Multi-Track Development Planning

A reusable skill that produces agent-proof project architecture for long-horizon, multi-track (frontend/backend/integration) development. It produces a documentation system + execution protocol that lets any zero-context LLM Agent pick up a long-running project and continue development correctly, even when multiple parallel agents work simultaneously across multiple tracks.

What It Does

Drop this skill into any brainstorming or planning session. It produces:

• 7 live documents (PLAN / PROGRESS / DECISIONS / AGENT_PROTOCOL / RECOVERY / BLOCKERS / MERGE_QUEUE)
• A task graph for the current phase with full specs (ID, deps, acceptance, steps, resume_hint)
• Milestone-level outlines for future phases
• A trigger-word protocol for auto-continuing any track (e.g. 前端 开始开发)
• A recovery protocol for any amnesiac agent
• An 8-mechanism parallel-execution safety system
• A task lifecycle state machine (pending → in_progress → done/blocked/failed/stale) with retry + stale detection
• A constitution change protocol preventing silent edits to STRATEGY/SPEC/ARCHITECTURE
• A zero-memory agent relay simulation test as the ultimate acceptance check

When to Use

Invoke this skill whenever the user asks to plan, design, or set up architecture for a project that matches any of:

• A long-running project (>1 session / days / weeks)
• A project that LLM AI Agents will execute
• A project with two or more parallel tracks (frontend + backend, etc.)
• A project where context compression or session switching is expected
• A project where the user wants parallel multi-agent / multi-session execution

Install

Copy the SKILL.md file into your project's skill folder (e.g. .trae/skills/multi-track-development-planning/SKILL.md or the equivalent location for your agent runtime).

Usage

After install, any agent in the project can invoke the skill by name. During brainstorming, the skill replaces the standard design step with a strategic architecture pack + execution protocol + task graph.

Trigger phrases (user → agent):

• <track-name> 开始开发 — start next ready task on that track
• 集成 开始开发 — start next integration task
• 继续 — resume last session's work
• <track> 状态 — report track status without doing work

Documentation

The complete methodology is documented inside SKILL.md. Read sections 0–10 for the full spec.

Quality

The skill ships with 13 invariant checks + a zero-memory agent relay simulation test. See §8 in SKILL.md.

Author

NoWint — https://github.com/NoWint

License

AGPL-3.0 — see LICENSE.