PR Title and Description Generator
Generate or update GitHub Pull Request titles and descriptions from the actual code changes in the final state. Use whenever a PR is being opened or its title/description written — when the user says 'open a PR', 'create a PR', 'make/raise/submit a PR', 'update the PR', or mentions generating or writing PR descriptions, titles, or summaries. If no PR exists yet it pushes the branch and creates one; otherwise it updates the existing open PR. Analyzes the git diff to document what's actually in the code, not just commit history.
next
Suggest the best next issue to work on. Considers workflow stage (Planned before Backlog), dependency graph, triage status, and what's unblocked. Optionally filters to a label — `/next specified` restricts to certified specs (what /auto runs). Use when the user says 'what's next', 'next issue', or invokes /next.
merge-queue
Inspect and drain the local deferred-merge queue. Shows worktree merges that /finish deferred (transient block — e.g. the main checkout was on the source branch with WIP) and that a local launchd drainer retries until they land. Use when the user says 'merge queue', 'what merges are pending', 'drain the merge queue', or invokes /merge-queue.
keeper
Interactive pickup for the config work autonomous runs cannot ship, with a mode per machine. On the keeper's machine (git -C ~/.claude config reflect.keeper prints true) it adjudicates three queues in one pass — uncommitted local changes in ~/.claude (typically /reflect's apply-now edits left for review), every `keeper`-labeled Linear issue workspace-wide, and open contributor proposal PRs on alienfast/claude — applying what holds up, merging accepted PRs, and committing and pushing the accepted set: invoking /keeper IS the commit/push grant for that adjudicated set (standards/git.md § Named exceptions). On any other machine it runs contributor mode instead — gathers the user's OWN keeper-labeled filings, adjudicates them for global value, builds the accepted set on a proposal branch in a linked worktree (the live checkout never leaves main), opens a PR to alienfast/claude, and closes the filed issues (Done with the PR link when included, Canceled when withdrawn); the grant is proposal-branch-only there, never main. Interactive-only; never runs unattended. Use when the user says 'keeper', 'process the keeper queue', 'review the keeper batches', 'what's waiting on me', 'propose my improvements', 'send my config changes upstream', or invokes /keeper.
quality-review
Adversarial implementation review with triage and fix loop, tiered by role. Hard-gates on `pnpm check`, delegates initial discovery to the quality-reviewer agent (opus/xhigh) for categorized findings (Critical/High/Medium/Nice-to-Have/Approved), then triages and fixes findings — mechanical, prose-only, and gate-clearing fix-now items applied directly by the orchestrator, substantive fixes dispatched to sonnet-tier developer agents. Re-reviews and confirmations run on the lighter quality-verifier agent and verify the fix delta (carried as an inline diff, or as a precomputed diff file above 200 lines) rather than re-reviewing from scratch; after cycle 2 only fix regressions and Critical/High findings on the change re-arm the loop (new Mediums are filed as severity-carrying deferred items), so the loop converges when a re-review surfaces no new substantive findings (an all-prose/mechanical batch — the initial review's included — converges without a re-review at all) — with a soft ceiling of 5 cycles that drains the mechanical/prose lanes before asking the user how to proceed (in auto mode, a decreasing recent findings trend instead extends the ceiling once); option 3 of that prompt terminates with verdict `escalated-to-architect`. Autonomous mode via the `auto` token (prompts resolve to documented defaults; used by /auto via /start auto). Use when the user says 'review my work', 'check this implementation', 'adversarial review', 'quality review', or invokes /quality-review.
reap-worktrees
Inspect and reclaim leftover /start wt worktrees. Shows which worktrees under .claude/worktrees/ are eligible for cleanup (PR merged, branch merged, or Linear issue Canceled/Done) and which are preserved (active or abandoned-for-resumption), and can reap the eligible ones now. Use when the user says 'reap worktrees', 'clean up worktrees', 'what worktrees are leftover', 'prune worktrees', or invokes /reap-worktrees.
reflect
Continuous-improvement reflection on the just-finished session — captures generalizable lessons and reconciles stale config, then auto-applies the small/safe shared-config edits (user-level `~/.claude` targets only on the keeper's machine; project-level edits inside a /start wt worktree are check-gated and committed so they ride the issue merge) and proposes the larger ones — filing the proposals as a certified (`specified`) Linear issue (Backlog — the human plans; stage-first ranking keeps it fleet-eligible); keeper batches (any `~/.claude` target) file uncertified with `keeper` instead, since `/auto` cannot ship cross-repo config work. Three modes — session (default; reflect on this session's friction), sweep (audit a project's CLAUDE.md/rules against the actual codebase and de-duplicate), and fleet (batched post-fleet reflection over a whole run's evidence — invoked by /fleet-retro, the only scheduled reflection surface since the per-issue /quality-review tail was retired 2026-08-15). Use when the user says 'reflect', 'reflect on this session', 'what did we learn', 'reflect sweep', 'reflect fleet', 'audit the config', or invokes /reflect.
