doctorlisted

# Doctor This skill audits the mechanical layer of the memory engine: are the hooks firing, have the scripts actually run, are the logs staying fresh, is the vault clean. The goal is to make a silent failure visible. ## How to run 1. Work from the vault root (the folder containing CLAUDE.md/AGENTS.md). All paths below are relative, never write an absolute path. 2. Run the checks below **with Bash, in order**. Use the commands as given, do not guess at them. 3. Classify each check's output as green / yellow / red. 4. Report the result as one table, with one fix line per red row. 5. Close with a single-sentence verdict. Checks are read-only. Never fix anything on your own, report first, then fix only if the user asks. ## Resolving Python Several checks below need a working Python interpreter, and this skill exists partly to detect the case where `python3` on PATH does not work: on Windows, `python3` is a Microsoft Store stub that resolves on PATH and then fails when actually run. A check that shells out to a bare `python3` cannot run on the very machine where that failure happens. Every check below that needs Python resolves one first, with this line, in this order: the interpreter recorded in `.claude/settings.local.json`'s `SessionStart` hook (the same value check 4 tests), then `python3`, then `python`, taking the first one that actually runs: ```bash PY=$(grep -o '"command":[[:space:]]*"[^"]*"' .claude/settings.local.json 2>/dev/null | head -1 | sed -E 's/.*"([^"]