Agent-readable docs index: /llms.txt. Full docs in one file: /llms-full.txt. Download /docs.zip to grep all markdown files locally.

Skills catalog

A skill is a markdown procedure your agent loads by name. Genie v6.261003.1 ships 18. genie install and genie update deliver them through skills.sh into every agent skills home Genie detects: ~/.claude/skills for Claude Code, and the shared ~/.agents/skills for Codex and the other agents that read it.
Claude Code runs a skill as /wish, Codex as $wish, and any other agent by its name or by a plain request. Each card shows both forms.

The loop

The skills that take a change from an idea to a pull request, plus council for the decisions along the way. Each has its own page.
brainstorm
Settles an open idea with you, one round at a time, into a DESIGN.md that an independent reviewer approves.
wish
Takes one decided task to a merge-ready pull request, or refuses it and names the route it needs.
work
Runs an approved WISH.md in dependency order, one scoped worker per group, and ends at an implementation review.
review
Reads a design, plan, diff or pull request without changing it and returns SHIP, FIX-FIRST or BLOCKED with evidence.
council
Puts a consequential decision through independent lenses, dissent included, and returns a synthesis. It edits nothing.
fix
Repairs the blocking gaps of a FIX-FIRST review within a bounded budget, each repair checked by a different reviewer.

More skills

authoring
Writes or revises a Genie skill so it meets the shipped contract: frontmatter, house size, starter card and runtime-neutral voice.
deslop
Removes what only adds reading work from the prose, code or interface in scope, keeping its meaning, voice and behaviour.
docs
Audits documentation and developer experience against the live product, and writes or fixes docs when you ask.
genie
Routes a Genie question, operation, bug or planned work to the right skill, and resumes a related wish.
genie-hacks
Searches the community's Genie patterns by problem or category, and helps you contribute your own.
merge
Finishes an in-progress merge or rebase by the intent of both sides, then re-runs the full gate on the merged tree.
refine
Rewrites a prompt with official OpenAI or Claude guidance, keeping its intent and scope. It never runs the prompt.
report
Investigates a failure to its root cause and hands the diagnosis to fix. It opens a GitHub issue only when asked.
research
Answers a question from the sources that own the facts, cites every claim and writes the notes into your repository.
skill-audit
Audits the shipped skill catalogue for overlap, staleness and drift, and proposes keep, improve, merge or retire for each skill.
verify
Proves a completion claim with fresh evidence: the gate's exit code, the real diff, the remote checks and the reviewer's verdict.
workfly
Turns a procedure worth repeating into a saved workflow, checked by a static test and adversarial refuters.

Workflows behind the skills

These skills start a saved workflow. In Claude Code the skill runs the workflow from ~/.claude/workflows, or from your repository's own .claude/workflows/ when it carries one. In Codex and other agents without saved workflows, the same skill runs the stages by hand.
SkillSaved workflow
brainstormbrainstorm
councilcouncil
docsdocs-audit
researchresearch-sweep
skill-auditskill-audit-sweep, and skill-intake for candidate skills from outside the catalog
wishwish
workflyworkfly
These workflows have no skill in front of them: evidence-gate, observability-review, pm-ledger-verify and test-simplify. Workflows describes each of the 12.