randlee/sc-startup

sc-startup Skill

Thin orchestration for the /sc-startup command. Validates config, launches background agents, aggregates statuses, and emits a concise startup report. Delegates all heavy lifting to agents via Agent Runner (registry enforced, audited).

Command

• /sc-startup [--pr] [--pull] [--fast] [--readonly]
• /sc-startup --init (config discovery and synthesized YAML preview)

Agents

• ci-pr-agent (PR list/fix; list-only when --readonly)
• sc-worktree-scan / sc-worktree-cleanup (worktree hygiene; scan-only when --readonly)
• ci-automation (pull-only master → develop; must complete before checklist updates)
• sc-checklist-status (report/update checklist; report-only when --readonly; no auto-commit)
• sc-startup-init (detection-only: config presence, candidates, package detection; returns fenced JSON with YAML payload)

Flow (best-effort)

1. If --init: Agent Runner → sc-startup-init (detection-only). Report detected prompt/checklist candidates, installed packages, missing keys, and synthesized YAML. Do not block on AskUserQuestion; exit after reporting findings.
2. Load .claude/sc-startup.yaml; validate required keys and enabled feature dependencies. If config is missing or invalid, return concise guidance with detected candidates and recommend --init or manual file creation. Fail closed with DEPENDENCY.MISSING if enabled package absent.
3. If --fast: read startup prompt only, summarize role, exit (no agents/checklist).
4. If --pr and enabled: Agent Runner → ci-pr-agent (--list --fix, or list-only when --readonly).
5. If worktree scan/cleanup enabled in config: Agent Runner → sc-worktree-{scan|cleanup} (scan/report-only when --readonly).
6. If --pull: Agent Runner → ci-automation (pull-only master → develop); wait for completion before checklist updates.
7. Agent Runner → sc-checklist-status (default update; report-only when --readonly). Checklist changes stay in workspace (no commit).
8. Read startup prompt + checklist (post-update). Aggregate task statuses in deterministic order; never abort on agent errors/timeouts.
9. Emit concise report: prompt summary, checklist deltas, PR/worktree/CI outcomes, partial failures, and next steps.

If sc-checklist-status reports VALIDATION.INVALID_PATH (path escape), explicitly flag it as a warning/issue in the final report (include the word "warning" or "issue").

Config (.claude/sc-startup.yaml)

• startup-prompt (string, required)
• check-list (string, required)
• worktree-scan (string: scan|cleanup|none|"")
• pr-enabled (bool, optional; must be false if PR package absent)
• worktree-enabled (bool, optional; must be false if worktree package absent)

Safety

• Path safety: repo-root-relative only.
• Default mutating; --readonly forces report-only everywhere.
• No auto-commit of checklist changes.
• Dependency validation required before launch; fail closed if missing.
• Missing or invalid config must not trigger an interactive question loop.
• Logging: Agent Runner audit logs under .claude/state/logs/sc-startup/; prune after ~14 days.

Output Contract

• Aggregate per-task status objects {agent_id, status: success|failure|timeout|partial, results, error?} plus narrative summary.
• Agents return fenced JSON minimal envelope; treat malformed/unfenced JSON as failure status.
• Dependency failure format (example):

  {
    "success": false,
    "data": null,
    "error": {
      "code": "DEPENDENCY.MISSING",
      "message": "Feature 'pr' enabled but required agent 'ci-pr-agent' is not installed",
      "recoverable": true,
      "suggested_action": "Install ci-pr-agent or set pr-enabled: false in .claude/sc-startup.yaml"
    }
  }