Skills

Skills are markdown files that teach Claude Code a reusable workflow. Each skills/<name>/SKILL.md file describes when to invoke the skill, what steps to follow, and which tools it may use. Claude Code discovers skills in ~/.claude/skills/ automatically.

Codex uses generated wrappers in codex-skills/<name>/SKILL.md. Those wrappers keep Codex frontmatter strict and point back to the canonical Claude skill in skills/, so the workflow stays in one place.

How to invoke a skill

Surface	Syntax
Local CLI	`/skill-name` or a bare keyword (e.g. `ardi`)
Plugin (other repos)	`/ai-config:skill-name`
`@claude` bot on PRs	`@claude skill-name` in a PR comment
Codex	skill trigger from `${CODEX_HOME:-$HOME/.codex}/skills`

Most skills also respond to plain-English triggers listed in their description field (e.g. “grab an issue”, “drive to clean”, “update memories and skills”).

Skill categories

PR / review workflow

Skill	Aliases	What it does
`ardi`	`dc`, `drive`, `clean`, `iterate`	Read the latest review, address/rebut/defer every finding, push, re-request review — loop until clean
`ard`	`adr`	Single ARD pass: address, rebut, defer, or acknowledge — without the iterate loop
`resolve-pr-threads`		Resolve inline review threads that are already settled, without a full ARD pass
`claim-pr`		Post a “paws off” comment before a work session; unclaim when done
`request-pr-review`		Add a human reviewer and @-mention them with context
`pr-status`		Report the current PR’s CI and review state
`pr-status-all`		Report status across all open PRs
`stack-prs`		Branch a new PR off an unmerged base PR, keep it in sync, and re-target it to `main` once the base merges

Issue management

Skill	Aliases	What it does
`brainstorm`		Socratic clarifying-question loop before any code/issue, ending in a plan file
`start-task`	`st`	File a tracking issue (if none exists), branch, implement, open PR, ARDI to clean
`grab-issue`	`gi`	Pick the highest-priority open issue and work it end-to-end
`grab-issues-in-parallel`	`gip`	Grab multiple issues and work them in parallel worktrees
`defer-issue`		File a follow-up issue for something out of scope in the current PR
`cancel-superseded`	`cb`	Close issues/PRs that a merged PR made redundant
`discussions`		Read and respond to GitHub Discussions topics; mark Q&A answers
`migrate-discussion`		Move an item between Discussions and Issues when it fits the other tracker better

Branch / sync

Skill	Aliases	What it does
`sync-pr-branch`	`sync`, `merge-main`, `resync-branch`	Fetch origin, merge main into the branch, push
`merge-it`	`merged`	Squash-merge a ready PR, then run post-merge cleanup
`mwc`	`merge-when-confident`, `maw`, `merge-at-will`	Grant session-scoped permission to merge fully-clean PRs without asking per PR
`post-merge`		After a merge: checkout main, pull, delete branch, run UMS
`clean-branches`	`prune`	Delete local branches already merged or closed
`resolve-conflicts`	`rmc`	Resolve merge/rebase/cherry-pick conflicts by consolidating the best of both branches

Memory and skills maintenance

Skill	Aliases	What it does
`ums`	`update-memories-and-skills`	Capture learnings from the session into memory files and skill definitions
`memorize`	`remember`, `always`	Save a specific fact or preference to memory
`push-memory`		Push a general-purpose memory into ai-config from a session working in another repo — on a branch + PR
`record-learnings`		Structured post-mortem: extract lessons from a completed task
`consolidate-memory`		Merge or deduplicate memory files that have drifted
`skill-audit`		Report which skills actually fire (and how often) vs. dead weight, and recommend pruning candidates
`learn`		Stage a candidate learning without yet deciding if it’s durable enough to commit
`promote-memory`		Review staged learnings and promote the durable ones into committed memory

Writing and style

Skill	Aliases	What it does
`use-preferred-style`	`style`	Apply the plain-prose writing guide to a draft
`find-ai-tells`	`ai-tells`	Scan text for AI-sounding patterns and flag them
`fact-check-prose`	`fcp`	Check prose claims, reasoning, and computed values against sources and rendered output
`fix-forward-references`	`ffr`	Detect and rearrange/reword forward-pointing references (“see below”) in prose
`detect-informal-definitions`		Find concepts defined with definitional precision that never got wrapped in a formal definition div
`detect-hypothetical-examples`		Find worked examples using invented numbers when a real dataset is already available
`use-math-macros`	`macroize`	Rewrite manuscript math onto the shared d-morrison/macros submodule
`tidy`		Clean up code or prose without changing behavior
`simplify`		Reduce a file or function to its essential parts

Repo / CI utilities

Skill	Aliases	What it does
`convert-repo-format`	`crf`	Convert a repo between the lab’s template formats (R package, Quarto website/book/manuscript)
`scout-peers`		Survey comparable repos for ideas and conventions
`check-dependency-updates`	`cdu`	Check for outdated dependencies and propose updates
`dependabot`		Add or update Dependabot config for a repo
`send-upstream`	`sup`	Open a PR or issue in an upstream repo for a bug found here
`reprexes`		Build minimal reproducible examples for a bug
`r-pkg-spellcheck`		Spell-check an R package’s documentation
`r-pkg-check`		Run `devtools::check()`/`R CMD check` and triage the results
`r-pkg-news`		Draft a new NEWS.md entry for an R package
`r-pkg-cran-checklist`		Walk through the standard CRAN submission checklist
`reproducibility-audit`		Audit a project for reproducibility gaps: hidden deps, hardcoded paths, undocumented prerequisites, output traceability

Session management

Skill	Aliases	What it does
`session-lock`	`deconflict-sessions`	Register/deregister this session to prevent parallel clobbering
`handoff`		Snapshot work state (branch, jobs, open decisions) so the next session can resume cleanly
`wrap-up`	`done`, `merged`	Verify true PR/issue/branch state, report a linked summary, then run UMS
`select-model`	`model-fit`, `assess-model-fit`	Choose the right Claude model for the task at hand
`away`		Grant session-scoped permission to stop asking clarifying questions, pick confident work, and consult a stronger model instead of blocking
`back`		Cancel the `away` grant and surface a decision log of what was resolved unattended
`delegate-to-codex`	`dtc`	Offload heavy read/draft/verify work to the `codex` CLI before spending Claude quota, with Claude fallback
`prompt-me`	`pm`	Surface the single most pressing open question (or top N) waiting on user input
`prompt-me-all`	`pma`	Restate every open question still waiting on user input, uncapped
`pending-decisions`	`pd`	Sweep issues/PRs (not just this conversation) for ones waiting on a decision from you, and ask about them one at a time
`checkpoint`		Save a deliberate mid-task snapshot without ending the session
`compress-session`		Distill the session into auto memory before context fills up
`permission-check`		Diagnose why Claude Code is (or isn’t) prompting for permission on an action

Full skill list

All 171+ canonical skills are in skills/ on GitHub. Generated Codex wrappers live in codex-skills/. Each canonical SKILL.md has a description field that lists the trigger phrases and a full procedure.