SuperAPI AI Directory

superskills-doctor

Deployment & operations

View repository

Updated 10-15 17:18

About this skill

/superskills-doctor

Read-only health check. Report what is, never repair it.

Rules

  1. Never mutate. Do not run ./setup, git pull, git reset, curl … | bash, bun install, or any login. If the user wants the fix, name the command from the table's "next action" column or point at /superskills-upgrade.
  2. Never claim readiness the script did not. The verdict is the script's verdict. "blocked" or "unverified" means not ready, even if most rows are fine.
  3. Report evidence, not impressions. Reproduce the table; do not summarise rows away or add checks you did not run.

Steps

Step 1: Locate and run the doctor (one command)

The script lives in the superskills repo at scripts/doctor.sh. Find the repo, in order: the directory this skill was loaded from (if your host told you this skill's base directory, put it in SKILL_DIR on the first line), the Claude Code plugin root, then this skill's ./setup symlink for Claude Code, Codex or OpenCode — never the working directory. Locate and run it in a single Bash call: separate calls do not share shell variables, so a path found in one call is gone in the next.

SKILL_DIR="${SKILL_DIR:-}"   # set to this skill's base directory when the host provides it
case "$SKILL_DIR" in /*) : ;; *) SKILL_DIR="" ;; esac   # only an absolute path is trusted, never resolved against the cwd
ROOT=""
for cand in "${SKILL_DIR:+$SKILL_DIR/../..}" "${CLAUDE_PLUGIN_ROOT}"; do
  [ -n "$cand" ] && [ -f "$cand/scripts/doctor.sh" ] && { ROOT="$(cd "$cand" && pwd -P)"; break; }
done
if [ -z "$ROOT" ]; then
  for probe in "$HOME/.claude/skills/superskills-doctor/SKILL.md" \
               "$HOME/.codex/skills/superskills-doctor/SKILL.md" \
               "$HOME/.config/opencode/skills/superskills-doctor/SKILL.md"; do
    # Only a link that resolves to a real file counts; never fall back to the cwd.
    [ -L "$probe" ] || continue
    real="$(readlink -f "$probe" 2>/dev/null || perl -MCwd=realpath -e 'print realpath($ARGV[0])' "$probe")"
    if [ -n "$real" ] && [ -f "$real" ]; then
      ROOT="$(cd "$(dirname "$real")/../.." 2>/dev/null && pwd -P)"
      [ -f "$ROOT/scripts/doctor.sh" ] && break
      ROOT=""
    fi
  done
fi
if [ -n "$ROOT" ] && [ -f "$ROOT/scripts/doctor.sh" ]; then
  echo "DOCTOR=$ROOT/scripts/doctor.sh"
  bash "$ROOT/scripts/doctor.sh" "$@"
else
  echo "DOCTOR=missing"
fi

If it prints DOCTOR=missing, that is itself the finding: report Repo: blocked with the reinstall command below and stop. Do not improvise checks.

git clone https://github.com/ariadoss/superskills.git ~/.claude/skills/superskills && cd ~/.claude/skills/superskills && ./setup

To inspect an install that is not the current user's (a container, a test tree, a colleague's copy), run the same block with the flags appended to the bash "$ROOT/scripts/doctor.sh" line: --root <superskills-checkout> --home <home-dir>. To call the script directly later, use the path from the DOCTOR= line.

Step 2: Read the output

The script prints a markdown table (Check | Status | Evidence / next action) and a final Verdict: line. Statuses:

Status Meaning
ready passed
warning usable; something optional is off
blocked a required piece is missing or broken — the row says what to run
unverified the probe could not run (e.g. no claude CLI) — unknown, not fine

Step 3: Report

Return the table verbatim, then the verdict, then one line per non-ready row naming the next action from that row. Say plainly:

  • Verdict: ready → "superskills is ready."
  • ready with warnings → ready, and list the warnings.
  • blocked → "Not ready" and the blocked rows first. The usual fix is ./setup from the repo, or /superskills-upgrade; offer it, do not run it.
  • unverified → say readiness is unknown and which probe could not run.