Launchpad

Launchpad

End-to-end orchestrator that chains SPARC or wave-planner (Stage 1), plan-review (Stage 2), linear (Stage 3), and hive-mind-execution (Stage 4) into a single interruptible workflow.

Stage 1 routing (ADR-109):

• Infra changes (Docker, CI, entrypoints, hooks, dev tooling) → SPARC researcher + architect → docs/internal/implementation/{slug}.md
• Feature work (application code, new endpoints, UI) → wave-planner → hive-mind YAML configs

Execution

When triggered, immediately:

1. Read ~/.claude/skills/launchpad/agent-prompt.md
2. Spawn a single Task with subagent_type: "general-purpose" passing the agent-prompt content as the prompt
3. Include in the prompt: the user's request (issue IDs, project name, or description), current working directory, and any flags (--from, --skip-review, --fresh, --infra, --feature)
4. Wait for the agent to complete
5. Present the agent's summary to the user

Do NOT execute the workflow in this session. The subagent handles all stage sequencing, gate prompts, resume detection, and skill dispatching.

Execution Context Requirements

This skill spawns a general-purpose subagent that coordinates stages via nested Task dispatches.

Foreground execution required: Yes. Interactive stage gates use AskUserQuestion which auto-denies in background mode.

Dispatcher tools (frontmatter): Read, Task, AskUserQuestion Subagent tools: Read, Write, Edit, Bash, Grep, Glob, Task, AskUserQuestion, TodoWrite

Reference: Subagent Tool Permissions

Sub-Documentation

Related Skills

Infra Trigger Criteria (ADR-109)

Stage 1a (SPARC) is selected when scope includes any of:

• docker-entrypoint.sh, Dockerfile, docker-compose.yml
• .github/workflows/*.yml
• .husky/*, pre-commit, pre-push hook scripts
• scripts/ files called by Docker/CI/hooks
• package.json scripts block changes
• Dev tooling config: .eslintrc, vitest.config.ts, turbo.json

Stage 1b (wave-planner) for everything else. Mixed scope → use Stage 1a.

Override with flags: --infra forces Stage 1a, --feature forces Stage 1b.

Changelog

v1.3.1 (2026-04-02)

• Replaced 3 hardcoded skillsmith-dev-1 references with {CONTAINER_NAME} placeholder (SMI-3787)
• Added Container Name Resolution section: docker compose ps primary, CLAUDE.md grep secondary, user prompt fallback
• Error guard: unresolved {CONTAINER_NAME} produces explicit error, never silent pass-through

v1.3.0 (2026-04-02)

• Stage 3: Changed subagent_type from Linear-specialist to general-purpose with explicit Linear skill prompt (no registered specialist type exists)
• Stage 3: Team UUID resolved from {TEAM_UUID} variable (read from project CLAUDE.md) instead of hardcoded value
• Stage 3: MCP tool names now use fully-qualified mcp__claude_ai_Linear__* form in subagent prompts
• Stage 3: Added rate-limit handling guidance (exponential backoff, batch limits) to issue creation prompt
• Stage 3: Added multi-issue verification step — checks min(3, TOTAL_ISSUES) issues (first, middle, last) before gate prompt (SMI-3776)
• Stage 3: Verification fallback uses query.ts (read-only GraphQL) instead of status command (which is mutating)
• Stage 3: Parse-failure pre-check — if WAVE_ISSUES is empty/malformed, skips ID lookup and goes to failure prompt
• Stage 3: Replaced "Continue anyway" option with "List recent project issues (verify manually)"
• Stage 3: All script paths use $HOME/... absolute paths instead of relative

v1.2.0 (2026-02-19)

• Section 2: Add MISSING/ENCRYPTED/OK skill classification with xxd-based binary detection (SMI-2676)
• Section 7: Replace silent Stage 4 fallback with explicit AskUserQuestion consent gate; add ENCRYPTED case with git-crypt unlock instructions (SMI-2676)
• Section 7: Add "Pause — unlock and re-run" option to Stage 4 consent gate (SMI-2676)

v1.1.0 (2026-02-18)

• Add Stage 0 infra detection routing (ADR-109)
• Add sparc-methodology to composes as Stage 1a for infra changes
• Add --infra / --feature override flags
• Update description and Related Skills table

v1.0.0

• Initial release: 4-stage pipeline (Plan, Review, Issues, Execute)
• Interruptible stage gates with standardized 5-option prompts
• Resume detection via multi-signal artifact checks
• Skill existence checks with graceful fallbacks