Stale Brain
Audit AI agent memory and instruction files: CLAUDE.md, AGENTS.md, .cursorrules, GEMINI.md, copilot-instructions.md and every other major agent's memory files (13 agents, 18+ file locations) against the live repository. Use when the user runs /stale-brain or asks to audit, verify, refresh, lint, or fix agent memory, rules, or instruction files; when they complain the agent "ignores rules", "keeps getting it wrong", or that instructions "feel outdated"; and mid-task, whenever an instruction in a loaded memory file contradicts observed reality (a referenced path is missing, a named script doesn'
Install
npx skills add stalebrainlabs/stalebrain --skill stale-braingit clone --depth 1 https://github.com/stalebrainlabs/stalebrain.git /tmp/stalebrain
mkdir -p ~/.claude/skills
cp -r /tmp/stalebrain/. ~/.claude/skills/stale-brainPersonal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.
From the SKILL.md
# stale-brain: provenance and decay for agent memory Agent memory files are written once and trusted forever. Code moves on; the memory rots; the agent gets confidently wrong. stale-brain is not a linter. It is a **trust model**: after an audit, every claim in agent memory carries an **age**, a **verification date**, and **cited evidence**. Four laws govern everything below: 1. **Every claim carries provenance.** Confirmed claims get a dated stamp; nothing is "just true". 2. **Confidence decays.** Every claim type has a half-life. Past its half-life, a fact becomes a hypothesis until re-verified. 3. **Contradictions cite evidence.** A CONTRADICTED verdict names the files and commits that prove it, never "this looks wrong". 4. **The human approves; the human does zero detective work.** You gather all evidence and prepare ready-to-apply diffs, but you never rewrite memory without explicit approval. ## Capabilities you need (any model, any harness) The protocol uses four generic capabilities. Map them to whatever you have: | Capability | Claude Code | Cursor / Copilot / Windsurf / Cline | No tools at all | |---|---|---|---| | Find files by pattern | `Glob` | workspace file search | ask the user to paste `git ls-files` output | | Search file contents | `Grep` | codebase/regex search | ask the user to paste the file | | Run git / read shell output | `Bash` / `PowerShell` | integrated terminal | ask the user to run the command and paste output |
Details
- Skill name
stale-brain- Repository
- stalebrainlabs/stalebrain
- License
- MIT
- Repo updated
- 31 July 2026
FAQ
What does the Stale Brain skill do?
Audit AI agent memory and instruction files: CLAUDE.md, AGENTS.md, .cursorrules, GEMINI.md, copilot-instructions.md and every other major agent's memory files (13 agents, 18+ file locations) against the live repository. Use when the user runs /stale-brain or asks to audit, verify, refresh, lint, or fix agent memory, rules, or instruction files; when they complain the agent "ignores rules", "keeps getting it wrong", or that instructions "feel outdated"; and mid-task, whenever an instruction in a loaded memory file contradicts observed reality (a referenced path is missing, a named script doesn'
How do I install the Stale Brain skill?
skills CLI (Claude Code, Cursor, Codex and more): npx skills add stalebrainlabs/stalebrain --skill stale-brain. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.
Is Stale Brain popular?
It lives in stalebrainlabs/stalebrain, which has 5 GitHub stars and 0 forks.
Related skills
Stele
5johnwangwyx/stele
Durable project and task state kept as plain markdown in the repository - a PROJECT_CONTEXT.md of standing facts, a generated TASKS.md census, and one file per task holding its goal, current state, failed attempts and steps. Every step is written before the work it describes rather than at session exit, so a session killed by a rate limit or a crash still leaves the next agent — in any harness — enough to carry on. Invoke only when stele is named — the user asks to manage(init) a project with it, asks to resume or continue where it left off, or a project instruction file points at a stele/ dir
Skill Guide
5VV1NN/skill-guide
Skill navigator and guide -- automatically scans all installed skills and slash commands, then explains them in the user's native language with usage examples, workflow order, and recommendations. Triggered when users ask about available skills, how to use a skill, what skills are installed, or need help choosing the right skill for a task. Also triggered by questions like 'what can you do', 'what tools do I have', 'how do I start', 'help me understand this skill'. Multilingual: responds in whatever language the user writes in.
Taoci Letter
5jielosc/taoci-letter
为保研、考研、申博、读研或暑研申请生成个性化的导师联系邮件。用户要给导师/教授写套磁信、联系信或申请邮件,或提到“套磁/套磁信/陶瓷信/contact a professor/cold email professor”时使用。
Sibyl Save
5Sibyl-Labs/sibyl-save
Session memory sync and skill extraction for the Sibyl Memory Plugin. Injects session context into the plugin's local memory DB (entities, state, journal) through the plugin interface, then extracts reusable patterns into harness-native skills. Use at the end of a session or after a major block of work. Triggers - /sibyl-save, save session, sync memory, /sibyl-save setup, /sibyl-save preview.
Trip Planner Cn
5ChenYiming-aaa/better-trip-planner
境内旅游计划全流程制作(仅支持中国大陆境内行程 / domestic-China-only build): 把"我想去某地玩 N 天"变成经过核验、逐项可预订的行程——城市路线骨架,高铁/机票/自驾 交通决策(12306 与携程/Trip.com/去哪儿深链),小时级每日时间线(开放时间、停留时长、 假日调休冲突、高德/百度逐跳导航链接+离线KML),按街区筛选的住宿,预算汇总,带深链的 预订清单,以及把成品渲染成设计页面(七种主题:illustrated / clay / noir / glass / journal / zine / splash — 插画/黏土/夜航/玻璃/手账/Zine/闪屏,全部纯图文、无视频)。 Use this whenever the user asks to plan a domestic trip (境内游/周边游/自驾游/高铁游), 排一个旅行日的小时级时间表,填一个空闲时段("我在 X 附近有 2 小时空档"),or turn a finished plan into a designed page, or asks 旅行规划/行程安排/机票火车票比价/去某地玩 N 天怎么安排/把行程做成好看的网页 — even if they only mention one piece, the playbook and verification rul
Parallel Lanes
5noderaven/parallel-lanes
Runs an approved implementation plan as parallel lanes of tasks in separate git worktrees, each task implemented and reviewed by its own agents, then integrated, checked and reviewed as a whole. Use when an approved implementation plan is about to be executed (at the execution-method handoff), or when resuming a stopped parallel-lanes run.
Browse the most popular Claude skills, all productivity & notes skills, or write your own with the skills generator.