Skip to content

Repository structure

Policy · Canonical English日本語

The policy branch of TakashiSasaki/templates is an orphan history with no common ancestor with the repository's independent site and composition authority histories. The branch contains the Policy toolchain and one repository-facing agent-policy skill. Agent Skill and Web application artifact semantics are owned by Composition rather than separate provider branches.

The skill is under skills/agent-policy/. Its runtime manifest pins a reviewed full commit SHA and the SHA-256 of that revision's runtime dependency lock, so it never executes the mutable policy branch tip.

The policy branch

The placeholder below displays the complete Git-tracked tree of the policy branch. During documentation publication, the preview manifest is generated from the same commit.

Loading tree…

Main directory responsibilities

Path Responsibility
policy/ Canonical shared policy. Each rule is maintained as Markdown with YAML front matter.
profiles/ Declares named selections of shared policy modules for an agent operation or risk context. Profiles determine which shared rules are included; rule metadata determines final rule ordering.
schemas/ JSON Schemas for product-repository .agent-policy.yml, adoption state, and stable release metadata.
src/agent_policy/ Canonical Python CLI implementing unified adopt inspect/prepare/preview/finalize, validate, render, and check; the hidden init command remains an internal fresh-adoption primitive.
templates/ Generation templates for AGENTS.md, product-specific policy, consumer workflows, and related outputs.
skills/agent-policy/ Single installable repository-facing skill for unmanaged bootstrap, managed command dispatch, immutable pin selection, and persistent runtime-cache management.
tests/ Validation for configuration, adoption transactions, rendering, lock state, path safety, release identity, runtime distribution, cache behavior, and single-skill trust boundaries.
docs/ Adoption guidance, architecture, ADRs, PWA assets, and repository-preview UI.
scripts/ Branch-maintenance, release verification, runtime-distribution verification, and publication helpers.
.github/workflows/ CI and build-only documentation validation for policy. This branch has no Pages deployment authority.

For a selection guide and the complete current profile catalog, see Policy profiles.

Single agent-policy skill

skills/agent-policy/
  SKILL.md
  README.md
  runtime-manifest.json
  build-closure.json
  scripts/
    bootstrap.py
    build_closure.py
    install.py
    prepare_runtime_wheel.py
    run.py
    runtime.py
    uninstall.py
Path Responsibility
SKILL.md Defines when to bootstrap an unmanaged repository, when to run managed commands, cache/pin semantics, and migration-finalization safety.
runtime-manifest.json Pins the stable full SHA of TakashiSasaki/templates, the SHA-256 of its requirements-runtime.lock, stable project identity, and the closed bootstrap route set. It contains no finalize route.
build-closure.json Binds exact pip frontend, Hatchling backend, and active transitive build wheels by filename, version, direct URL, and SHA-256.
scripts/build_closure.py Validates the reviewed wheel closure, build-system declaration, artifact bytes, and prebuilt wheel record.
scripts/prepare_runtime_wheel.py Uses a disposable builder and the reviewed closure to build the exact selected Policy source without index resolution or build isolation.
scripts/runtime.py Resolves the stable or repository-pinned full SHA, includes the build closure in runtime/cache identity, constructs or reuses the persistent cache, and verifies the exact installed distribution set.
scripts/bootstrap.py Inspects unmanaged repository state and applies either the state-derived fresh adoption or migration adoption strategy when authorized. Migration bootstrap stops after preview.
scripts/run.py Runs normal managed commands through the cached runtime selected from .agent-policy.lock.
scripts/install.py Atomically installs or replaces the skill from a reviewed checkout after identity and path-safety checks.
scripts/uninstall.py Removes an installed skill after checking its identity marker.
tests/test_agent_policy_skill.py Verifies pin precedence, cache identity, offline cache hits, environment isolation, bootstrap safety, managed dispatch, and install/uninstall guards.

Runtime and control transfer

before adoption
  installed agent-policy skill
      ↓ stable full SHA + runtime-lock digest from runtime-manifest.json
  validated persistent runtime cache
      ↓ adopt inspect
  ├─ unmanaged-empty
  │    └─ fresh adoption --apply
  │         └─ internal init primitive → managed
  │
  └─ unmanaged-existing
       └─ migration adoption prepare --apply
            ↓ preview and semantic review
          separate explicit adopt finalize --apply

after adoption
  same installed agent-policy skill
      ↓ full SHA from product-repository .agent-policy.lock
  validated persistent runtime cache
      ↓ normal managed commands
  .agent-policy.yml + generated outputs + repository-local CI

Before adoption, runtime-manifest.json supplies the reviewed stable default toolchain. After .agent-policy.lock exists, the same skill prefers the repository's full-SHA pin. A malformed or mutable managed pin fails closed instead of silently falling back.

Runtime cache identity includes repository, full revision, runtime-lock SHA-256, Python major/minor, and platform. A valid cache entry is reusable without network access; a cache miss is built in a staging directory and moved into place only after dependency and installed-set validation succeed.