forge-adopt

Bring an existing codebase into the Forge pipeline. Adopt detects the project type, samples a bounded set of files, fans out extractors to infer an SRS and an architecture map, seeds pipeline/state.md, and enters the normal 12-stage flow at Stage 1 so you confirm the inferred artifacts before building on them.

When to Use

• User runs /forge:adopt on a repo that already has code
• User wants reverse-engineered SRS / architecture drafts to start from
• The project predates Forge and needs onboarding (not greenfield /forge:init)

Guarantees

• Read-only to your source. Adopt writes ONLY under pipeline/ and .forge/ — it never modifies your code.
• Everything is marked INFERRED with a confidence score and the files it was derived from. Forge proposes; you confirm.
• Bounded. Sampling is capped (adopt.max_files, default 40); what was sampled vs skipped is reported — no silent truncation.
• Refuses if pipeline/state.md already exists (that project is already initialized).

How to run

Preview first — --dry-run reports the plan and spends nothing:

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/adopt.py --cwd . --dry-run

Then run it for real (fans the extractors out via the orchestration primitive; the dimensions are cost-gated):

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/adopt.py --cwd .

It prints a JSON summary (project_type, sampled/skipped, written, dropped_aspects, cost_usd). Relay it, then tell the user the inferred drafts under pipeline/01-srs/ and pipeline/03-architecture/ are drafts to confirm, and that the pipeline is seeded at Stage 1. If any aspect was dropped, say so — don't imply full coverage. From here, /forge:srs / /forge:architecture refine the drafts with gates intact.