kit `mm-tui`)
47+
48+From the project you want `mm` to work on:
49+
50+```bash
51+sbx run --pull never ./docker-sbx/kits/mm-tui . -- agent.dmr.yaml
52+```
53+
54+The first time, `sbx` creates the sandbox, mounts the current directory as its workspace and attaches; then it runs the kit's entry point with what follows `--` as arguments — here `mm -tui agent.dmr.yaml`, in the workspace. Any `mm` flag passes the same way: `-- -config other.yaml`.
55+
56+Without anything after `--`, `mm` reads the kit's `config` argument, then `./agent.yaml` in the workspace, then its built-in defaults:
57+
58+```bash
59+sbx run --pull never ./docker-sbx/kits/mm-tui . --kit-arg config=agent.dmr.yaml
60+```
61+
62+| `--kit-arg` | Default | Effect |
63+|---|---|---|
64+| `config=<path>` | empty | The configuration file, as `mm`'s `AGENT_CONFIG`: relative to the workspace, or absolute inside the sandbox. |
65+| `tag=<version>` | the version last built (`0.0.2`) | Which `k33g/mm-sandbox:<version>` image the sandbox is made from. |
66+
67+The sandbox keeps its name (`mm-tui-<directory>` unless you pass `--name`); `sbx run --name <name>` re-attaches to it later. Inside, the window is the one described in [Use the window mode](use-the-window-mode.md). Sessions, when `sessions.enabled` is on, are written under the workspace's `.mm/sessions` — that is your host directory, so they survive the sandbox.
68+
69+### 3. From an editor (kit `mm-acp`)
70+
71+An editor talks to `mm -acp` over the agent's raw stdin and stdout. `sbx run` cannot carry that: it always allocates a terminal. So the sandbox is created once without attaching, and the editor starts `mm` inside it with `sbx exec -i`, which pipes stdin and stdout untouched.
72+
73+1. From the project the editor will open, create the sandbox with the configuration file baked in (`.` is the workspace, sbx makes it absolute; the kit path must be an explicit `./`, `../` or absolute path):
74+
75+ ```bash
76+ cd /path/to/your/project
77+ sbx create --pull never --name mm-acp /path/to/mini-me/docker-sbx/kits/mm-acp . \
78+ --kit-arg config=agent.dmr.yaml
79+ ```
80+
81+ A workspace path that does not exist on your machine makes sbx ask "The selected workspace does not exist. Would you like to create it?" — answer `N` and fix the path.
82+
83+ `--kit-arg config` becomes the container's `AGENT_CONFIG`, relative to the workspace because that is where `sbx exec` starts `mm`. The workspace is fixed at creation: one sandbox per project, each with its own `--name`.
84+
85+2. Register it in the editor. In Zed's `settings.json`:
86+
87+ ```json
88+ {
89+ "agent_servers": {
90+ "mini-me (sandbox)": {
91+ "type": "custom",
92+ "command": "sbx",
93+ "args": ["exec", "-i", "mm-acp", "mm", "-acp"]
94+ }
95+ }
96+ }
97+ ```
98+
99+ Compared with [the host-side setup](use-from-an-editor.md), the command is `sbx` and no `env` is needed: `AGENT_CONFIG` is already in the container. If the editor cannot find `sbx`, put the absolute path from `which sbx` in `"command"`.
100+
101+3. Open the agent panel and pick **mini-me (sandbox)**. The editor starts `mm -acp` in the sandbox for every conversation; `sbx exec` starts the sandbox first when it is stopped. Tool calls, permission dialogs, `@file` mentions and the slash-command menu work as they do on the host — the difference is where the commands run.
102+
103+Attaching to this sandbox with `sbx run --name mm-acp` does not start `mm`: its entry point prints the wiring above with the sandbox's real name and paths, then opens a shell inside the sandbox. `sbx exec -it mm-acp bash` is the shorter way to that shell.
104+
105+## The model server stays on the host
106+
107+Inside a sandbox `localhost` is the sandbox itself. Your model server keeps running on your machine and is reached through `host.docker.internal`, which the sandbox proxy maps to your host's `localhost`.
108+
109+- **Docker Model Runner** — nothing to change: `agent.dmr.yaml` already has `fallback: http://host.docker.internal:12434/engines/v1`, and `mm` switches to it when `localhost:12434` does not answer (the banner says which URL won).
110+- **llama-server** — `agent.llamacpp.yaml` has no fallback. Either edit its `baseUrl` to `http://host.docker.internal:8080/v1`, or pass `--env AGENT_BASE_URL=http://host.docker.internal:8080/v1` to `sbx run` / `sbx create` (see [Run with llama.cpp](run-with-llama-cpp.md)).
111+
112+Both kits allow ports 12434 and 8080 on the host in the sandbox's network policy. For another port: `sbx policy allow network host.docker.internal:<port>`.
113+
114+## Publishing a version
115+
116+The Release workflow (`.github/workflows/release.yml`, job `docker-sandbox`) publishes the image and the kits on every `v*` tag, after the binaries job. It needs two repository secrets, exposed to the job as environment variables: `DOCKER_HANDLE`, the Docker Hub account (also the image owner), and `DOCKER_PAT`, an access token with read/write scope. It then runs the scripts a laptop would run:
117+
118+```bash
119+docker login -u k33g # paste an access token
120+PUSH=1 ./docker-sbx/build.sh v0.0.2 # buildx, linux/amd64 + linux/arm64, pushed as k33g/mm-sandbox:0.0.2
121+./docker-sbx/push-kits.sh v0.0.2 # sbx kit validate + sbx kit push → k33g/mm-tui:0.0.2, k33g/mm-acp:0.0.2
122+```
123+
124+`push-kits.sh` needs the `sbx` CLI on the PATH (or `SBX=/path/to/sbx`; the workflow downloads the Linux tarball of `docker/sbx-releases`). `sbx kit push` talks to the registry directly and authenticates from the Docker credential store — no sandbox daemon is involved, which is what makes it usable in CI. Both scripts call `sync-kits.sh` first, which rewrites the kits' `version`, default `tag` and image owner, so a published kit always names the image pushed with it. `IMAGE_OWNER=<handle>` publishes under another account.
