claude-code-config: Skill for Claude Code

/pm-clean is the repo's single janitor. It runs two independent staleness scans and presents one recommend-then-confirm report:

1. Open GitHub issues — solved-by-merged-PR, inactive, superseded, potential duplicates.
2. On-disk workspace — stale worktrees and stale local/remote branches, via stale-cleanup.sh (this skill holds that script's only direct invocation — one implementation, no divergence).

The two scans never gate each other: one finding nothing does not suppress the other, and the summary always reports both. Nothing is closed or deleted without explicit confirmation.

Resolve both scan engines before Step 0:

resolve_script() {
  local name="$1" candidate
  for candidate in \
    "$HOME/.claude/skills-worktree/.claude/scripts/$name" \
    "$HOME/.claude/scripts/$name" \
    ".claude/scripts/$name"; do
    if [[ -x "$candidate" ]]; then echo "$candidate"; return 0; fi
  done
  return 1
}
BACKLOG_STALENESS_SH=$(resolve_script backlog-staleness.sh || true)
STALE_CLEANUP_SH=$(resolve_script stale-cleanup.sh || true)
[[ -n "$BACKLOG_STALENESS_SH" ]] || { echo "ERROR: backlog-staleness.sh not found (checked all three paths) — issue scan unavailable" >&2; exit 1; }
[[ -n "$STALE_CLEANUP_SH" ]] || { echo "ERROR: stale-cleanup.sh not found (checked all three paths) — workspace scan unavailable" >&2; exit 1; }

Invoking-repo scope (issue #687). Both scans stay in the invoking repo's lane: the issue scan (backlog-staleness.sh) runs gh issue list cwd-repo-scoped, and the workspace sweep (stale-cleanup.sh) enumerates worktrees/branches via git -C "$ROOT" where $ROOT is this repo's root — never another project's worktrees. /pm-clean therefore never offers to close an issue or delete a worktree/branch outside the invoking repo, whether run directly or inline from /pm.

Step 0: Parse arguments

Parse $ARGUMENTS:

• [days] — the first bare (non-flag) token. If it is a positive number, use it as the issue-inactivity threshold and the default worktree/branch age. If empty, non-numeric, or non-positive (zero or negative), default to 30 days:
  • Warn if non-numeric: "Invalid argument '{value}', defaulting to 30 days"
  • Warn if non-positive: "Threshold must be positive, defaulting to 30 days"
• --worktree-days N — optional flag. If present and a positive integer, it overrides the worktree/branch age only (the issue scan still uses [days]). If present but invalid (non-numeric or non-positive), warn ("Invalid --worktree-days '{value}', reusing the [days] threshold of {DAYS}") and fall back to [days]. If the flag is absent, the worktree/branch age reuses [days].

Record two values for the rest of the run: DAYS (issue-inactivity threshold) and WORKTREE_DAYS (workspace threshold — equal to DAYS unless --worktree-days overrode it).

Threshold note. Reusing [days] means /pm-clean defaults the workspace age to 30 days (its issue default), whereas stale-cleanup.sh itself defaults to STALE_DAYS=7. That is intentional: this skill owns the threshold it passes, while the script owns the implementation. Use --worktree-days 7 for the script's own aggressive default.