doidor/agentrig

address-review-comments

Read PR review feedback, fix or reject each requested change, reply per thread, resolve every thread, and verify zero threads remain.

no-self-approve

Never cast an approving review on a PR your own (shared) agent identity authored — approval must come from an independent reviewer.

resolve-conflicts

Rebase the PR branch onto the default branch, resolve conflicts keeping both intents, and verify mergeability before pushing.

harness-eval

Evaluate THIS repository's agent harness — a deterministic structure audit (A1) plus content quality probes (A2), plus an isolated producer/judge dynamic eval (B) with paired sign-test A/B variant comparison.

skill-authoring

Admission bar and structure for writing a new skill, so the skill library stays lean and discoverable.

skill-improver

Turn a reviewer/judge failure into an instruction-surface change that passes a prevention test.

log-gotcha

Record a newly-discovered gotcha to `.agents/wiki/` BEFORE handoff — the harness's feedback loop. The wiki is how the next agent doesn't repeat your discovery.

self-verify

Run the project's own build/test/lint and converge before handing work to a reviewer. Requires explicit baseline → after evidence — the suite must be shown to change state, not just be "green at the end".

verify-loop

General wait → inspect → fix (max 3) → self-park loop for any post-action verification (build, tests, CI, visual, lint).

fix-ci

Diagnose and fix a failing CI run for the current branch, then re-verify.