125+
126+## Variants
127+
128+- **Another configuration for one editor session.** Add `"-e", "AGENT_CONFIG=agent.lsp.yaml"` before the sandbox name in `"args"`; `sbx exec -e` overrides the baked-in value for that run only.
129+- **Skills.** `skillsDir` is relative to the directory `mm` runs in, which is the workspace in both kits — the same rule as on the host.
130+- **Docker inside the sandbox.** Build the template on the Docker-in-Docker shell instead: `SHELL_TEMPLATE=docker/sandbox-templates:shell-docker ./docker-sbx/build.sh`.
131+- **Read the logs.** In the window mode they are on screen; over ACP they go to stderr, which Zed shows under `dev: open acp logs`.
132+- **Remove a sandbox.** `sbx rm mm-acp`. The project directory is a bind mount and is not touched.
133+
134+## See also
135+
136+- The kit format these two files follow: the Docker Sandboxes kit specification (`schemaVersion: "2"`), bundled with the `sbx` sources.
137+- What `mm -acp` exposes to the editor: [ACP reference](../reference/acp.md).
138+- The flags and environment variables `mm` reads: [command line reference](../reference/cli.md).
new file mode 100644
@@ -0,0 +1,138 @@
1+# How to run mini-me in a Docker Sandbox
2+
3+This guide shows how to run `mm` inside a [Docker Sandbox](https://docs.docker.com/ai/sandboxes/), so that every command the model runs and every file it edits happens in a container, on a mounted copy of your project. It covers the two ways of using it: the window mode in your terminal, and from an editor through the Agent Client Protocol. It assumes Docker Desktop with the `sbx` command, and a model server (Docker Model Runner or llama-server) running on your machine.
4+
5+Everything lives under [`docker-sbx/`](../../../docker-sbx/README.md) in the repository:
6+
7+| Piece | What it is |
8+|---|---|
9+| `template/Dockerfile` | The **template**: Docker Sandboxes' `shell` image with `mm` built from this checkout, plus the two entry points below. |
10+| `kits/mm-tui/spec.yaml` | A **sandbox kit**: `sbx run` on it opens `mm`'s window mode on the workspace. |
11+| `kits/mm-acp/spec.yaml` | A **sandbox kit** for an editor: created once with `sbx create`, the editor then runs `mm -acp` inside it. |
12+| `build.sh` | Builds the template image, `k33g/mm-sandbox:<version>`; `PUSH=1` builds it for linux/amd64 and linux/arm64 and pushes it. |
13+| `sync-kits.sh`, `push-kits.sh` | Point the kits at one image, and publish them to Docker Hub as OCI kit artifacts. |
14+
15+## Use the published kits
16+
17+Every release publishes the image and both kits to Docker Hub at mm's version — `k33g/mm-sandbox:0.0.2`, `k33g/mm-tui:0.0.2`, `k33g/mm-acp:0.0.2` — so the shortest path needs no build at all. Docker Hub is an allowed kit source by default (`sbx settings get kit.allowedSources` shows `docker.io/`), and a reference with a `/` but no registry host is read as `docker.io/…`:
18+
19+```bash
20+cd /path/to/your/project
21+sbx run k33g/mm-tui:0.0.2 . -- agent.dmr.yaml
22+```
23+
24+for the window mode, and for an editor:
25+
26+```bash
27+sbx create --name mm-acp k33g/mm-acp:0.0.2 . --kit-arg config=agent.dmr.yaml
28+```
29+
30+then the editor registration of step 3 below. No `--pull never`: the kit names `k33g/mm-sandbox:0.0.2` and sbx pulls it. `sbx kit inspect k33g/mm-tui:0.0.2` shows what a kit declares before you run it. The steps that follow build the same thing locally, for a work-in-progress mm.
31+
32+## Steps
33+
34+### 1. Build the template
35+
36+From the repository root:
37+
38+```bash
39+./docker-sbx/build.sh
40+```
41+
42+This reads mm's tag from `release.env` (`v0.0.2`) and builds `k33g/mm-sandbox:0.0.2`: the image tag is the version without the `v`, like the release assets, and `mm` inside is stamped with the tag — `mm --version` is run in the image as proof. The script then rewrites the kits' default `tag` argument to `0.0.2`, so the kits and the image move together. `./docker-sbx/build.sh v0.0.3` builds another version; an earlier image stays reachable with `--kit-arg tag=0.0.1`.
43+
44+The image is in your local Docker image store: every `sbx run` or `sbx create` on these kits takes `--pull never` until the image is pushed (`docker push k33g/mm-sandbox:0.0.2`), or `sbx` tries to pull it and fails. If `sbx` runs on another machine, `EXPORT=1 ./docker-sbx/build.sh` also writes `tmp/mm-sandbox-0.0.2.tar`, to load there with `sbx template load`.
45+
46+### 2. Window mode in the terminal (kit `mm-tui`)
47+
48+From the project you want `mm` to work on:
49+
50+```bash
51+sbx run --pull never ./docker-sbx/kits/mm-tui . -- agent.dmr.yaml
52+```
53+
54+The first time, `sbx` creates the sandbox, mounts the current directory as its workspace and attaches; then it runs the kit's entry point with what follows `--` as arguments — here `mm -tui agent.dmr.yaml`, in the workspace. Any `mm` flag passes the same way: `-- -config other.yaml`.
