@adlc/init

Idempotently bootstrap the committable .adlc/ runtime and the project-scoped .codex/agents/ roles used by the native Codex integration.

From the current checkout (before the matching suite release):

node packages/init/bin/adlc-init.mjs --root /path/to/repo

After the matching @adlc/cli release:

adlc init
adlc init --root /path/to/repo --json

The initializer preserves existing files, refuses symlink escapes from the project root, and never edits user-global ~/.codex implicitly.

Security modes

Every scaffolded .adlc/config.json sets securityMode: "unsigned-fallback". This is deliberate, not an oversight: the alternative, "signed", requires a dedicated signed-runner pool (ADLC_SIGNED_RUNNER_POOL=1, a pinned runner binary sha256) that cannot be bootstrapped by adlc init itself — a repo's very first .adlc/ commit is refused in signed mode (packages/rails-guard/lib/ci/bootstrap.mjs) and must upgrade into it later through a protected-base runner ceremony.

What unsigned-fallback actually weakens: gate-manifest evidence (.adlc/manifest.jsonl / manifest.d/) is accepted without a cryptographic signature from that runner pool — a P5 review record, a mutation-gate pass, or a cross-model attestation is trusted at face value rather than verified against a signed chain. signed mode closes that gap once the runner-pool ceremony has run. Repos with a real trust boundary between the agent and the CI runner (untrusted contributors, a compliance requirement) should plan the upgrade; unsigned-fallback is the correct starting point for everyone else.