spec
Groom a Linear issue into a certified spec — research the issue and codebase context, interview the user for problem/outcomes/success criteria, rewrite the description in the canonical spec shape, and graduate it with the `specified` label that gates autonomous /auto pickup. Decision-shaped issues ("Decide X") take the decision-grade path instead — architect-backed research, a debate brief with recommendations, a recorded decision, and filed follow-ups rather than certification. With no args, surfaces the top-ranked uncertified issues (including the Triage inbox) to pick from. Interactive-only — never runs unattended. Use when the user says 'spec', 'spec next', 'groom this issue', 'certify PL-XX', 'spec the backlog', 'debate this decision issue', or invokes /spec.
standardize-tooling
>-
start
Start working on a Linear issue — check blockers, assign, move to In Progress, create branch, plan implementation, execute with checkpoint updates, review and triage findings. Autonomous mode via the `auto` token (skips plan approval; converts prompts to documented defaults; used by /auto). The `interactive` token (implies worktree mode) sets up the worktree and claims the issue, then hands off to the user without planning/implementing, so you can work in an isolated worktree without interfering with background `wt` agents. Use when the user says 'start issue', 'work on PL-XX', 'begin PL-XX', or invokes /start.
update
Bring this machine current — pull the latest project code and ~/.claude, then run the update script (the project's .claude/update.sh when it has one, ~/.claude/update.sh directly when it doesn't) to update Claude Code, skills, and the CLI tools. A start-of-day habit, and the first thing to try when a tool is missing or a skill behaves like an older version. Use when the user says 'update', 'update my setup', 'pull the latest', 'bring me up to date', or invokes /update.
finish
Finish a Linear issue — check off requirements, add completion comment, commit/push, mark Ready For Release. Autonomous mode via the `auto` token (every prompt resolves to the conservative default — abort, never override; used by /auto). Use when the user says 'finish issue', 'done with this issue', 'complete PL-XX', or invokes /finish.
fleet-forecast
Estimate — explicitly not a plan — of what a fleet of parallel /loop /auto sessions would ship over a time horizon. Simulates the certified backlog draining across N sessions (blockers resolve and make way for their dependents), reporting the projected pick order as waves, roughly when the Planned/Todo stage burns down and Backlog picks begin, what a given horizon (e.g. 12h) cannot reach, and which candidates are stranded behind blockers the fleet can never ship. Read-only; nothing is launched or written. Use when the user says 'fleet forecast', 'forecast the fleet', 'what would a fleet run look like', 'what order would issues run', 'how many hours to burn the Planned issues', 'what would ship overnight/in 12 hours', or invokes /fleet-forecast.
fleet-launch
Launch a fleet of parallel /loop /auto sessions as background agents in `claude agents`, staggered so each session's first pick sees the previous session's claim, with an optional time budget that winds the fleet down cleanly (in-flight issues finish; no new picks). Count defaults to /auto-prep's persisted recommendation; an explicit count is the quota throttle. Ending a running fleet early is /fleet-stop, not this skill. Use when the user says 'fleet launch', 'launch the fleet', 'spawn N auto loops', 'start 5 loops for 10 hours', or invokes /fleet-launch.
fleet-retro
Post-mortem on a finished fleet of parallel /loop /auto sessions — measures each session with scripts/fleet-metrics.py (blind-sleep burn, dispatch mode, heartbeat compliance, classifier blocks, state-vs-reality drift, review churn with findings origins and the implementing-tier join, token and estimated-dollar attribution by agent type and model plus the developer-lane split (implementation vs fix batch) — cache-aware, with main-loop thinking share, cost per shipped issue, the context-size distribution (the autocompact gauge), shipped-issue provenance (the treadmill share), and a cross-run trend ledger diffing the last six fleets' headline gauges), reconciles the shipped ledger against git and Linear, audits the issues the run FILED for duplicates and stranded states, then reports ranked findings and applies the fixes you approve. The bookend to /auto-prep. Use when the user says 'fleet retro', 'review the fleet run', 'how did the fleet do', 'post-mortem the auto run', or invokes /fleet-retro.
