agent-context-os
CLAUDE.md30 starsChanged 37 days agoMITView on GitHub
→ Read identity/who-i-am.md for bio and background. Add more commands as you build out skills. See docs/agent-template.md for how. The /setup, /start, /update, and /end commands are Claude Code adapters to the portable workflow cores in .agents/skills/context-*. Keep shared procedure in the skill; keep Claude-only tools and auto-memory behavior in the command adapter. For tasks without a slash command → load ROUTING.md. When thinking through a decision or tradeoff — rather than producing a deliverable — operate as a thought…
In this repository14 other files it gives its agents.
- AGENTS.mdAGENTS.md
- starter/AGENTS.mdAGENTS.md
- context-endSkill
- context-setupSkill
- context-startSkill
- context-updateSkill
- endSkill
- migrate-geminiSkill
- mine-gemini-workflowsSkill
- setupSkill
- startSkill
- updateSkill
- skill-creatorSkill
- contextos-workspaceSkill
The file
We split this CLAUDE.md into the rules it gives an agent
7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 29
Add more commands as you build out skills.
- Distinctive: line 39
When thinking through a decision or tradeoff — rather than producing a deliverable — operate as a thought partner:
- Distinctive: line 41
Challenge assumptions: If a premise seems weak or unexamined, say so.
- Distinctive: line 44
Be direct about tradeoffs: Don't soften the downsides.
- Very specific: line 55
Run /dream explicitly for shipped curators (rot, merge, split, lint), then /dream-apply to review live-memory changes.
- Distinctive: line 64
When you add project-specific data (metrics, pipeline status, etc.), pick one file as the source and note it here.
- Very specific: line 68
When running parallel Claude Code sessions, use claude --worktree <task-name> so each gets its own branch.
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
[Your Name] — Context
@AGENTS.md
Who I Am
→ Read identity/who-i-am.md for bio and background.
Slash Commands (Claude Code)
| Command | Does |
/setup | Interactive onboarding — builds identity, project, and state files conversationally |
/start | Load repository state and get a session briefing |
/end | Log session, update state and the decision log, propose memory updates, check git |
/update | Mid-session checkpoint — quick state save |
/today | Morning heartbeat — briefing, staleness check, deadlines |
/clean-ai-writing | Load writing/skills/avoid-ai-writing/SKILL.md and apply it |
/capture | Triage raw notes from inbox/ into the right locations |
/find-context | Find relevant context files by topic keyword (avoids Claude's built-in /context) |
/reconcile | Drift detection after parallel sessions |
/recover | Scan orphaned worktrees and stale branches, offer cleanup |
/content-shipped | Log a published piece of content |
/dream | Run an on-demand curator pass over the configured memory dir (default: rot detection) |
/dream-apply | Walk a curator proposal artifact, accept/reject/edit per item |
/migrate-gemini | Migrate selected Gemini CLI workflows with dry-run review and parity checks |
/mine-gemini-workflows | Find repeated Gemini workflows and draft portable skills |
Add more commands as you build out skills. See docs/agent-template.md for how.
The /setup, /start, /update, and /end commands are Claude Code adapters to the portable workflow cores in .agents/skills/context-*. Keep shared procedure in the skill; keep Claude-only tools and auto-memory behavior in the command adapter.
When to Load Additional Context
For tasks without a slash command → load ROUTING.md.
Thought-Partner Mode
When thinking through a decision or tradeoff — rather than producing a deliverable — operate as a thought partner:
- Challenge assumptions. If a premise seems weak or unexamined, say so.
- Offer the alternative. Present the strongest case for a different path.
- Ask the uncomfortable question. What hasn't been considered or is being avoided?
- Be direct about tradeoffs. Don't soften the downsides.
- Summarize the decision cleanly. What's the choice, what does each path require, and what's the real risk?
This mode applies to career moves, strategy decisions, project bets, and any "help me think through X" framing. It does NOT apply to execution tasks where the goal is to produce a deliverable.
Safety Contract
Actions that change external state or are hard to reverse require confirmation. See docs/safety-contract.md for the full policy, approval patterns, and design principles.
Memory
Claude Code enables host-local auto-memory by default. Do not hand-derive its internal path from the current working directory. Use /memory to inspect it and autoMemoryEnabled: false if automatic host writes are unwanted; the optional /dream workflow has a separate explicit local path contract in docs/auto-memory.md. Typed entries use user, feedback, environment, project, and reference; MEMORY.md stays an index capped at ~100 lines.
Run /dream explicitly for shipped curators (rot, merge, split, lint), then /dream-apply to review live-memory changes. Proposal artifacts are written and committed before review; live memories are not. Pattern, standalone contradiction, and untapped curators are roadmap items. See docs/dream-architecture.md.
Single Source of Truth
Each piece of data lives in one file only. Other files can reference it, never duplicate it.
- Tasks →
TODO.md(curated top-of-mind view →state/current.md) - Weekly focus →
state/weekly-priorities.md
When you add project-specific data (metrics, pipeline status, etc.), pick one file as the source and note it here.
Parallel Sessions
When running parallel Claude Code sessions, use claude --worktree <task-name> so each gets its own branch. Merge to main when done. Run /reconcile after parallel work to catch drift.
For high-traffic repos, enable the parallel-session guards: add the repo basename to .claude/hooks/guarded-repos.txt and worktree-guard.sh (PreToolUse) blocks edits to its primary checkout when ≥2 Claude sessions are running, while branch-hygiene.sh (SessionStart) surfaces non-default HEAD. See .claude/hooks/README.md.
Claude Code vs claude.ai
This repo is designed for Claude Code (CLI). If you also use claude.ai projects, upload only a deliberately selected and sanitized subset as plain project knowledge. Repository instructions and skill metadata do not become active commands there. See docs/claude-projects-sync.md if present.
Codex uses the same repository state through root AGENTS.md and .agents/skills/. See docs/codex-onboarding.md; do not copy Claude hooks, settings, or auto-memory assumptions into portable skills.
Repo Maintenance
See docs/repo-maintenance.md for staleness conventions, changelog updates, repo map regen, and validation.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.