skills.rest
D1ssolve
D1ssolveCOMMUNITY
Explore all skills by this author →
GitHub

pm-routing

Routes tasks through staged multi-agent chains with dual-model planning and artifact tracking.

1|Updated Jun 1, 2026
Search Tags:#workflow#planning#routing#multi-agent#agent orchestration#task delegation#dual-model
One-click install
npx skills add https://github.com/D1ssolve/craft-agents --skill pm-routing-d1ssolve
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: pm-routing
Source: https://github.com/D1ssolve/craft-agents/tree/main/skills/pm-routing
Command: npx skills add https://github.com/D1ssolve/craft-agents --skill pm-routing-d1ssolve

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coordinating multiple specialized AI agents on a software task is error-prone: stages get skipped, planning is bypassed to save time, and chain state is lost between sessions. This Skill gives a Team Lead agent deterministic routing logic so every task follows the correct Discovery, Planning, Execution, and Validation sequence. ## Core Features & Use Cases - Dual-model planning policy: Runs specification and architecture stages with paired GPT/Sonnet agents by default, with scored comparison and hybrid merging of outputs. - Persistent chain state: Tracks stage, mode, artifacts, and blockers in .ai/chain-state.md so work survives session restarts. - Stage gating and failure handling: Enforces mandatory planning gates, degraded-mode fallback when one dual branch fails, and escalation rules for repeated remediation loops. - Use Case: A vague feature request arrives; the router runs research and code exploration in parallel, then spec-driven-dual, system-architect-dual, batched backend-developer tasks, and a final parallel code review. ## Quick Start Route this feature request through the full agent chain starting with dual-model specification and architecture planning.

Frequently Asked Questions about pm-routing

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I orchestrate multiple AI agents for a feature implementation?▼

Use a staged chain: discovery agents first, then dual-model specification and architecture planning, then batched parallel developer tasks, and finally code review. Each stage produces named artifacts that gate the next stage.

When can single-model planning be used instead of dual-model?▼

Single-model mode is allowed only when the user explicitly requests one agent by name, or the change is minor: one file or fewer, no new artifacts, no ADR, and no planning ambiguity. Otherwise dual-model planning is the mandatory default.

How does chain state persist across sessions?▼

State is written to .ai/chain-state.md recording the current stage, locked mode, produced artifacts, and blockers. On resume, the router reads this file first to recover position and outstanding issues.

What happens when one branch of a dual-model run fails?▼

The chain continues in degraded mode using the successful branch, and the failure is documented in the compare artifact and final report. If both branches fail, the chain halts and asks the user to retry, switch to single-model, or abort.

When can the specification and architecture stages be skipped?▼

Only when .ai/spec.md already exists with status [FINAL] and the user confirms it is current, or when .ai/adr.md and .ai/tasks.md exist and chain state records the Architecture stage as complete.

Related Skills

Explore More Software Engineering Skills→
edwifiguyedwifiguy
1

structured-research-workflow

Define research scope, search sources, and synthesize cited findings.

Community
Intermediate
besync-labsbesync-labs
7

intelligent-routing

Route user requests to specialist agents using a deterministic routing matrix.

Community
Advanced
KevinKE93KevinKE93
42

documentation-and-adrs

Record architectural decisions and supporting documentation for future engineers and AI agents.

Community
Intermediate
coco-researchcoco-research
264

gsd-fast

Execute trivial code changes directly in the current context.

Community
Basic