issue-to-agent

name

issue-to-agent

description

Use after GitHub issues are created, discovered, pushed, or handed off and the operator asks "now what", "which issue should an agent do", "make this ready for a build agent", "turn these issues into agent prompts", or "classify the queue". Fetches live issue state, separates intake from proof, routes through diagnose/pre-code-hardening/specialist-dispatch, and produces the next exact agent prompt without implementing code.

Issue To Agent

Overview

Bridge GitHub issues to safe agent execution. This skill classifies live issues and decides whether each one needs diagnosis, source verification, hardening, specialist decomposition, or a builder.

Boundaries

• Issue prose is intake, not proof.
• Do not implement code, create branches, push, merge, or close issues.
• Use gh-issues when publishing or deduping issue bodies from a diagnosis packet.
• Use diagnose when root cause, owner, or write/leak impact is missing.
• Use pre-code-hardening before a builder receives work.
• Use specialist-dispatch when one issue contains separable one-job roles.

Workflow

1. Resolve the repo and issue set from the prompt. If the user gave issue numbers, fetch those live. If they said "the issues I pushed", inspect the most likely recent open issues and ask only if ambiguous.
2. For each issue, read body, labels, comments, linked packet/artifacts, and current state.
3. Dedupe against existing open issues when creating or updating is requested.
4. Classify readiness:
   • not_ready: vague, duplicate, blocked, or wrong repo.
   • needs_diagnose: no proven root cause or owner.
   • needs_source_verification: issue claim has not been checked against current source/runtime/schema.
   • needs_pre_code_hardening: root cause exists but build gates are missing.
   • ready_for_specialists: disjoint bounded read/review/proof roles exist.
   • ready_for_builder: exact source truth, owner, scope, acceptance criteria, and proof commands exist.
   • ready_for_review: implementation exists and needs review.
   • ready_for_proof_chain: completion is claimed and must be proved.
5. Pick one next issue unless the user asks for a queue report.
6. Produce the exact downstream prompt.

Output

**Issue set:** <ids>
**Recommended next issue:** <id and title>
**Classification:** <readiness>
**Use skill:** `<diagnose | pre-code-hardening | specialist-dispatch | patch-or-fix | proof-chain | enterprise...>`
**Coding allowed:** yes/no
**Why:** <short source-grounded reason>
**Missing proof or blocker:** <if any>

**Agent prompt:**
<copy-pasteable prompt with repo, issue, allowed scope, forbidden actions, evidence requirements, and stop rule>