メニュー
Capsem development toolchain -- all just recipes, what they do, when to use which, and dependency chains. Use when you need to know how to build, run, test, or ship Capsem, or when deciding which just command to run for a given change. This is the toolchain reference.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | dev-just |
| description | Capsem development toolchain -- all just recipes, what they do, when to use which, and dependency chains. Use when you need to know how to build, run, test, or ship Capsem, or when deciding which just command to run for a given change. This is the toolchain reference. |
All workflows use just (not make). The justfile is the single entry point.
| Command | What it does |
|---|---|
just doctor | Check all required tools, colored output, structured recap |
just doctor fix | Doctor + auto-fix all fixable issues in dependency order |
just shell | Daily driver: cross-compile + repack initrd + build + sign + boot temp VM + shell (~10s) |
just exec "CMD" | Run CMD in a fresh temp VM (auto-provisioned and destroyed) |
just run-service | Start capsem-service daemon (builds, signs, launches or reuses) |
just ui | Tauri dev with hot reload (service + Astro dev server on :5173 in Tauri webview) |
just dev-frontend | Frontend-only dev server on :5173 (no Tauri, no VM, mock data) |
just build-ui [release] | Frontend build + cargo build -p capsem-app in lockstep. Use after any frontend change when running the Tauri binary directly. |
just run-ui -- [args] | build-ui then launch ./target/debug/capsem-app with args (e.g. --connect <id>) |
just build-assets [arch] | Full VM asset rebuild through capsem-admin/profile materialization and the private Python builder backend. Default: both arches. |
just smoke | Hermetic smoke gate: audit + doctor + injection + integration + parallel pytest groups |
just test | ALL tests: unit (warnings-as-errors) + cov + cross-compile + frontend + python + injection + integration + bench + install e2e |
just test-gateway | Gateway unit + mock-UDS tests (no VM needed) |
just test-gateway-e2e | Gateway E2E tests (real service + VMs) |
just test-install | Install e2e in Docker + systemd, then hermetic local release glow-up from generated stable/nightly channels |
just coverage | HTML coverage report across all Rust crates (opens target/llvm-cov/html/index.html) |
just cross-compile [arch] | Full Linux build in container (agent + deb) |
just benchmark | Standard artifact-recording benchmark suite across host-native, in-VM, lifecycle/fork/parallel, and Security Engine lanes |
just bench | Alias for just benchmark |
just inspect-session [args] | Session DB integrity + event summary |
just list-sessions | Table of recent sessions with event counts |
just query-session "SQL" [id] | Run SQL against a session DB (latest with a DB by default) |
just update-fixture <src> | Copy + scrub real session DB as test fixture |
just update-prices | Refresh model pricing JSON |
just update-deps | cargo update + pnpm update |
just logs | Tail ~/.capsem/run/service.log |
just sandbox-logs <id> | View process + serial logs for a specific sandbox |
just build-host-image | Build/refresh the capsem-host-builder Docker image |
just install | Build release .pkg/.deb + install it locally (postinstall handles codesign, PATH, service registration) |
just release [tag] | Wait for CI to build + publish a pushed tag |
just cut-release | Run test, bump version, stamp changelog, commit, and create a local tag |
just clean | Remove all build artifacts |
just clean all | clean + Docker prune (full reset) |
| What changed | Command |
|---|---|
| Rust host code | just smoke (E2E) or just test (full) |
| Guest binary (agent, net-proxy, mcp-server) | just smoke (auto-repacks initrd) |
capsem-init | just smoke (auto-repacks) |
In-VM diagnostics (guest/artifacts/diagnostics/) | just smoke |
Profile payloads (config/profiles/<id>/) or rootfs packages | just build-assets then just shell |
| Frontend components | just ui (iterate) then just test (validate) |
| Frontend standalone (no VM) | just dev-frontend |
| Tauri binary (not dev) | just build-ui then just run-ui |
| Telemetry pipelines | just exec "<cmd>" then just inspect-session |
| Gateway code | just test-gateway (unit) or just test-gateway-e2e (real VMs) |
| Service HTTP API / CLI / MCP | just smoke (parallel pytest groups cover all three) |
| Install / postinst / systemd / release glow-up flow | just test-install |
| Pre-release | just test |
| Ship | just cut-release, then manually push main and the tag |
shell -> _check-assets + _pack-initrd + _ensure-service (_sign + build)
ui -> _ensure-setup + _pnpm-install + run-service
run-service -> _check-assets + _pack-initrd + _ensure-service
exec -> run-service
build-assets -> _install-tools + _clean-stale (inline: doctor, capsem-admin image build)
build-ui -> _frontend-dist (pnpm build + cargo build -p capsem-app)
smoke -> _install-tools + _frontend-dist + _check-assets + _pack-initrd + _ensure-service
test -> _install-tools + _clean-stale + _frontend-dist + _generate-settings
+ _check-assets + _pack-initrd
bench -> _ensure-setup + _check-assets + _pack-initrd + _ensure-service
test-gateway-e2e -> _check-assets + _pack-initrd + _sign
test-install -> Docker package install + generated local stable/nightly glow-up
install -> _pnpm-install + _stamp-version + _check-assets + _pack-initrd
cut-release -> test + _stamp-version
_-prefixed recipes are internal (hidden from just --list).
