Latest Agent Skills
Everything added to the registry recently, newest first.
- eric-lingren SkillImprove Skill Learnings
Apply captured learnings from unified-learnings.jsonl to their target skill, agent, process, or contract files. Shows a ranked table of targets, lets the user pick one or "run all" (offered when TOTAL ≤ 10). A picked target with ≤ 10 learnings also offers "run all" for that target. Validation is lazy — runs per-learning just before drafting, not upfront in batch. Run-all loops through every target and learning automatically with a y/n per diff. Covers all four improves_type values: skill, agent, process, contract.
- eric-lingren SkillRun Task Followups
Wizard-style interactive walkthrough of manual follow-up actions from a tasks JSON file. Guides user through each follow-up one at a time with concrete steps (exact URL, click, value), progress framing, and confirm gates on irreversible actions. Enriches steps with codebase context, offers help when stuck, and captures friction to improve shared runbooks. For steps that capture secrets or write .env / gh values, generates a bundled bash snippet so secrets never pass through chat. Use when user wants guided help completing manual follow-ups after build-code.
- eric-lingren SkillCc Usage Analytics
View and explore Claude Code usage analytics — model mix, estimated cost, request intent, weekly trend, and per-skill tier adherence across the office (cco) and home (cch) profiles. Use when the user wants to see usage stats, run the benchmark, view the weekly report, check if model spend/tiers are trending right, or invokes /cc-usage-analytics.
- eric-lingren SkillImprove Component
Analyze React components or TypeScript/HTML files for reusability, modularity, and hygiene improvements. Separates business logic from presentation, enforces design system usage, and proposes focused refactors. Use when user wants to improve a specific component, clean up a file, or make code more modular.
- eric-lingren BundleClean Scaffolding
Archive consumed doc scaffolding (seeds, PRDs, tasks, handoffs) in the current working tree. Bundles each fully-consumed lineage chain verbatim into docs/archive/. Leaves docs/adr/, docs/prototype/, docs/standups/, and docs/wizard/ untouched. Deletes browser-checks, task trace logs, and docs/visual-changes/<branch-slug>/ directories for branches whose PRs are merged — without archiving. Use when done with a workflow and ready to clear the clutter. Pass --force or -y to skip the confirmation prompt.
- eric-lingren SkillTasks To Linear
Convert a docs/tasks/ JSON file into Linear tickets, preserving blocked-by relationships and writing Linear URLs back to the JSON. Use when the user wants to push tasks to Linear, create Linear issues from a task file, or sync tasks to Linear.
- eric-lingren SkillRegister Skill
Register a new skill or agent into the shared infrastructure so tier-advisor, the weekly usage report, and sync-model-tiers never miss it. Invoke as /register-skill <name>. Covers both the skill registration path (tiered via model-tiers.json skills map) and the agent registration path (tiered via model-tiers.json agents map). This file is the canonical reference for the full registration chain.
- eric-lingren SkillPr Code Review
Comprehensive code review of staged changes or a PR. Checks for bugs, security issues, performance, and adherence to project conventions. Use when user says "review my changes", "review this PR", "code review", or invokes /review.
- eric-lingren SkillDispatch Tasks
Orchestrate a task file by partitioning items by task_type and running one branch per invocation. Routes code items to build-code (inline), reply items to relay (inline), and triage items to export-tasks (agent). Re-entrant — run once per branch until all items are done. Use when user wants to dispatch a mixed task file, route deferred triage items to destinations, or run one pass of the pipeline.
- eric-lingren SkillTo E2e Tasks
Analyze branch changes, discover critical user-facing workflows, grill the test plan, and generate a Playwright e2e tasks JSON file for build-code. Output is a tasks JSON plan, not implemented tests. Use when user wants to plan e2e coverage after implementing a feature.
- eric-lingren SkillSprout Seed
End-to-end unattended pipeline: converts a seed file into tasks, executes every AFK task via build-code, and returns a pr-code-review of the resulting diff — all without user interaction. Use when you want one command to go from seed to reviewed PR.
