claude-academic-setup
CLAUDE.md48 starsChanged 7 days agoMITView on GitHub
Precedence: system constraints > explicit user request > project CLAUDE.md > this file. If a named skill or tool is unavailable, use the closest available workflow. When starting a new task, check for clearly relevant skills and invoke them first. Key triggers: - Session start → napkin - New data → datacheck - Plots → viz - Prose → plain-writing. Default for docs and general-audience prose. For expert or academic writing, keep field terms and stay concise. - Creative work…
In this repository9 other files it gives its agents.
- AGENTS.mdAGENTS.md
- agySkill
- claudeSkill
- codexSkill
- datacheckSkill
- hallucinatorSkill
- napkinSkill
- plain-writingSkill
- vizSkill
The file
Show its rules and text
We split this CLAUDE.md into the rules it gives an agent
14 rules. About 0 are common, 11 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈11
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
- Distinctive: line 4
If a named skill or tool is unavailable, use the closest available workflow.
- Distinctive: line 8
When starting a new task, check for clearly relevant skills and invoke them first.
- Very specific: line 15
Use AskUserQuestion (via ToolSearch) for structured questions.
- Distinctive: line 24
No features, abstractions, or error handling beyond what's needed.
- Distinctive: line 25
Prefer concise solutions. If it looks overcomplicated, simplify.
- Distinctive: line 29
Don't improve, refactor, or reformat code you weren't asked to change.
- Distinctive: line 31
Remove only orphans YOUR changes created.
- Distinctive: line 36
When running substantive analysis (not one-liners or sanity checks):
- Very specific: line 38
Keep logs/ in .gitignore.
- Distinctive: line 39
Never report results without evidence the code actually ran (log, notebook output, or test report).
- Distinctive: line 44
Do not discard work you did not create.
- Very specific: line 45
Never include Co-Authored-By in commit messages.
- Distinctive: line 49
Never hard-wrap prose. Write each paragraph as one continuous line and let the editor soft-wrap.
- Distinctive: line 52
When editing an existing prose file that is hard-wrapped and the user asks for content changes, don't silently reflow it.
≈ “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
CLAUDE.md
Precedence: system constraints > explicit user request > project CLAUDE.md > this file. If a named skill or tool is unavailable, use the closest available workflow.
1. Use Skills When Relevant
When starting a new task, check for clearly relevant skills and invoke them first. Key triggers:
- Session start →
napkin - New data →
datacheck - Plots →
viz - Prose →
plain-writing. Default for docs and general-audience prose. For expert or academic writing, keep field terms and stay concise. - Creative work →
brainstorming
Use AskUserQuestion (via ToolSearch) for structured questions. Fall back to plain text if unavailable.
2. Think Before Coding
- State assumptions. If uncertain, ask.
- Multiple interpretations → present them, don't pick silently.
3. Simplicity First
- No features, abstractions, or error handling beyond what's needed.
- Prefer concise solutions. If it looks overcomplicated, simplify.
4. Surgical Changes
- Don't improve, refactor, or reformat code you weren't asked to change.
- Match existing style. Assume the worktree may be dirty.
- Remove only orphans YOUR changes created.
- Exception: visual/UI tasks should present cohesive updates, not fragmented line-edits.
5. Reproducible Evidence
When running substantive analysis (not one-liners or sanity checks):
- Capture output via
teetologs/YYYY-MM-DD_HH-MM-SS_<description>.log - Keep
logs/in.gitignore. - Never report results without evidence the code actually ran (log, notebook output, or test report).
6. Git
- Inspect status and diffs before changing or committing files.
- Do not discard work you did not create.
- Never include
Co-Authored-Byin commit messages.
7. Prose & Markdown Line Breaks
- Never hard-wrap prose. Write each paragraph as one continuous line and let the editor soft-wrap.
- Insert newlines only between block elements (paragraphs, list items, headings, table rows, code blocks) — never mid-sentence or at a fixed column width.
- Applies to all
.md,.txt, and general prose output. Exceptions: git commit message bodies (wrap at ~72), and files whose existing content is already hard-wrapped (match the file). - When editing an existing prose file that is hard-wrapped and the user asks for content changes, don't silently reflow it; ask first.
8. Independent Review
- For important plans or results, get a review from a different model with
/codexor/agy. - Present the findings and wait for approval before changing files.
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.

