writing-skillslisted

# Writing Skills **Read Anthropic's [skill authoring best practices](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices) for the format itself.** It is the authority on frontmatter validation, progressive disclosure, naming and structure, and it is maintained. This skill deliberately does not restate it — a vendored copy used to live here and had drifted out of date within three weeks. Everything below is what that document does *not* cover: findings from wording tests run in this repo, and conventions enforced by this repo's tests. The checkable numbers worth remembering: SKILL.md body **under 500 lines**, description **under 1,024 characters**, references **one level deep** from SKILL.md, reference files over 100 lines get a table of contents. ## Do not author guidance until a control exhibits the failure **Always run a no-guidance control first. If the control doesn't exhibit the failure, there is nothing to fix — stop, and don't author the guidance.** This is the most load-bearing rule here, and it deletes more text than it writes. Two examples from this repo, both of which removed a rule rather than adding one: - `test-driven-development` carried "You MUST write the test first" as an absolute. A no-guidance Opus 5 control, given valuable token-bucket logic three hours in, a sprint ending and a blocked colleague, declined to delete the work and instead invented a mutation-testing recovery, disclosed the violation unprompted, and named t