55+
56+Without anything after `--`, `mm` reads the kit's `config` argument, then `./agent.yaml` in the workspace, then its built-in defaults:
57+
58+```bash
59+sbx run --pull never ./docker-sbx/kits/mm-tui . --kit-arg config=agent.dmr.yaml
60+```
61+
62+| `--kit-arg` | Default | Effect |
63+|---|---|---|
64+| `config=<path>` | empty | The configuration file, as `mm`'s `AGENT_CONFIG`: relative to the workspace, or absolute inside the sandbox. |
65+| `tag=<version>` | the version last built (`0.0.2`) | Which `k33g/mm-sandbox:<version>` image the sandbox is made from. |
66+
67+The sandbox keeps its name (`mm-tui-<directory>` unless you pass `--name`); `sbx run --name <name>` re-attaches to it later. Inside, the window is the one described in [Use the window mode](use-the-window-mode.md). Sessions, when `sessions.enabled` is on, are written under the workspace's `.mm/sessions` — that is your host directory, so they survive the sandbox.
68+
69+### 3. From an editor (kit `mm-acp`)
70+
71+An editor talks to `mm -acp` over the agent's raw stdin and stdout. `sbx run` cannot carry that: it always allocates a terminal. So the sandbox is created once without attaching, and the editor starts `mm` inside it with `sbx exec -i`, which pipes stdin and stdout untouched.
72+
73+1. From the project the editor will open, create the sandbox with the configuration file baked in (`.` is the workspace, sbx makes it absolute; the kit path must be an explicit `./`, `../` or absolute path):
74+
75+ ```bash
76+ cd /path/to/your/project
77+ sbx create --pull never --name mm-acp /path/to/mini-me/docker-sbx/kits/mm-acp . \
78+ --kit-arg config=agent.dmr.yaml
79+ ```
80+
81+ A workspace path that does not exist on your machine makes sbx ask "The selected workspace does not exist. Would you like to create it?" — answer `N` and fix the path.
82+
83+ `--kit-arg config` becomes the container's `AGENT_CONFIG`, relative to the workspace because that is where `sbx exec` starts `mm`. The workspace is fixed at creation: one sandbox per project, each with its own `--name`.
84+
85+2. Register it in the editor. In Zed's `settings.json`:
86+
87+ ```json
88+ {
89+ "agent_servers": {
90+ "mini-me (sandbox)": {
91+ "type": "custom",
92+ "command": "sbx",
93+ "args": ["exec", "-i", "mm-acp", "mm", "-acp"]
94+ }
95+ }
96+ }
97+ ```
98+
99+ Compared with [the host-side setup](use-from-an-editor.md), the command is `sbx` and no `env` is needed: `AGENT_CONFIG` is already in the container. If the editor cannot find `sbx`, put the absolute path from `which sbx` in `"command"`.
100+
101+3. Open the agent panel and pick **mini-me (sandbox)**. The editor starts `mm -acp` in the sandbox for every conversation; `sbx exec` starts the sandbox first when it is stopped. Tool calls, permission dialogs, `@file` mentions and the slash-command menu work as they do on the host — the difference is where the commands run.
102+
103+Attaching to this sandbox with `sbx run --name mm-acp` does not start `mm`: its entry point prints the wiring above with the sandbox's real name and paths, then opens a shell inside the sandbox. `sbx exec -it mm-acp bash` is the shorter way to that shell.
104+
105+## The model server stays on the host
106+
107+Inside a sandbox `localhost` is the sandbox itself. Your model server keeps running on your machine and is reached through `host.docker.internal`, which the sandbox proxy maps to your host's `localhost`.
108+
109+- **Docker Model Runner** — nothing to change: `agent.dmr.yaml` already has `fallback: http://host.docker.internal:12434/engines/v1`, and `mm` switches to it when `localhost:12434` does not answer (the banner says which URL won).
110+- **llama-server** — `agent.llamacpp.yaml` has no fallback. Either edit its `baseUrl` to `http://host.docker.internal:8080/v1`, or pass `--env AGENT_BASE_URL=http://host.docker.internal:8080/v1` to `sbx run` / `sbx create` (see [Run with llama.cpp](run-with-llama-cpp.md)).
111+
112+Both kits allow ports 12434 and 8080 on the host in the sandbox's network policy. For another port: `sbx policy allow network host.docker.internal:<port>`.
113+
114+## Publishing a version
115+
116+The Release workflow (`.github/workflows/release.yml`, job `docker-sandbox`) publishes the image and the kits on every `v*` tag, after the binaries job. It needs two repository secrets, exposed to the job as environment variables: `DOCKER_HANDLE`, the Docker Hub account (also the image owner), and `DOCKER_PAT`, an access token with read/write scope. It then runs the scripts a laptop would run:
117+
118+```bash
119+docker login -u k33g # paste an access token
120+PUSH=1 ./docker-sbx/build.sh v0.0.2 # buildx, linux/amd64 + linux/arm64, pushed as k33g/mm-sandbox:0.0.2
121+./docker-sbx/push-kits.sh v0.0.2 # sbx kit validate + sbx kit push → k33g/mm-tui:0.0.2, k33g/mm-acp:0.0.2
122+```
123+
124+`push-kits.sh` needs the `sbx` CLI on the PATH (or `SBX=/path/to/sbx`; the workflow downloads the Linux tarball of `docker/sbx-releases`). `sbx kit push` talks to the registry directly and authenticates from the Docker credential store — no sandbox daemon is involved, which is what makes it usable in CI. Both scripts call `sync-kits.sh` first, which rewrites the kits' `version`, default `tag` and image owner, so a published kit always names the image pushed with it. `IMAGE_OWNER=<handle>` publishes under another account.
125+
126+## Variants
127+
128+- **Another configuration for one editor session.** Add `"-e", "AGENT_CONFIG=agent.lsp.yaml"` before the sandbox name in `"args"`; `sbx exec -e` overrides the baked-in value for that run only.
129+- **Skills.** `skillsDir` is relative to the directory `mm` runs in, which is the workspace in both kits — the same rule as on the host.
130+- **Docker inside the sandbox.** Build the template on the Docker-in-Docker shell instead: `SHELL_TEMPLATE=docker/sandbox-templates:shell-docker ./docker-sbx/build.sh`.
