Rig — Testing skill for Claude Code

rig

[](https://github.com/alexjpower74-create/rig/actions/workflows/test.yml)

Orchestration for builds where several Claude Code sessions work the same repo at once.

**Rig 3.0 builds the workflow's desk in:** review before merge, a QA worktree per project and per run that never refuses, negative controls re-proved on the sha being called done, one GitHub issue per slice opened and closed by the rig, the app registry and the decision log written at the finish gate, and `rig roll` for one brief across many repos. See [CHANGELOG.md](CHANGELOG.md) for what changed and why.

Three sessions in one checkout will quietly overwrite each other, review their own blind spots, and hand you a green test suite for a broken page. `rig` is the set of rules that stops each of those, made executable. Every one of them came from a build where the absence of it cost real time.

rig init     scaffold a brief-first PLAN.md, the AGENTS.md rulebook, .rig/config.json, the registry file
             (--hook: pre-commit guard; --slug, --name, --no-registry)
rig up       a worktree + branch + GitHub issue per slice, an agent launched in each  (--no-issues)
rig status   agent state, commits ahead, uncommitted work, anything edited outside its slice
rig guard    refuse work that reaches outside its slice
rig qa       rig qa  --run "" --negative "": grade from a clean worktree pinned to that
             commit, one per run (qa, qa-2…); record both exits  (--fresh: git clean -x first)
rig review   rig review : a read-only, fresh-eyes review brief for one slice's diff
rig finish   the done gate: reviewed, merged, clean, QA and negatives green on this sha; closes the
             issues, jots, touches the registry, writes docs/FINISH.md  (--no-review, --no-desk, --wrap "")
rig rule     rig rule "": add a learned rule to the rulebook every agent reads
rig brief    print an