← ClaudeAtlas

workflow-routerlisted

Load when a non-trivial request needs intent recognition before work starts; classify it into exactly one workflow owner, or ask a clarification; do not load for explicit owner-skill invocations, trivial chat, or execution steps already routed.
JasonxzWen/harness-hub · ★ 2 · AI & Automation · score 65
Install: claude install-skill JasonxzWen/harness-hub
# Workflow Router Classify the user's request into exactly one workflow owner before substantive work starts. This skill owns intent recognition and handoff only. It does not implement, test, review, deliver work, or replace the selected owner. ## Owners | State | Owner | |---|---| | question | `answer-workflow` | | sdd-change | `sdd-workflow` | | diagnosis | `diagnosis-workflow` | | review | `review-workflow` | | delivery | `delivery-workflow` | | harness-hub-maintenance | `hub-maintenance-workflow` | ## Workflow 1. For a concrete request, run or mirror `scripts/workflow-check.mjs --prompt "<user request>" --json` before substantive work when a terminal is available. 2. Use `scripts/route-intent.mjs --prompt "<user request>" --json` when only classification is needed. 3. Use `scripts/skill-activation-check.mjs --prompt "<user request>" --json` for one helper skill trigger audit, or `scripts/skill-activation-check.mjs --cases-file "<cases.json>" --json` for a prompt matrix, positive and boundary helper coverage report, and excluded owner summary from installed `SKILL.md` metadata. 4. Use `scripts/owner-contract-check.mjs --state "<state>" --json` only when auditing whether the installed owner `SKILL.md` still contains required sections, ordered gates, and boundary phrases. 5. Use `scripts/helper-contract-check.mjs --json` only when auditing whether high-risk helper skills still carry explicit side-effect boundaries. 6. Read `references/intent-taxonomy.md` when the state