fleet-status
One-screen readout of a running (or just-finished) /fleet-launch — time remaining on the deadline, per-session shipped/canceled/failed ledgers with liveness, in-flight issues from live worktrees, the shipped ledger cross-checked against git and the merge queue, stalled flags, and remaining certified runway. Read-only. Use when the user says 'fleet status', 'how's the fleet doing', 'what has the fleet shipped', 'what's in flight', 'how much time is left on the fleet', or invokes /fleet-status.
fleet-stop
Gracefully wind down a running fleet of /loop /auto sessions — writes an already-passed deadline so every session finishes its in-flight issue, then stops picking new work. Nothing is killed. Quota rationing is the primary use. Use when the user says 'fleet stop', 'stop the fleet', 'wind down the fleet', 'end the fleet early', 'discontinue the fleet', or invokes /fleet-stop.
full
End-to-end Linear issue macro — runs /start then /finish in sequence, gated on the /quality-review verdict. Worktree mode is opt-in via the `wt` token, mirroring /start. Pauses only for plan approval and the deferred-items filing decision; otherwise autonomous. The `auto` token removes those two gates too (documented defaults instead of prompts; used by /auto). Use when the user says 'full PL-XX', 'ship PL-XX end-to-end', or invokes /full.
linear
Linear via linear-cli — MUST READ before running Linear commands, especially for reading comments or dependencies (the obvious commands silently miss both)
auto
Autonomous Linear-backlog iteration — ships exactly ONE issue per invocation (preflight → /next specified → /full auto wt → record outcome), with a skip-and-circuit-breaker failure policy. Accepts an optional issue ID (`/auto BF-123`) for targeted mode — skip the pick and ship that specific certified issue unattended. Run continuously via `/loop /auto`; the loop ends itself on NO-CANDIDATES or AUTO-HALTED. Invoking /auto IS the run-scoped commit/push grant (standards/git.md). Use when the user says 'auto', 'work autonomously', 'work through the backlog', 'ship the next issue unattended', 'ship BF-123 unattended', or invokes /auto (typically as /loop /auto).
checkpoint
Save progress checkpoint — git commit WIP and update Linear issue with progress comment. Use when the user says 'checkpoint', 'save progress', 'wip commit', or invokes /checkpoint.
dependency-updater
Runs dependency updates end to end: parallel per-package research scaled to semver, applies the whole set at once, then fixes what the quality gate actually reports. Files a Linear issue to track the run, then asks up front whether to work in an isolated worktree and whether to finish by merge or PR, shipping through /finish and removing its own worktree when done. Invoked when users request package updates, dependency updates, version bumps, or mention 'ncu' or npm-check-updates.
auto-prep
Prepare a team's certified backlog for a fleet of parallel /loop /auto sessions — audit `specified` labels for unattended-shippability (route human-gated issues by shape — `needs decision` for pending decisions, `human` for human-performed work — mark fleet-hostile ones `solo`, flag decision-gated ones), consolidate same-defect-family point fixes into class-scoped sweep issues, wire `blocks` edges between file-colliding candidates, validate through next-candidates.sh, audit the release scope (every unstarted Planned/Todo issue classified workable/needs-keeper-action/draining, blocker chains walked to their root causes, fan-out first — the keeper's personal unblock list leads the report) plus the second-order certified-Backlog fleet-drain edges, and recommend a parallel-session count. Use when the user says 'auto-prep', 'fleet prep', 'prep the backlog for auto', or before launching multiple /loop /auto sessions.
Deprecation Handler
Handles deprecated APIs, types, and modules by applying safe migration patterns. Use when encountering deprecation warnings, migrating from deprecated code, updating dependencies with breaking changes, or modernizing legacy code to use current APIs.
k8s-installer
自動化安裝 Kubernetes 叢集的 AI Agent Skill
chezmoi
Manage dotfiles with chezmoi. Use when working with dotfiles, chezmoi templates, machine-specific configuration, syncing configs across machines, or when the user mentions chezmoi, dotfiles, or configuration management.
gcp-troubleshoot
Troubleshoot GCP services using tool-first access (via MCP when available), falling back to the CLI only when necessary. Focus on Firestore, Cloud Run, networking, load balancers, IAM, Pub/Sub, Cloud SQL, and Storage.
