Paul Graham, distilled — 5 Claude Code skills + 6 principles from his essays
Skill Claude CodeCodex
Apply when an attempt at a task has failed, when Claude is debugging, when a previous answer did not work, when iterating on a solution that is not converging, or when the user reports that the last response missed the mark. Critical trigger: any moment where the temptation is to retry the same approach with cosmetic…
Skill Claude CodeCodex
Apply when Claude is in any mode of investigation, analysis, critique, or disagreement. Triggers especially when: the user pushes back on a previous Claude position, Claude finds a conclusion that contradicts what the user seems to want, Claude is reviewing the user's work, or Claude is being asked to evaluate…
Skill Claude CodeCodex
Apply at the start of any non-trivial task, and any time something looks 'clean,' 'simple,' or 'obvious' during execution. Triggers include: receiving a user request before executing it, encountering a clean dataset, seeing an elegant solution that resolves quickly, finding an obvious explanation that fits the surface…
Skill Claude CodeCodex
Apply whenever Claude is making any claim, recommendation, factual statement, or analysis. Always-on for any non-trivial assertion. Triggers especially when: Claude is producing content in domains where its training data is uneven, when the user is asking for facts Claude is not certain of, when the analysis depends…
Skill Claude CodeCodex
Apply when Claude is producing any substantive output: analysis, recommendations, reports, messages, code reviews, or written responses of any non-trivial length. Always-on for output quality. Specifically triggers when Claude is tempted to use elevated vocabulary, default to bulleted lists, add transitional…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: