skilldoctorlisted

# skilldoctor A read-only, deterministic health check over every skill and slash-command Claude Code loads. It answers the one question the agent cannot answer about itself: *is this skill even visible to me?* ## Run it ```bash uvx claude-skills-doctor # full check; exits non-zero if there are errors uvx claude-skills-doctor --fix # same, plus a concrete fix under each finding uvx claude-skills-doctor budget # the budget bar + the biggest descriptions to trim uvx claude-skills-doctor --json # machine-readable, for parsing ``` State check first: if `uvx` is missing, use `pip install claude-skills-doctor` and run `skilldoctor` instead. The tool only reads files — it never edits a skill, and makes no network or model calls, so it is always safe to run. Useful flags: `--project PATH` checks a repo's `.claude/skills`; `--skills PATH` checks a `skills/` directory you are authoring (a plugin repo); `--strict` fails on warnings too. ## The failure this exists for Claude Code injects every skill and slash-command description into the system prompt under one shared character budget (15,000 by default, `SLASH_COMMAND_TOOL_CHAR_BUDGET`). Past it, some skills are **silently** left out of the prompt — no error, no warning — and Claude is told not to use skills it wasn't shown. So the answer to "why doesn't my skill trigger?" is often "Claude was never told it exists." Check the budget line first. If it is over, that is the cause, and no a