skills

Convene a board of frontier AI models — Claude, Codex, Gemini, and Grok — that each review the same decision independently, debate across rounds, and hand back one clear recommendation.

30 skills

Find what a change breaks somewhere else before it ships: past where grep stops, risks graded with real file:line cites, and the one fact the change is safe because of proven by running real code.

30 skills

Portable skills for AI agents — a multi-model advisory board, a seventeen-skill team-workflow pack for tracked multi-session development, a media-ingest skill that turns recordings and URLs into evidence, a pair of writing references: documents agents consume, and pages humans read — and a plain-language decoder for agent messages that didn't land.

30 skills

Decode an agent message that didn't land — a plain-sentence restatement, every piece of invented shorthand expanded, what's being asked of you separated from what happened, and unevidenced claims flagged.

30 skills

Turn a video, recording, voice memo, or media URL into an evidence packet — authoritative whisper transcript, timestamped frames, manifest — plus a routing recommendation. Six intents behind a mandatory purpose gate; the pipeline script is resumable and encodes the hard-won gotchas.

30 skills

Write like a person in everything the agent emits: plain words, concrete claims, no AI tells. Always on for chat replies, status updates, close-outs, review comments, and commit messages; owns the repo's single merged tell catalog behind a pointer.

30 skills

Seventeen skills that ship as one pack — a portable discipline for tracked, multi-session, agent-assisted development: the orchestrator seat a non-engineer works from (one session routing, auditing, and merging parallel lanes), a router entry point, a once-per-repo setup interview that records whether you lead from that seat or read code and pick skills, per-repo institutional memory written as side effects of work, relentless grilling, decision maps, throwaway prototypes, autonomous research lanes, ticket filing, human-step wizards, session handoffs, a build discipline of seam-scoped test-first tracer slices, a disciplined bug-diagnosis loop where no fix ships without a named cause, an adversarial review that breaks changes before they ship, a skeptic-verified state review of the codebase itself, an append-only decision trail that lets a human audit an unattended run after stepping away, and a PR-writing standard that leads with the problem, shows before-and-after proof, and says whether a revert undoes the change. Versioned together (team-workflow/vX.Y.Z); this version mirrors the latest pack tag.

30 skills

Write and prune documents an agent consumes — a SKILL.md, an AGENTS.md or CLAUDE.md, a reference reached by a pointer. Context pointers, the two loads, the information hierarchy, leading words, and the no-op test.

30 skills

Write copy a human reads — a landing page, a README's front half, a launch post. Guide structure over catalog structure, the warmth moves, three named failure modes, standing voice rules, and a last-mile AI-tell scrub.

30 skills

24 Click AI skills for Codex desktop, adapted for Astra workflows: planning, research, writing, implementation, independent review, and checkpoint recovery.

24 skills