run

ForgeProof: Provenance-Tracked Code Generation

Execute a cryptographically signed development workflow. Every action is recorded in a tamper-evident Ed25519 hash chain. The output is an .rpack provenance bundle that proves what was done, why, and that nothing was altered after signing.

The provenance engine script is at ${CLAUDE_PLUGIN_ROOT}/skills/run/scripts/forgeproof.py. Reference it as $FP in all commands below for brevity.

Interpreter setup (do this once, before anything else): determine the Python interpreter: run python3 --version; if that fails or reports that Python is not found, run python --version. Set $FP_PY to whichever succeeded and use it for every engine invocation below. The examples below use bash syntax ("$FP_PY" "$FP" <subcommand> ...); if your shell is PowerShell, adapt the invocation (& $FP_PY $FP <subcommand> ...) and any command substitutions accordingly — execute the intent, not the literal bash syntax.

Issue Selection

If $ARGUMENTS is empty (no issue number provided):

1. Run: "$FP_PY" "$FP" issues --assignee @me
2. Present the list to the user as a numbered list showing issue number, title, and labels
3. Ask the user to pick one
4. Set $ISSUE to the chosen issue number and continue

If $ARGUMENTS contains an issue number, set $ISSUE to that number and continue.

Phase 0 — Preflight

Run the dependency check:

"$FP_PY" "$FP" preflight

If any check fails, stop and tell the user exactly what is missing and how to install it. Do not proceed until all checks pass.

Then detect the project toolchain:

"$FP_PY" "$FP" detect

Parse the JSON output. Store the detected test_runner.command and linter.command for use in Phase 3. If no language is detected, ask the user to specify their test command and lint command.

Phase 1 — Parse & Plan

Fetch the issue:

gh issue view $ISSUE --json title,body,labels,assignees,comments

Read the issue body carefully. Extract structured requirements from it. Number them as REQ-1, REQ-2, etc. Each requirement should be a single, testable statement.