setup-agent-routing

v2026.09.24

Sets up an `## Agent skills` routing block in CLAUDE.md/AGENTS.md plus docs/agents/ so the dev-loop skills (executing-plans, feature-intake, prd-writer, qa-reviewer) know this repo's GitHub issue tracker, kanban label vocabulary, and domain doc layout. Run once per repo before first use of the loop, or when those skills appear to lack tracker, label, or domain context.

GitHub
Install command
npx skhub add shipshitdev/setup-agent-routing
Markdown
SKILL.md

Setup Agent Routing

Configure one thin routing index and consumer tracker/domain documentation for the prepared issue workflow. Reuse the canonical readiness and delivery contracts; setup records where to find them, not another execution procedure.

Authorized Scope

Preserve the user's target and existing write authorization. Read-only/report-only requests stay read-only. Existing approval for this setup covers the requested local files; ask only for consequential missing choices or expanded authority. Model, effort, account, checkout and provider policy remain harness-owned.

Contract

Inputs:

  • Target repository, existing agent instructions and docs/agents/ layout.
  • Live tracker/project fields, labels, and user-approved overrides.

Outputs:

  • One ## Agent skills index and populated tracker, triage and domain documents.
  • Present/missing state, canonical resource locations and unresolved setup blockers.

Creates/Modifies:

  • The existing CLAUDE.md or AGENTS.md routing block, updated in place.
  • docs/agents/{issue-tracker,triage-labels,domain}.md, preserving unrelated content.

External Side Effects:

  • Read-only GitHub inspection and authorized local documentation writes.
  • No issue labels, board changes, workflows, model configuration or dispatch writes.

Confirmation Required:

  • Ask only for a missing target/authority or consequential choice not settled by session context or repository convention. Prepare drafts before any required approval; do not ask again between writes already covered by that approval.

Delegates To:

  • File pointer: executing-plans and its installed delivery reference define execution.
  • File pointer: prd-quality-gate and its installed readiness reference define preparation.
  • Recommend domain-modeling to maintain the discovered domain layout separately.

1. Inspect Existing Context

Read remote identity, current agent instructions, docs/agents/, domain documents, labels, and the configured project. Inspect live board fields and paginate queries when complete inventory is needed. Preserve existing vocabulary and use native priority where configured; do not invent board IDs or duplicate fields.

Resolve executing-plans and prd-quality-gate through the active skill catalog. Resolve references relative to each selected installed directory, never an assumed source checkout. When provisioned, .github/agent-dispatch.md and its setup-pinned .github/agent-skills/ resources own workflow dispatch. Missing required resources are setup blockers, not permission to seed a weaker fallback contract.

2. Settle Missing Setup Choices

Use existing evidence for repository/project, label strings, domain layout and instruction-file selection. Prefer the established instruction file; if both exist, follow their documented canonical/generated relationship. If neither exists and repository convention is absent, ask which one to create. Ask only unresolved consequential questions; do not poll the user for known decisions.

Use the triage seed for the label roles. It records dispatch:plan as the OpenAI planner workflow, with models/effort selected by the harness. Execution gates require a current prepared contract and explicit authority; AFK is earned by readiness, not implied by a label. Identify human-only and planner-owned blockers separately.

3. Write a Thin Index and Consumer Docs

Within existing authorization, update the ## Agent skills block in place and seed only missing or stale sections of docs/agents/. Preserve user overrides and other content. Use the reference templates as location/vocabulary guidance; do not copy execution steps or complete issue templates into multiple documents.

## Agent skills

### Issue tracker

GitHub Issues on <owner/repository>, project #<number>.
See docs/agents/issue-tracker.md for verified tracker configuration.

### Prepared delivery

Resolve executing-plans and prd-quality-gate through the active skill catalog;
read their installed delivery/readiness references. For provisioned workflows,
read .github/agent-dispatch.md and use its pinned skill resources.
See docs/agents/triage-labels.md for local label roles and delivery-state mapping.

### Domain docs

<Single-context or multi-context layout>. See docs/agents/domain.md.

Ensure the seeded docs preserve these invariants: confirm the run ended before explicit claim recovery; implementation PRs use Refs #<issue> and hand off as review_pending; Done requires independent actual review from a different model provider/lab, green required CI at the final head, verified merge and required deployment evidence. Keep the epic open until its integrated acceptance is complete. A review request, self-QA or reviewer assignment cannot substitute for review.

4. Verify and Report

Read the resulting index and referenced docs. Check that resource resolution works in the target installation, the index is not duplicated, and no generated seed contradicts the shared contracts. Report changed paths, missing resources, and unresolved setup choices. Setup completion does not claim a live dispatch or review route has been exercised; verification of those routes requires actual evidence.

Reference Seeds

Discovery
Tags

No tags published for this skill.

Version
Latest version metadata

Version

v2026.09.24

Published

Sep 24, 2026

Category

Uncategorized

License

Not specified

Source path

skills/setup-agent-routing

Default branch

master

Latest commit

a0f9899

Tree SHA

f05942e