agent-workspace-starter
CLAUDE.md1 starsChanged 36 days agoMITView on GitHub
Always-loaded context that Claude Code reads at session start. Keep it short — every line costs tokens on every turn. Verbal phrases that map directly to a destination — go there without asking. Rule: when a new shortcut emerges, ask once, then add it to this table — so the shortcut works next time. Iron Law: credentials never live in files. Passwords, API keys, tokens, recovery codes all live in your password manager. This file references credentials only by their…
The file
We split this CLAUDE.md into the rules it gives an agent
4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 19
After any user correction, prepend a rule to tasks/lessons.md before continuing other work.
- Very specific: line 24
Write the plan to tasks/todo.md with checkable items.
- Distinctive: line 25
Verify the plan with the user before implementing.
- Distinctive: line 47
If a credential gets pasted into chat (happens occasionally), refuse to save it, explain why, and recommend regeneration — the transcript …
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 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>'s workspace
Always-loaded context that Claude Code reads at session start. Keep it short — every line costs tokens on every turn.
What this is
<!-- Replace with one paragraph describing this workspace. Example: -->Multi-project workspace: personal admin, a side project, and a writing project. Each project folder has its own
CLAUDE.mdandCONTEXT.md.
Working principles
- Simplicity first — prefer editing existing files; no new abstractions without clear need.
- Minimal impact — only touch what the task requires.
- Ask before destructive actions — delete, force-push, mass-rewrite all need confirmation.
- Honest communication — no performative agreement, no empty affirmations. When corrected, state what changes and why.
Self-improvement loop
- After any user correction, prepend a rule to
tasks/lessons.mdbefore continuing other work. Format:## YYYY-MM-DD — Short title, what went wrong, the rule to follow. - At session start, read
tasks/lessons.mdand apply its rules for the duration of the session.
Task management
- Write the plan to
tasks/todo.mdwith checkable items. - Verify the plan with the user before implementing.
- Mark items complete as you go.
- Append a review section on completion (the
wrapskill does this).
Command shortcuts
Verbal phrases that map directly to a destination — go there without asking.
| Phrase | Target |
| "add to tasks" / "add to the list" | tasks/notes.md (append under the right ## section) |
| "add a lesson" / "lesson learned: ..." | tasks/lessons.md (prepend entry) |
| "orient" / "get me up to speed" | run the orient skill |
| "tasks" / "what's on the list" | run the tasks skill |
| "wrap" / "close out" | run the wrap skill |
Rule: when a new shortcut emerges, ask once, then add it to this table — so the shortcut works next time.
Credentials
Iron Law: credentials never live in files. Passwords, API keys, tokens, recovery codes all live in your password manager. This file references credentials only by their password-manager item name, never their value.
If a credential gets pasted into chat (happens occasionally), refuse to save it, explain why, and recommend regeneration — the transcript has already leaked it.
Protected paths
The PreToolUse hooks in .claude/settings.json block agent writes to anything matching .claude/protected-paths.txt. If a legitimate edit is blocked, loosen the list deliberately — don't bypass the hook.
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.