- eric-lingren SkillInvestigate
Investigates a question, claim, external signal, or incident by routing it to the investigator orchestrator agent (Opus tier), which decomposes it into sub-claims, routes each to the appropriate source, and returns a schema-valid investigation-result with a unified verdict and merged evidence.
- eric-lingren SkillBuild Code
Execute tasks from a tasks JSON file sequentially, one build-runner spawn per task. Handles branching, status updates, blocker detection, leaf/blocker failure policy, and deferred HITL tasks without halting independent work. Use when user wants to run AI tasks from a docs/tasks/ file.
- eric-lingren SkillTldr Tech
Summarize the following technical content across four outputs. Use simple markdown throughout — bullets, bold, short lines. No long prose blocks. Everything must be skimmable at a glance.
- eric-lingren BundlePrototype
Explore ideas, hypotheses, and design options before committing to a PRD. Runs throwaway code (terminal or UI) on a local-only git branch, then collapses findings into docs/prototypes/<filename>.md. Use when the user wants to prototype, sanity-check a data model or state machine, mock up a UI, explore design options, or says "prototype this", "let me play with it", "try a few designs".
- eric-lingren SkillPr Revise
Post-publish PR feedback ingestion skill. Harvests reviewer and bugbot comments from an already-published PR, runs provisional triage, applies mandatory diligence, presents an HITL confirmation gate, fires attribution-tracer for confirmed escapes, and stops at a seed (via /to-seed) whose provenance captures both the code fixes and the PR replies to post. Use when the user says "revise PR", "process PR feedback", "review my PR comments", or invokes /pr-revise.
- eric-lingren BundleFind Work
List active worktrees (and their Linear URLs) powered by gxlist. Falls back to listing spike/*, feat/*, and fix/* branches when gxlist is unavailable. Use when user wants to find previous spikes, feature branches, fix branches, or in-progress research branches.
- eric-lingren SkillWorktree
Resolve or create a git worktree for a branch, list worktrees, or remove a worktree. Delegates entirely to the ~/.dotfiles/.scripts/worktree engine — same logic as the `wt` shell alias. Use when the user types /worktree <branch>, /worktree ls, /worktree rm <branch>, or asks to open/switch to a worktree from inside a Claude session.
- eric-lingren BundleTo Tasks
Break a seed or PRD into independently-grabbable tasks using tracer-bullet vertical slices, then write a JSON task file to docs/tasks/. Use when user wants to convert a seed file or PRD into AI-ready tasks.
- eric-lingren BundleTo Spec
Render a seed .json file into an interactive HTML PRD and save it to docs/prd/. Use when user wants to create a PRD from a seed file with rich interactive output.
- eric-lingren SkillTo Seed
Distill the current conversation or grill session into a standardized JSON IR (seed) and save it to docs/seeds/. Use when user wants to capture decisions and context from a conversation before moving to /to-spec, /to-tasks, or /prototype.
- eric-lingren SkillOffload
Offload a skill invocation to a remote cloud agent. Accepts a target skill name and optional input content; looks up the skill's model and effort from model-tiers.json; spawns an Agent with isolation: 'remote' and the resolved model; instructs the remote agent to run the target skill at the looked-up effort level and return the full output as text; then writes the result to docs/seeds/ (for seeds) or docs/tasks/ (for tasks).
- eric-lingren SkillCapture
Capture an idea, fix or bug as a GitHub issue without breaking flow. Use when the user says "capture idea", "capture fix", "capture bug", "capture that", "file that issue", "document that idea", "capture flush", "capture review", "capture import", or invokes /capture. Optional type word; a bare `capture` asks what the idea is. Builds the issue, shows one compact plain-text preview, and files it in the background on y.
- eric-lingren SkillCI Fix
Diagnose and fix CircleCI pipeline failures on the current branch. Pulls failing job logs via CircleCI MCP, applies all code fixes, verifies locally for non-hook-covered steps, and auto-pushes via gxpush. One shot per invocation.