131+- **Read the logs.** In the window mode they are on screen; over ACP they go to stderr, which Zed shows under `dev: open acp logs`.
132+- **Remove a sandbox.** `sbx rm mm-acp`. The project directory is a bind mount and is not touched.
133+
134+## See also
135+
136+- The kit format these two files follow: the Docker Sandboxes kit specification (`schemaVersion: "2"`), bundled with the `sbx` sources.
137+- What `mm -acp` exposes to the editor: [ACP reference](../reference/acp.md).
138+- The flags and environment variables `mm` reads: [command line reference](../reference/cli.md).
modified docs/fr/README.md +1 -0
@@ -12,6 +12,7 @@ Cette documentation est organisée selon les quatre quadrants [Diátaxis](https:
1212
1313 - [Faire tourner mini-me sur llama.cpp plutôt que Docker Model Runner](how-to/run-with-llama-cpp.md)
1414 - [Utiliser mini-me depuis un éditeur (ACP)](how-to/use-from-an-editor.md)
15+- [Faire tourner mini-me dans un Docker Sandbox](how-to/run-in-a-docker-sandbox.md) — le template et les deux kits sous `docker-sbx/`, mode fenêtre et éditeur
1516 - [Ajouter un skill](how-to/add-a-skill.md)
1617 - [Mesurer si le modèle choisit le bon skill](how-to/measure-skill-routing.md) — l'eval de routage : le lancer, lire son tableau, s'en servir en avant/après
1718 - [Garder une longue session dans la fenêtre de contexte](how-to/manage-the-context-window.md)
@@ -12,6 +12,7 @@ Cette documentation est organisée selon les quatre quadrants [Diátaxis](https:
12 12
13 - [Faire tourner mini-me sur llama.cpp plutôt que Docker Model Runner](how-to/run-with-llama-cpp.md)13 - [Faire tourner mini-me sur llama.cpp plutôt que Docker Model Runner](how-to/run-with-llama-cpp.md)
14 - [Utiliser mini-me depuis un éditeur (ACP)](how-to/use-from-an-editor.md)14 - [Utiliser mini-me depuis un éditeur (ACP)](how-to/use-from-an-editor.md)
15+- [Faire tourner mini-me dans un Docker Sandbox](how-to/run-in-a-docker-sandbox.md) — le template et les deux kits sous `docker-sbx/`, mode fenêtre et éditeur
15 - [Ajouter un skill](how-to/add-a-skill.md)16 - [Ajouter un skill](how-to/add-a-skill.md)
16 - [Mesurer si le modèle choisit le bon skill](how-to/measure-skill-routing.md) — l'eval de routage : le lancer, lire son tableau, s'en servir en avant/après17 - [Mesurer si le modèle choisit le bon skill](how-to/measure-skill-routing.md) — l'eval de routage : le lancer, lire son tableau, s'en servir en avant/après
17 - [Garder une longue session dans la fenêtre de contexte](how-to/manage-the-context-window.md)18 - [Garder une longue session dans la fenêtre de contexte](how-to/manage-the-context-window.md)
added docs/fr/how-to/run-in-a-docker-sandbox.md +138 -0
new file mode 100644
@@ -0,0 +1,138 @@
1+# Faire tourner mini-me dans un Docker Sandbox
2+
3+Ce guide montre comment exécuter `mm` dans un [Docker Sandbox](https://docs.docker.com/ai/sandboxes/), pour que chaque commande lancée par le modèle et chaque fichier qu'il modifie le soient dans un conteneur, sur votre projet monté. Il couvre les deux usages : le mode fenêtre dans votre terminal, et depuis un éditeur via l'Agent Client Protocol. Il suppose Docker Desktop avec la commande `sbx`, et un serveur de modèles (Docker Model Runner ou llama-server) qui tourne sur votre machine.
4+
5+Tout est sous [`docker-sbx/`](../../../docker-sbx/README.md) dans le dépôt :
6+
7+| Élément | Ce que c'est |
8+|---|---|
9+| `template/Dockerfile` | Le **template** : l'image `shell` de Docker Sandboxes avec `mm` compilé depuis ce dépôt, plus les deux points d'entrée ci-dessous. |
10+| `kits/mm-tui/spec.yaml` | Un **kit sandbox** : `sbx run` dessus ouvre le mode fenêtre de `mm` sur l'espace de travail. |
11+| `kits/mm-acp/spec.yaml` | Un **kit sandbox** pour un éditeur : créé une fois avec `sbx create`, l'éditeur y lance ensuite `mm -acp`. |
12+| `build.sh` | Construit l'image du template, `k33g/mm-sandbox:<version>` ; `PUSH=1` la construit pour linux/amd64 et linux/arm64 et la pousse. |
13+| `sync-kits.sh`, `push-kits.sh` | Alignent les kits sur une image, et les publient sur Docker Hub comme artefacts OCI de kit. |
14+
15+## Utiliser les kits publiés
16+
17+Chaque release publie l'image et les deux kits sur Docker Hub à la version de mm — `k33g/mm-sandbox:0.0.2`, `k33g/mm-tui:0.0.2`, `k33g/mm-acp:0.0.2` — le chemin le plus court ne demande donc aucune construction. Docker Hub est une source de kits autorisée par défaut (`sbx settings get kit.allowedSources` montre `docker.io/`), et une référence avec un `/` mais sans hôte de registre est lue comme `docker.io/…` :
18+
19+```bash
20+cd /chemin/de/votre/projet
21+sbx run k33g/mm-tui:0.0.2 . -- agent.dmr.yaml
22+```
23+
24+pour le mode fenêtre, et pour un éditeur :
25+
26+```bash
27+sbx create --name mm-acp k33g/mm-acp:0.0.2 . --kit-arg config=agent.dmr.yaml
28+```
29+
30+puis la déclaration dans l'éditeur de l'étape 3 ci-dessous. Pas de `--pull never` : le kit nomme `k33g/mm-sandbox:0.0.2` et sbx la télécharge. `sbx kit inspect k33g/mm-tui:0.0.2` montre ce qu'un kit déclare avant de le lancer. Les étapes qui suivent construisent la même chose en local, pour un mm en cours de travail.
