This file loads only when the cwd is inside this repo (~/Code/claude-tools). It does not load when the skills are invoked from elsewhere via the ~/.claude/skills/* symlinks.
When you modify a skill in skills/<skill-name>/ (SKILL.md or its scripts), also update:
README.md(repo root) — the one-line skill description should reflect any new flags, behaviors, or caveats that a user would want to know before installing.skills/<skill-name>/how-i-prompted-this.md— append a new "Round N" section describing what the user asked for and what shipped. This file is the skill's dev-diary — a narrative record of how it evolved via prompting, not a changelog. Quote the user's actual prompt verbatim when it's instructive. Outside of quotes, these bullet points should be 1-3 short sentences with a focus on the abstract process. For example: "Gave Claude a doc URL and asked it to list every way it could try to access it and the results. Used those results to add warnings around non-functional approaches to getting doc URLs." is a good single line. Listing every single item tried and the results is not.
Don't batch doc updates across multiple rounds — land them with the code change they describe.
Applies only to files under this repo. Don't write a how-i-prompted-this.md or touch READMEs in the user's other projects unless explicitly asked.