- eric-lingren SkillRetro
Generate sprint retro talking points (3 things that went well, 3 ideas to improve) from your current-cycle GitHub PRs and Linear tickets. Prints inline for saying out loud in a synchronous retro. Use when the user says /retro, "sprint retro", "retro talking points", or "what went well this sprint".
- eric-lingren SkillRelay
Channel-agnostic egress orchestrator for reply tasks. Reads the task file produced by /pr-revise, determines the target channel for each reply task from reply_url, presents combined drafts for final HITL approval, then routes each approved draft to the appropriate channel adapter under agents/egress/. Use when dispatch-tasks routes the reply branch, or the user invokes /relay <task-file>.
- eric-lingren SkillPick
Rank the next ticket to pick up from the current repo's issue tracker. Reads issue_tracker and pick_buckets from repo-policy.json, shows a "Which bucket?" menu, and prints a top-5 list with copy-paste start lines. Read-only, terminal output only. Use when the user types /pick.
- alecastione-code BundleCaca Bugs
Hunts and fixes bugs on its own. Sweeps a live web app signed in and read-only (every page and tab, console, network, broken values, stuck loaders, slowness), cross-checks the server logs (5xx, 499, slow endpoints, repeated errors), scores health per page, ranks the bugs and fixes each one with a regression test, a PR, the CI gate and a live re-check. Use for "qa", "find bugs", "sweep the app", "what is broken", "caça bugs", "procura bugs", "testa o painel", /caca-bugs, and after every big deploy.
- crownjoker07 SkillEnglish Passage Study
用简单英语逐句讲解用户提供的英文段落,包括句意、句子结构、实用语法和同结构例句;适用于英语阅读与语法学习。
- crownjoker07 BundleInterview Questions
根据用户粘贴或上传的简历生成表格化的精简简历总结(含匹配度与面试建议、证据与待确认项)和固定分类的面试题清单。成果缺少技术机制与量化数据时一律按存疑处理,不计入能力证据。收到简历或面试准备请求时使用。
- crownjoker07 BundleQuery Repository
只读查询远程 Git 仓库最新分支中的源码、配置或文档,并给出 commit、文件和行号证据。用于从指定仓库查询规则、数值、位置或行为;不创建 worktree、不切换分支、不修改代码。
- crownjoker07 SkillClone Repository
将用户指定的 Git 仓库和分支首次克隆到当前 workspace 的 repos/ 目录。用户要求克隆、下载或准备远程仓库时使用;不更新或覆盖已有仓库。
- crownjoker07 BundleProject Docs
Query, explain, create, update, or review authoritative Markdown documentation under this repository's docs/ directory, and perform evidence-grounded product or game-design analysis from project documentation and implementation evidence. Use when answering project questions, tracing documented behavior to code or configuration, maintaining documentation, or reviewing product value, game mechanics, balance, economy, monetization, or retention. Do not use generic design heuristics as evidence of project intent or behavior.
- crownjoker07 BundleRepo Bugfix
诊断和修复 Git 仓库中的 BUG。分析请求只做只读诊断;用户明确要求修复时,在独立 worktree 中完成最小根因修复,经 Review 和测试后提交并推送。
- 4mgln BundleRust Cache Management
Diagnose Rust build-cache growth and plan safe selective cleanup while preserving Cargo configuration and reusable artifacts. Use for Rust disk pressure, cache cleanup, or repeated-build storage concerns.
- ada-lc826 BundleButton Graveyard
Finds buttons, links and clickable elements in a web app that do nothing when clicked (missing handlers, TODO stubs, handlers that throw, controls hidden under invisible overlays). It clicks every one in a real headless browser and records evidence, with no AI judging the results. Then it fixes them and rescans to prove they work. Use when the user wants to check whether their UI actually works, find dead, broken or fake buttons, smoke-test an AI-generated or vibe-coded frontend before a demo, PR or launch, or confirm that button fixes landed.