31+
32+## Étapes
33+
34+### 1. Construire le template
35+
36+Depuis la racine du dépôt :
37+
38+```bash
39+./docker-sbx/build.sh
40+```
41+
42+Cela lit le tag de mm dans `release.env` (`v0.0.2`) et construit `k33g/mm-sandbox:0.0.2` : le tag de l'image est la version sans le `v`, comme les binaires de release, et le `mm` à l'intérieur est estampillé avec le tag — `mm --version` est exécuté dans l'image comme preuve. Le script réécrit ensuite la valeur par défaut de l'argument `tag` des kits à `0.0.2`, pour que kits et image avancent ensemble. `./docker-sbx/build.sh v0.0.3` construit une autre version ; une image antérieure reste joignable avec `--kit-arg tag=0.0.1`.
43+
44+L'image est dans votre store d'images Docker local : chaque `sbx run` ou `sbx create` sur ces kits prend `--pull never` tant qu'elle n'est pas poussée (`docker push k33g/mm-sandbox:0.0.2`), sinon `sbx` tente de la télécharger et échoue. Si `sbx` tourne sur une autre machine, `EXPORT=1 ./docker-sbx/build.sh` écrit aussi `tmp/mm-sandbox-0.0.2.tar`, à charger là-bas avec `sbx template load`.
45+
46+### 2. Mode fenêtre dans le terminal (kit `mm-tui`)
47+
48+Depuis le projet sur lequel `mm` doit travailler :
49+
50+```bash
51+sbx run --pull never ./docker-sbx/kits/mm-tui . -- agent.dmr.yaml
52+```
53+
54+La première fois, `sbx` crée le sandbox, monte le répertoire courant comme espace de travail et s'y attache ; puis il lance le point d'entrée du kit avec ce qui suit `--` en arguments — ici `mm -tui agent.dmr.yaml`, dans l'espace de travail. N'importe quel drapeau de `mm` passe de la même façon : `-- -config autre.yaml`.
55+
56+Sans rien après `--`, `mm` lit l'argument `config` du kit, puis `./agent.yaml` dans l'espace de travail, puis ses valeurs par défaut :
57+
58+```bash
59+sbx run --pull never ./docker-sbx/kits/mm-tui . --kit-arg config=agent.dmr.yaml
60+```
61+
62+| `--kit-arg` | Défaut | Effet |
63+|---|---|---|
64+| `config=<chemin>` | vide | Le fichier de configuration, comme `AGENT_CONFIG` de `mm` : relatif à l'espace de travail, ou absolu dans le sandbox. |
65+| `tag=<version>` | la dernière version construite (`0.0.2`) | L'image `k33g/mm-sandbox:<version>` dont le sandbox est fait. |
66+
67+Le sandbox garde son nom (`mm-tui-<répertoire>` sauf `--name`) ; `sbx run --name <nom>` s'y rattache plus tard. À l'intérieur, la fenêtre est celle décrite dans [Utiliser le mode fenêtre](use-the-window-mode.md). Les sessions, quand `sessions.enabled` est actif, sont écrites sous `.mm/sessions` dans l'espace de travail — c'est votre répertoire hôte, elles survivent donc au sandbox.
68+
69+### 3. Depuis un éditeur (kit `mm-acp`)
70+
71+Un éditeur parle à `mm -acp` sur l'entrée et la sortie standard brutes de l'agent. `sbx run` ne peut pas porter cela : il alloue toujours un terminal. Le sandbox est donc créé une fois sans s'y attacher, et l'éditeur y lance `mm` avec `sbx exec -i`, qui transmet stdin et stdout tels quels.
72+
73+1. Depuis le projet que l'éditeur ouvrira, créez le sandbox avec le fichier de configuration intégré (`.` est l'espace de travail, sbx le rend absolu ; le chemin du kit doit être explicite : `./`, `../` ou absolu) :
74+
75+ ```bash
76+ cd /chemin/de/votre/projet
77+ sbx create --pull never --name mm-acp /chemin/de/mini-me/docker-sbx/kits/mm-acp . \
78+ --kit-arg config=agent.dmr.yaml
79+ ```
80+
81+ Un chemin d'espace de travail qui n'existe pas sur votre machine fait demander à sbx « The selected workspace does not exist. Would you like to create it? » — répondez `N` et corrigez le chemin.
82+
83+ `--kit-arg config` devient l'`AGENT_CONFIG` du conteneur, relatif à l'espace de travail parce que c'est là que `sbx exec` démarre `mm`. L'espace de travail est fixé à la création : un sandbox par projet, chacun avec son `--name`.
84+
85+2. Déclarez-le dans l'éditeur. Dans le `settings.json` de Zed :
86+
87+ ```json
88+ {
89+ "agent_servers": {
90+ "mini-me (sandbox)": {
91+ "type": "custom",
92+ "command": "sbx",
93+ "args": ["exec", "-i", "mm-acp", "mm", "-acp"]
94+ }
95+ }
96+ }
97+ ```
98+
99+ Par rapport à [la configuration côté hôte](use-from-an-editor.md), la commande est `sbx` et aucun `env` n'est nécessaire : `AGENT_CONFIG` est déjà dans le conteneur. Si l'éditeur ne trouve pas `sbx`, mettez le chemin absolu donné par `which sbx` dans `"command"`.
100+
101+3. Ouvrez le panneau agent et choisissez **mini-me (sandbox)**. L'éditeur lance `mm -acp` dans le sandbox à chaque conversation ; `sbx exec` démarre d'abord le sandbox s'il est arrêté. Appels d'outils, dialogues de permission, mentions `@fichier` et menu des commandes fonctionnent comme sur l'hôte — la différence est l'endroit où les commandes s'exécutent.
