prime-discipline
Skill2 starsChanged 5 months agoMITView on GitHub
Reference card of discipline patterns from across the research — Boris Cherny's CLAUDE.md / compaction / rewind rules, Karpathy's surgical-changes rule, Lovable's good/bad example format, Cursor's semantic-search-first, Devin's plan↔execute toggle. Single source of habits. Invoke manually via /prime-discipline; the skill-router will also hint it on `prime me`, `load context`, `refresh state`.
In this repository60 other files it gives its agents.
- .claude/CLAUDE.mdCLAUDE.md
- CLAUDE.mdCLAUDE.md
- analyzeSkill
- animation-disciplineSkill
- api-versioningSkill
- autopilotSkill
- browser-e2eSkill
- characterizeSkill
- clarifySkill
- codemodSkill
- constitution-compactSkill
- constitutionSkill
- context-budgetSkill
- creative-webSkill
- db-migrationSkill
- debugSkill
- detect-stackSkill
- dispatch-criteriaSkill
- dreamSkill
- experimentSkill
- fastapiSkill
- flag-cleanupSkill
- flag-rolloutSkill
- flask-realtimeSkill
- framework-detectorSkill
- front-end-designSkill
- grill-meSkill
- handoffSkill
- image-generationSkill
- implementSkill
- initiativeSkill
- instinctSkill
- loopSkill
- nextjsSkill
- observabilitySkill
- okr-alignSkill
- parallel-swarmSkill
- pivotSkill
- planSkill
- pr-bodySkill
- prod-observabilitySkill
- react-nodeSkill
- react-viteSkill
- release-trainSkill
- researchSkill
- reviewSkill
- security-guardSkill
- self-healSkill
- shipSkill
- skill-creatorSkill
- specifySkill
- tailwind-disciplineSkill
- tasksSkill
- tdd-loopSkill
- token-budgetSkill
- typographySkill
- verify-loopSkill
- verifySkill
- wip-checkpointSkill
- <kebab-case-skill-name>Skill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 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.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 34
When defining behavior in a skill or prompt, include BOTH good and bad examples:
- Distinctive: line 56
No preamble ("I'll now read..."), no postamble ("Let me know if...").
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill 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
Prime Discipline
Habits that compound. From the best-engineered setups in the ecosystem.
On corrections (Boris Cherny tips #74, #16)
- Don't say "try X instead." That keeps the failure in context. Use
/rewindor double-tap Esc, then issue a fresh prompt. - After every correction, update CLAUDE.md so the agent doesn't repeat the mistake. Pattern: "Note for future sessions: when you see X, do Y not Z. — added
<date>"
On compaction (Boris tips #75, #76)
- Same task continuation →
/compact "focus on X, drop test debugging"(hint-driven) - New task →
/clear CLAUDE_CODE_AUTO_COMPACT_WINDOW=400000is set in settings.json. Context rot starts at 300-400K on the 1M model.
On exploration (Cursor pattern)
- Always semantic-search before grep.
Globfor file shapes, thenReadthe candidate, onlyGrepwhen you have a specific symbol. - Grep on a 50K-file repo without a path filter is a token sink.
On planning (Devin pattern)
- Plan mode → Execute mode toggle. State the plan in 3-5 bullets. Get explicit user "go" (or implicit in autopilot). Don't interleave planning and execution turn-by-turn.
On examples (Lovable pattern)
- When defining behavior in a skill or prompt, include BOTH good and bad examples:
✅ "Edited src/auth.ts:42-58 — fixed null-check order." ❌ "I've now updated the file. Here's what it contains: <500 lines>" - Show the failure mode you're trying to prevent.
On surgical changes (Karpathy)
- Every changed line traces to the user's request.
- "While I was in there I also fixed..." — undo it. Open a new task for that fix.
- Dead code: mention, never delete unilaterally.
On verification (superpowers)
- "Looks fine" is not verification.
- "Tests pass" alone is not verification of a user-facing change. Boot the app.
- Evidence files in
verify/<date>-<feature>/: at least one screenshot or HTTP response per acceptance criterion.
On token discipline (token-budget skill)
- Cite paths, don't paste files.
- No preamble ("I'll now read..."), no postamble ("Let me know if...").
- Subagents for heavy reads. Their context dies with them.
On autonomy (autopilot skill)
- 3x same error → ABORT, log, move on. No infinite retry.
- WIP commits every 5-15 minutes during long features.
- Hooks > permission prompts in autopilot mode.
Cross-tool consistency
These same rules apply when running in Claude Code, Codex CLI, Cursor agents, etc. The harness deliberately uses AGENTS.md-compatible patterns (CLAUDE.md @imports AGENTS.md is a valid setup) so the discipline ports.
References
- howborisusesclaudecode.com tips #1, #14, #16, #74, #75, #76, #81, #88, #89
- multica-ai/andrej-karpathy-skills/CLAUDE.md (surgical changes)
- x1xhlol/system-prompts-and-models-of-ai-tools (Cursor, Devin, Lovable patterns)
- obra/superpowers (verification + TDD)
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.

