unc-skills

Self-QA a repo's running app end-to-end — discover run/auth/features, build a feature×modality test matrix, execute it live, and produce a witnessed pass/fail report.

1 skill

Find what a diff breaks beyond the diff and prove the one safety fact by running real code instead of writing it up.

1 skill

Cascading development loops — plan, advance, or take over a chain of bounded, evidence-gated development loops.

1 skill

Portable evidence-backed codebase cleanup with Peter O'Malley's official Desloppify engine.

1 skill

Route AI coding-assistant questions and approvals to a phone call through Vapi. Works with Claude Code, Codex, and pi.

1 skill

Multi-model coding orchestration: plan, route to the cheapest capable executor model, verify with code, review with fresh eyes.

1 skill

Render Parable's multi-model cast or solo contract inside Claude Code at session start.

hooks

Recap a task before doing it — a grounded, past-tense account of the finished work that long-running agents check themselves against for drift.

1 skill

Indexed search over Claude Code + Codex session history — find, continue, repeat, or skill-ify prior work

1 skill

Evidence-backed end-to-end comprehension of one exact coding-agent session.

1 skill

Drive every reviewer thread on a GitHub PR (Greptile, Devin, humans) to zero unresolved in bounded iterations, with a fixed-format final report.

1 skill

Test-driven development at pre-agreed seams: red before green, one vertical slice per cycle, regression-first bug fixes.

1 skill

Keep Slack threads attached to their Codex, Claude Code, Zellij, Hermes, or headless agents without exposing Slack credentials.

1 skill

Cut AI tells from text written or edited for a human reader before committing, posting, or sending it.

1 skill

Verify that the intended release candidate is live in its target environment and that the changed path works there, with live evidence.

1 skill