_ensure-service honors CAPSEM_HOME / CAPSEM_RUN_DIR for isolated
smoke/test runs and assigns the gateway an ephemeral port in that mode. This
keeps test services from colliding with the user's installed gateway on the
default developer port.
Docker builds (build-assets, cross-compile, test-install) accumulate images, build cache, and stopped containers inside the Colima VM. The _docker-gc helper runs automatically after each of these recipes to prevent unbounded disk growth:
fstrim on the Colima VM disk to release freed space back to macOSThe Colima VM uses a Virtualization.framework raw disk that only grows, never shrinks on its own. Without fstrim, Docker prune frees space inside the VM but macOS never gets it back. This is why _docker-gc always trims after pruning.
For a full manual reset: just clean all (removes all build artifacts + aggressive Docker prune).
tauri::generate_context!() reads tauri.conf.json frontendDist: ../../frontend/dist and bakes every file under that directory into the Rust binary during cargo build. Consequences:
just smoke, just test) must build _frontend-dist before cargo clippy --workspace --all-targets; otherwise capsem-app fails during macro expansion if frontend/dist is missing.pnpm run build) has zero effect on a running ./target/**/capsem-app -- the binary still carries the old bundle.frontend/**, you must cargo build -p capsem-app for the change to reach the Tauri app.just ui (cargo tauri dev) sidesteps this by serving http://localhost:5173 directly -- no embedding happens in dev mode.just build-ui / just run-ui, never raw pnpm run build followed by re-running an already-compiled binary.Symptom you'll see when you forget: edits to Svelte/CSS don't appear in the window, but http://localhost:5173 in a browser shows the new version. That's the embed-vs-live split.
All build infrastructure (runner, code signing, generation scripts) logs to target/build.log. This is a unified diagnostic log -- never write to stdout from build scripts. The runner (scripts/run_signed.sh) and _generate-settings both append here.
When debugging build issues, check target/build.log first. When writing new build scripts or recipes, always log to this file, never stdout (which contaminates binary output like mcp-export).
just doctor # Check tools (colored output, shows fixable issues)
just doctor fix # Auto-fix missing targets, cargo tools, config files
just build-assets # Build kernel + rootfs (~10 min, needs docker)
just shell # Boot a temp VM and drop into a shell
Or use bootstrap which does all of this:
sh bootstrap.sh # Installs deps + runs doctor fix
just shell is the daily driver. It cross-compiles the guest agent, repacks the initrd, builds the host binary, codesigns, boots the VM, and drops into a shell. For a one-shot command use just exec "CMD". For UI iteration use just ui (Tauri dev with hot reload).
The capsem-builder Python package is the backend implementation. Product image
truth enters through capsem-admin and profile-owned config, not direct
builder authoring commands:
capsem-admin profile check --profile config/profiles/<profile-id>/profile.toml --config-root config
just build-assets # Build profile-owned VM assets through the profile-derived build rail
just _materialize-config # Materialize generated runtime profile config
The only public capsem-builder helper commands are backend support commands
used by just/CI: doctor, validate-skills, agent, and audit.
There is no public capsem-builder build, validate, inspect, --dry-run,
mcp, or render-only rail. If the product contract needs a new image input,
add it to the profile/corp/settings config model and the capsem-admin
validation path.
On macOS, agent binaries are compiled inside a Linux container (docker) via cross_compile_agent() in docker.py. This avoids needing rust-lld, musl targets, or llvm-tools on the host. On Linux (CI), cargo builds natively.
just cross-compile [arch] is a debug/verification tool that builds everything in a container: agent binaries, frontend, and the full Tauri .deb. It's not in the daily just shell path -- _pack-initrd calls cross_compile_agent() directly for agent-only builds.
Guest binaries target aarch64-unknown-linux-musl and x86_64-unknown-linux-musl. Per-arch named volumes (capsem-agent-target-{arch}) cache build artifacts separately to prevent cache clobbering.
Capsem benchmarking with capsem-bench and capsem-bench-rs. Use when running benchmarks, adding new benchmark categories, interpreting results, or investigating performance regressions. Covers benchmark categories (disk, rootfs, storage, startup, protocol/http, snapshot, all), the JSON output format, and how to add new benchmarks.
Asset building, manifest format, hash verification, and boot-time resolution for Capsem VM images. Use when debugging boot failures, manifest issues, hash mismatches, or understanding how assets flow from build to boot.
Capsem native package installer -- package install, service registration, self-update, manifest-driven asset download, corp config provisioning, and the install test harness. Use when working on package install/update/uninstall commands, service install/uninstall, asset management, corp config, install test infrastructure, or the installed layout (~/.capsem/).
Capsem release process, CI pipeline, Apple code signing, notarization, documentation site, and post-release verification. Use when preparing a release, debugging CI failures, working with Apple certificates, updating the documentation site, or cutting a new version. Covers the full release lifecycle from pre-release checklist through post-release verification.
Capsem system architecture -- service daemon, per-VM processes, CLI, MCP server, guest agent, vsock, network proxy. Use when you need to understand the system design to write code, review changes, write documentation, or debug cross-component issues. Covers the service architecture, IPC protocols, vsock ports, storage modes, network policy, MITM proxy, and key source files.
Capsem documentation site infrastructure and conventions. Use when writing, editing, or maintaining docs in the docs/ directory (docs.capsem.org), adding new doc pages, updating the sidebar, or working with Astro Starlight. Covers site structure, frontmatter, writing style, sidebar config, release pages, and dev workflow.