ARD + Iterate: apply the ARD framework within an iterate loop on a single PR/MR. Read the latest review, Address/Rebut/Defer every finding, push fixes, post the ARD summary, then re-request review — repeating until the verdict is clean. Runs a per-round checklist before each re-review trigger (latest review analyzed for current head, all findings dispositioned, fixes pushed and summary posted) and a fully-clean exit checklist before declaring clean (all CI green, zero findings, all inline threads resolved, only final all-clear exchange open). Use when asked to 'ardi', 'dc', 'drive', 'clean', 'drive to clean', 'iterate', 'iterate until clean', 'iterate this MR', 'drive this PR to clean', 'address the review comments', '@claude review again and fix what it finds', or after receiving a review you want to resolve completely / carry a PR all the way to mergeable.
Merge two or more genuinely-redundant memory entries in the `memories/` corpus into a single canonical entry — union the facts, keep one copy in the right scope, and repoint any `[[links]]` — so the corpus shrinks without losing a fact or a cross-reference. Delegates detection to `find-overlap` (scope = `memories/`), proposes a plan for approval, then ships it via branch + PR. The memory-corpus counterpart of `consolidate-skills`. Use when asked to "consolidate memory", "consolidate memories", "merge duplicate memories", "dedupe memories", or "collapse redundant memory entries". Invoke explicitly with /consolidate-memory.
Persist a fact or preference to memory that survives across sessions, machines, and agents — routed by relevance to project-specific or general scope. Use when the user says 'memorize', 'remember that …', '/remember', 'from now on …', 'always/never …', 'note that …', or 'add to memories'. (`remember` / `/remember` and `always` / `/always` are synonyms.)
Merge a ready pull request, then automatically wrap up — verify the merge landed, tidy the branch, and run UMS — without asking. Use when the user says 'merge it', 'merge this', 'merge the PR', or 'go ahead and merge'. Performs the actual merge when the PR isn't merged yet (squash by default), then chains into the post-merge skill. If the PR is already merged, it skips straight to post-merge.
Wrap up a just-merged PR/MR: verify the merge actually landed (never assume), tidy the local branch (switch to main, pull, delete the merged branch), confirm any deferred follow-up issues are tracked, then run UMS to capture what the PR's review lifecycle taught — mistakes corrected and guidance given along the way. Use right after a PR merges, or when asked to 'post-merge', 'wrap up the merged PR', or 'clean up after the merge'. For the directive to actually perform the merge ('merge it' / 'merge this'), use the merge-it skill, which merges then chains into this one.
Audit a target for hallucinations — concrete, checkable references that don't actually resolve (a missing file path, an undefined R function/object, a non-existent `uses:` action ref or version, a dead URL, a `[[memory-link]]` with no target, a fabricated skill name, citation, package, flag, config key, or SDK method) — then interactively propose a fix for each one. Conservative: only flags references PROVEN absent; unverifiable ≠ hallucination. Scope is any of the memory/instruction corpus (ai-config memories/, skills/, CLAUDE.md), the current repo's code & docs, or an explicit file / PR diff / pasted AI output. Use when asked to 'purge hallucinations', 'ph', 'check for hallucinations', 'verify the references', 'find made-up / fabricated references', 'fact-check this AI output', 'does everything in this file actually exist', 'audit my memories for stale references', or 'what did you make up'.
Push a general-purpose memory into the ai-config repo when you're working primarily in ANOTHER repo — a `CLAUDE.md` standing rule or a `memories/*.md` reference fact — delivered on its own branch + PR (or via the GitHub file API) without disturbing the repo you're in. Use when the user says 'push this to ai-config', 'remember this globally from here', 'add this to ai-config's memory / CLAUDE.md', 'record this in ai-config even though we're in <other repo>', or '/push-memory'. For the normal case where ai-config IS your working repo, use `memorize` instead.
Persist discoveries, debugging insights, and working patterns to memory and shared instruction files as you work. Ensures knowledge survives across sessions and is accessible to other AI agents via the shared ai-config repo. Use continuously — after solving a tricky bug, discovering a codebase convention, or learning a tool quirk.