executing-plans
Use when given a complete implementation plan to execute in controlled batches with review checkpoints - loads plan, reviews critically, executes tasks in batches, reports for review between batches
knowledge-base
Maintain a markdown knowledge base at ~/git/knowledge-base/. Use when the user asks to interact with 'kb' or 'knowledge base' - adding content, deleting obsolete information, searching, or maintaining organization. Triggers on phrases like "add this to the kb", "delete from kb", "update kb", "maintain kb", "search kb", or any knowledge base operation.
get-secret
Retrieve and decrypt secrets from the local encrypted credential store (~/.get_vars.json). Use when the user needs to access stored credentials, API keys, passwords, or other secrets that have been exported from 1Password. This reads from the encrypted JSON file created by the get_vars.sh script.
kubernetes-troubleshoot
Troubleshoot and manage Kubernetes clusters, including resource inspection, debugging, pod logs, events, and cluster operations. Use when the user needs to diagnose issues, inspect workloads, analyze pod failures, or perform Kubernetes cluster operations.
loki-troubleshoot
Help craft efficient Grafana Loki / LogQL queries for debugging logs — with label‑based filtering, narrow time windows, and best‑practice guidance to avoid expensive or overly broad scans.
mimir-prometheus-troubleshoot
Help craft efficient Mimir/Prometheus queries, troubleshoot metric issues, avoid high-cardinality problems, and recommend best practices for aggregation, recording rules, and performance.
postgres
PostgreSQL database management for day-to-day operations. Use when the user needs to run SQL queries, inspect schemas, manage tables, monitor performance, handle backups, or administer a PostgreSQL database via psql or DATABASE_URL.
pr-create
Use when the user asks to create a pull request or PR for their work - creates PR using available tools with clear, concise title and description following standard template
repo-documenter
Provide repository-wide documentation guidelines under `docs/`. Use tool-based inspection first. Keep README.md updated with a brief project overview and links to detailed docs created under `docs/`.
search-notes
Always use this skill at the start of a task to check whether relevant information already exists. Use when users ask to recall, search memories, or subscribe to updates. Triggers on "recall", "search memories", "list keys", "share", "subscribe to", or any persistent storage request.
sql-server
Microsoft SQL Server database management for day-to-day operations. Use when the user needs to run SQL queries, inspect schemas, manage tables, monitor performance, handle backups, or administer a SQL Server database via sqlcmd or a connection string.
test-driven-development
Use when implementing any feature or bugfix, before writing implementation code - write the test first, watch it fail, write minimal code to pass; ensures tests actually verify behavior by requiring failure first
user-story
Creates well-structured user stories for software development and project management. Use when the user asks to write, create, or format a user story, or needs to document requirements, features, or tasks in user story format.
using-git-worktrees
Use when starting feature work that needs isolation from current workspace or before executing implementation plans - creates isolated git worktrees with smart directory selection and safety verification
writing-plans
Use when design is complete and you need detailed implementation tasks - creates comprehensive implementation plans with exact file paths, complete code examples, and verification steps assuming minimal codebase familiarity
writing-skills
Use when creating new skills, editing existing skills, or verifying skills work before deployment - applies TDD to process documentation by testing with subagents before writing, iterating until bulletproof against rationalization
1password
Manage personal secrets and passwords using 1Password CLI (op). Use when the user asks to query, retrieve, create, or manage secrets in 1Password, 1p, or op. This is for personal secrets only - not for cloud provider secret managers like Azure Key Vault, AWS Secrets Manager, or GCP Secret Manager.
add-note
Use this skill whenever important information is learned during a task or when the user explicitly asks to store something. Use when users ask to remember. Triggers on "remember this", "update memory", "share" or any persistent storage request.
tempo-troubleshoot
Help craft efficient Tempo trace queries, troubleshoot distributed traces, link logs and metrics (Loki/Mimir), and perform request-level root cause analysis.
architecture-documenter
Maintain and update architecture documentation in docs/architecture/. Uses tool-based inspection first, falling back to CLI only when needed.
qmd-knowledge-base
Maintain a markdown knowledge base at ~/git/knowledge-base/. Use when the user asks to interact with 'kb' or 'knowledge base' - adding content, deleting obsolete information, searching, or maintaining organization. Triggers on phrases like "add this to the kb", "delete from kb", "update kb", "maintain kb", "search kb", or any knowledge base operation.
qmd
Search markdown knowledge bases, notes, and documentation using QMD. Use when users ask to search notes, find documents, or look up information.
aws-troubleshoot
Troubleshoot AWS services using the AWS CLI. Focus on EKS, S3, ECR, EC2, SSM, networking, site-to-site VPNs, IAM Identity Center, and IAM.
Page 1424 of 1709 · 85437 results