102+
103+S'attacher à ce sandbox avec `sbx run --name mm-acp` ne lance pas `mm` : son point d'entrée affiche le câblage ci-dessus avec le vrai nom et les vrais chemins du sandbox, puis ouvre un shell à l'intérieur. `sbx exec -it mm-acp bash` est le chemin le plus court vers ce shell.
104+
105+## Le serveur de modèles reste sur l'hôte
106+
107+Dans un sandbox, `localhost` est le sandbox lui-même. Votre serveur de modèles continue de tourner sur votre machine et se joint par `host.docker.internal`, que le proxy du sandbox fait pointer vers le `localhost` de l'hôte.
108+
109+- **Docker Model Runner** — rien à changer : `agent.dmr.yaml` a déjà `fallback: http://host.docker.internal:12434/engines/v1`, et `mm` bascule dessus quand `localhost:12434` ne répond pas (la bannière dit quelle URL a gagné).
110+- **llama-server** — `agent.llamacpp.yaml` n'a pas de fallback. Soit vous changez son `baseUrl` en `http://host.docker.internal:8080/v1`, soit vous passez `--env AGENT_BASE_URL=http://host.docker.internal:8080/v1` à `sbx run` / `sbx create` (voir [Faire tourner sur llama.cpp](run-with-llama-cpp.md)).
111+
112+Les deux kits autorisent les ports 12434 et 8080 de l'hôte dans la politique réseau du sandbox. Pour un autre port : `sbx policy allow network host.docker.internal:<port>`.
113+
114+## Publier une version
115+
116+Le workflow Release (`.github/workflows/release.yml`, job `docker-sandbox`) publie l'image et les kits à chaque tag `v*`, après le job des binaires. Il a besoin de deux secrets du dépôt, exposés au job comme variables d'environnement : `DOCKER_HANDLE`, le compte Docker Hub (aussi le propriétaire de l'image), et `DOCKER_PAT`, un jeton d'accès en lecture/écriture. Il lance ensuite les scripts qu'un portable lancerait :
117+
118+```bash
119+docker login -u k33g # coller un jeton d'accès
120+PUSH=1 ./docker-sbx/build.sh v0.0.2 # buildx, linux/amd64 + linux/arm64, poussée comme k33g/mm-sandbox:0.0.2
121+./docker-sbx/push-kits.sh v0.0.2 # sbx kit validate + sbx kit push → k33g/mm-tui:0.0.2, k33g/mm-acp:0.0.2
122+```
123+
124+`push-kits.sh` a besoin de la CLI `sbx` sur le PATH (ou `SBX=/chemin/vers/sbx` ; le workflow télécharge l'archive Linux de `docker/sbx-releases`). `sbx kit push` parle directement au registre et s'authentifie depuis le magasin d'identifiants Docker — aucun démon sandbox n'intervient, c'est ce qui le rend utilisable en CI. Les deux scripts appellent d'abord `sync-kits.sh`, qui réécrit `version`, le `tag` par défaut et le propriétaire de l'image des kits, de sorte qu'un kit publié nomme toujours l'image poussée avec lui. `IMAGE_OWNER=<compte>` publie sous un autre compte.
125+
126+## Variantes
127+
128+- **Une autre configuration pour une session d'éditeur.** Ajoutez `"-e", "AGENT_CONFIG=agent.lsp.yaml"` avant le nom du sandbox dans `"args"` ; `sbx exec -e` remplace la valeur intégrée pour cette exécution seulement.
129+- **Skills.** `skillsDir` est relatif au répertoire où `mm` tourne, c'est-à-dire l'espace de travail dans les deux kits — la même règle que sur l'hôte.
130+- **Docker dans le sandbox.** Construisez le template sur le shell Docker-in-Docker : `SHELL_TEMPLATE=docker/sandbox-templates:shell-docker ./docker-sbx/build.sh`.
131+- **Lire les logs.** En mode fenêtre ils sont à l'écran ; en ACP ils vont sur stderr, que Zed montre sous `dev: open acp logs`.
132+- **Supprimer un sandbox.** `sbx rm mm-acp`. Le répertoire du projet est un montage et n'est pas touché.
133+
134+## Voir aussi
135+
136+- Le format suivi par ces deux fichiers : la spécification des kits de Docker Sandboxes (`schemaVersion: "2"`), livrée avec les sources de `sbx`.
137+- Ce que `mm -acp` expose à l'éditeur : [référence ACP](../reference/acp.md).
138+- Les drapeaux et variables d'environnement que `mm` lit : [référence de la ligne de commande](../reference/cli.md).
new file mode 100644
@@ -0,0 +1,138 @@
1+# Faire tourner mini-me dans un Docker Sandbox
2+
3+Ce guide montre comment exécuter `mm` dans un [Docker Sandbox](https://docs.docker.com/ai/sandboxes/), pour que chaque commande lancée par le modèle et chaque fichier qu'il modifie le soient dans un conteneur, sur votre projet monté. Il couvre les deux usages : le mode fenêtre dans votre terminal, et depuis un éditeur via l'Agent Client Protocol. Il suppose Docker Desktop avec la commande `sbx`, et un serveur de modèles (Docker Model Runner ou llama-server) qui tourne sur votre machine.
