← ClaudeAtlas

plan-orchestratorlisted

Turns a bullet list of tasks/issues/bugs into one ordered plan with full coverage verification. Fans out specialist sub-agents in parallel. Designed for Claude Code Plan Mode. Read-only — produces a plan, never edits.
anthril/official-claude-plugins · ★ 3 · AI & Automation · score 82
Install: claude install-skill anthril/official-claude-plugins
<!-- Runtime dependencies: bash, python3. The bare `Bash` entry in `allowed-tools` is required to invoke the helper scripts under `scripts/` (parse-bullets, classify-tasks, verify-coverage, compile-plan, detect-stack, stop-hook). Bash is never used to mutate project source — the read-only guarantee is enforced by the "no Write, no Edit" principle in the body. --> # Plan Orchestrator ultrathink ## Before You Start > **CRITICAL — Execute immediately, do not meta-plan.** This skill IS the plan generator. When invoked (whether or not Plan Mode is already active), do NOT run the harness's default Plan-Mode workflow (Phase 1 Explore subagents → Phase 2 Plan subagents → Phase 4 write plan file → Phase 5 ExitPlanMode). Skip all of that. Go directly to *this skill's* Phase 1 (Input Validation & Task Inventory) on your very first tool call. The orchestrator dispatches its own `*-investigator` sub-agents in Phase 3 — never spawn top-level `Explore` or `Plan` agents to "research the request first". The compiled plan emitted at the end of Phase 5 IS the deliverable; it becomes the argument to `ExitPlanMode` directly. 1. **Enter Plan Mode if not already active.** Call `EnterPlanMode` as your first action so the read-only guarantee is enforced by the harness, not just by convention. If the session is already in Plan Mode, the call is a safe no-op — proceed regardless. Do not pause for user confirmation here; the user invoked the slash command knowing this skill produces a plan. 2. **L