4+
5+Tout est sous [`docker-sbx/`](../../../docker-sbx/README.md) dans le dépôt :
6+
7+| Élément | Ce que c'est |
8+|---|---|
9+| `template/Dockerfile` | Le **template** : l'image `shell` de Docker Sandboxes avec `mm` compilé depuis ce dépôt, plus les deux points d'entrée ci-dessous. |
10+| `kits/mm-tui/spec.yaml` | Un **kit sandbox** : `sbx run` dessus ouvre le mode fenêtre de `mm` sur l'espace de travail. |
11+| `kits/mm-acp/spec.yaml` | Un **kit sandbox** pour un éditeur : créé une fois avec `sbx create`, l'éditeur y lance ensuite `mm -acp`. |
12+| `build.sh` | Construit l'image du template, `k33g/mm-sandbox:<version>` ; `PUSH=1` la construit pour linux/amd64 et linux/arm64 et la pousse. |
13+| `sync-kits.sh`, `push-kits.sh` | Alignent les kits sur une image, et les publient sur Docker Hub comme artefacts OCI de kit. |
14+
15+## Utiliser les kits publiés
16+
17+Chaque release publie l'image et les deux kits sur Docker Hub à la version de mm — `k33g/mm-sandbox:0.0.2`, `k33g/mm-tui:0.0.2`, `k33g/mm-acp:0.0.2` — le chemin le plus court ne demande donc aucune construction. Docker Hub est une source de kits autorisée par défaut (`sbx settings get kit.allowedSources` montre `docker.io/`), et une référence avec un `/` mais sans hôte de registre est lue comme `docker.io/…` :
18+
19+```bash
20+cd /chemin/de/votre/projet
21+sbx run k33g/mm-tui:0.0.2 . -- agent.dmr.yaml
22+```
23+
24+pour le mode fenêtre, et pour un éditeur :
25+
26+```bash
27+sbx create --name mm-acp k33g/mm-acp:0.0.2 . --kit-arg config=agent.dmr.yaml
28+```
29+
30+puis la déclaration dans l'éditeur de l'étape 3 ci-dessous. Pas de `--pull never` : le kit nomme `k33g/mm-sandbox:0.0.2` et sbx la télécharge. `sbx kit inspect k33g/mm-tui:0.0.2` montre ce qu'un kit déclare avant de le lancer. Les étapes qui suivent construisent la même chose en local, pour un mm en cours de travail.
31+
32+## Étapes
33+
34+### 1. Construire le template
35+
36+Depuis la racine du dépôt :
37+
38+```bash
39+./docker-sbx/build.sh
40+```
41+
42+Cela lit le tag de mm dans `release.env` (`v0.0.2`) et construit `k33g/mm-sandbox:0.0.2` : le tag de l'image est la version sans le `v`, comme les binaires de release, et le `mm` à l'intérieur est estampillé avec le tag — `mm --version` est exécuté dans l'image comme preuve. Le script réécrit ensuite la valeur par défaut de l'argument `tag` des kits à `0.0.2`, pour que kits et image avancent ensemble. `./docker-sbx/build.sh v0.0.3` construit une autre version ; une image antérieure reste joignable avec `--kit-arg tag=0.0.1`.
43+
44+L'image est dans votre store d'images Docker local : chaque `sbx run` ou `sbx create` sur ces kits prend `--pull never` tant qu'elle n'est pas poussée (`docker push k33g/mm-sandbox:0.0.2`), sinon `sbx` tente de la télécharger et échoue. Si `sbx` tourne sur une autre machine, `EXPORT=1 ./docker-sbx/build.sh` écrit aussi `tmp/mm-sandbox-0.0.2.tar`, à charger là-bas avec `sbx template load`.
45+
46+### 2. Mode fenêtre dans le terminal (kit `mm-tui`)
47+
48+Depuis le projet sur lequel `mm` doit travailler :
49+
50+```bash
51+sbx run --pull never ./docker-sbx/kits/mm-tui . -- agent.dmr.yaml
52+```
53+
54+La première fois, `sbx` crée le sandbox, monte le répertoire courant comme espace de travail et s'y attache ; puis il lance le point d'entrée du kit avec ce qui suit `--` en arguments — ici `mm -tui agent.dmr.yaml`, dans l'espace de travail. N'importe quel drapeau de `mm` passe de la même façon : `-- -config autre.yaml`.
55+
56+Sans rien après `--`, `mm` lit l'argument `config` du kit, puis `./agent.yaml` dans l'espace de travail, puis ses valeurs par défaut :
57+
58+```bash
59+sbx run --pull never ./docker-sbx/kits/mm-tui . --kit-arg config=agent.dmr.yaml
60+```
61+
62+| `--kit-arg` | Défaut | Effet |
63+|---|---|---|
64+| `config=<chemin>` | vide | Le fichier de configuration, comme `AGENT_CONFIG` de `mm` : relatif à l'espace de travail, ou absolu dans le sandbox. |
65+| `tag=<version>` | la dernière version construite (`0.0.2`) | L'image `k33g/mm-sandbox:<version>` dont le sandbox est fait. |
66+
67+Le sandbox garde son nom (`mm-tui-<répertoire>` sauf `--name`) ; `sbx run --name <nom>` s'y rattache plus tard. À l'intérieur, la fenêtre est celle décrite dans [Utiliser le mode fenêtre](use-the-window-mode.md). Les sessions, quand `sessions.enabled` est actif, sont écrites sous `.mm/sessions` dans l'espace de travail — c'est votre répertoire hôte, elles survivent donc au sandbox.
68+
69+### 3. Depuis un éditeur (kit `mm-acp`)
70+
71+Un éditeur parle à `mm -acp` sur l'entrée et la sortie standard brutes de l'agent. `sbx run` ne peut pas porter cela : il alloue toujours un terminal. Le sandbox est donc créé une fois sans s'y attacher, et l'éditeur y lance `mm` avec `sbx exec -i`, qui transmet stdin et stdout tels quels.
72+
73+1. Depuis le projet que l'éditeur ouvrira, créez le sandbox avec le fichier de configuration intégré (`.` est l'espace de travail, sbx le rend absolu ; le chemin du kit doit être explicite : `./`, `../` ou absolu) :
74+
75+ ```bash
76+ cd /chemin/de/votre/projet
77+ sbx create --pull never --name mm-acp /chemin/de/mini-me/docker-sbx/kits/mm-acp . \
78+ --kit-arg config=agent.dmr.yaml
79+ ```
80+
81+ Un chemin d'espace de travail qui n'existe pas sur votre machine fait demander à sbx « The selected workspace does not exist. Would you like to create it? » — répondez `N` et corrigez le chemin.
82+
83+ `--kit-arg config` devient l'`AGENT_CONFIG` du conteneur, relatif à l'espace de travail parce que c'est là que `sbx exec` démarre `mm`. L'espace de travail est fixé à la création : un sandbox par projet, chacun avec son `--name`.
84+
85+2. Déclarez-le dans l'éditeur. Dans le `settings.json` de Zed :
86+
87+ ```json
88+ {
89+ "agent_servers": {
90+ "mini-me (sandbox)": {
91+ "type": "custom",
92+ "command": "sbx",
93+ "args": ["exec", "-i", "mm-acp", "mm", "-acp"]
94+ }
95+ }
96+ }
97+ ```
98+
99+ Par rapport à [la configuration côté hôte](use-from-an-editor.md), la commande est `sbx` et aucun `env` n'est nécessaire : `AGENT_CONFIG` est déjà dans le conteneur. Si l'éditeur ne trouve pas `sbx`, mettez le chemin absolu donné par `which sbx` dans `"command"`.
100+
101+3. Ouvrez le panneau agent et choisissez **mini-me (sandbox)**. L'éditeur lance `mm -acp` dans le sandbox à chaque conversation ; `sbx exec` démarre d'abord le sandbox s'il est arrêté. Appels d'outils, dialogues de permission, mentions `@fichier` et menu des commandes fonctionnent comme sur l'hôte — la différence est l'endroit où les commandes s'exécutent.
102+
103+S'attacher à ce sandbox avec `sbx run --name mm-acp` ne lance pas `mm` : son point d'entrée affiche le câblage ci-dessus avec le vrai nom et les vrais chemins du sandbox, puis ouvre un shell à l'intérieur. `sbx exec -it mm-acp bash` est le chemin le plus court vers ce shell.
104+
105+## Le serveur de modèles reste sur l'hôte
106+
107+Dans un sandbox, `localhost` est le sandbox lui-même. Votre serveur de modèles continue de tourner sur votre machine et se joint par `host.docker.internal`, que le proxy du sandbox fait pointer vers le `localhost` de l'hôte.
108+
109+- **Docker Model Runner** — rien à changer : `agent.dmr.yaml` a déjà `fallback: http://host.docker.internal:12434/engines/v1`, et `mm` bascule dessus quand `localhost:12434` ne répond pas (la bannière dit quelle URL a gagné).
110+- **llama-server** — `agent.llamacpp.yaml` n'a pas de fallback. Soit vous changez son `baseUrl` en `http://host.docker.internal:8080/v1`, soit vous passez `--env AGENT_BASE_URL=http://host.docker.internal:8080/v1` à `sbx run` / `sbx create` (voir [Faire tourner sur llama.cpp](run-with-llama-cpp.md)).
111+
112+Les deux kits autorisent les ports 12434 et 8080 de l'hôte dans la politique réseau du sandbox. Pour un autre port : `sbx policy allow network host.docker.internal:<port>`.
113+
114+## Publier une version
115+
116+Le workflow Release (`.github/workflows/release.yml`, job `docker-sandbox`) publie l'image et les kits à chaque tag `v*`, après le job des binaires. Il a besoin de deux secrets du dépôt, exposés au job comme variables d'environnement : `DOCKER_HANDLE`, le compte Docker Hub (aussi le propriétaire de l'image), et `DOCKER_PAT`, un jeton d'accès en lecture/écriture. Il lance ensuite les scripts qu'un portable lancerait :
117+
118+```bash
119+docker login -u k33g # coller un jeton d'accès
120+PUSH=1 ./docker-sbx/build.sh v0.0.2 # buildx, linux/amd64 + linux/arm64, poussée comme k33g/mm-sandbox:0.0.2
121+./docker-sbx/push-kits.sh v0.0.2 # sbx kit validate + sbx kit push → k33g/mm-tui:0.0.2, k33g/mm-acp:0.0.2
122+```
123+
124+`push-kits.sh` a besoin de la CLI `sbx` sur le PATH (ou `SBX=/chemin/vers/sbx` ; le workflow télécharge l'archive Linux de `docker/sbx-releases`). `sbx kit push` parle directement au registre et s'authentifie depuis le magasin d'identifiants Docker — aucun démon sandbox n'intervient, c'est ce qui le rend utilisable en CI. Les deux scripts appellent d'abord `sync-kits.sh`, qui réécrit `version`, le `tag` par défaut et le propriétaire de l'image des kits, de sorte qu'un kit publié nomme toujours l'image poussée avec lui. `IMAGE_OWNER=<compte>` publie sous un autre compte.
125+
126+## Variantes
127+
128+- **Une autre configuration pour une session d'éditeur.** Ajoutez `"-e", "AGENT_CONFIG=agent.lsp.yaml"` avant le nom du sandbox dans `"args"` ; `sbx exec -e` remplace la valeur intégrée pour cette exécution seulement.
129+- **Skills.** `skillsDir` est relatif au répertoire où `mm` tourne, c'est-à-dire l'espace de travail dans les deux kits — la même règle que sur l'hôte.
130+- **Docker dans le sandbox.** Construisez le template sur le shell Docker-in-Docker : `SHELL_TEMPLATE=docker/sandbox-templates:shell-docker ./docker-sbx/build.sh`.
131+- **Lire les logs.** En mode fenêtre ils sont à l'écran ; en ACP ils vont sur stderr, que Zed montre sous `dev: open acp logs`.
132+- **Supprimer un sandbox.** `sbx rm mm-acp`. Le répertoire du projet est un montage et n'est pas touché.
133+
134+## Voir aussi
135+
136+- Le format suivi par ces deux fichiers : la spécification des kits de Docker Sandboxes (`schemaVersion: "2"`), livrée avec les sources de `sbx`.
137+- Ce que `mm -acp` expose à l'éditeur : [référence ACP](../reference/acp.md).
138+- Les drapeaux et variables d'environnement que `mm` lit : [référence de la ligne de commande](../reference/cli.md).