| name | agent-issues-workflow |
| description | (Codex) Build a ranked backlog from open issues, or execute a ranked subset of issues by delegating to the Orchestrator in sequence. |
Issues Workflow
Use this skill when working with project issues — either building a ranked backlog or executing ranked issues with an explicit target.
Tool Access
If you need a tool that is not available (e.g. gh CLI not installed or not authenticated), stop and tell the user what tool you need, why, and what they should do to provide access. Do not invent ad-hoc alternatives.
Modes
Backlog mode (default)
Fetch all open issues, score and rank them, and return a prioritized list with a short rationale for each ranking decision. Do not execute any work.
Execute mode
Take a ranked issue list (from a prior backlog run or provided by the user) and execute each issue in order by delegating to the Orchestrator skill. Require an explicit target and explicit issue scope before starting.
Inputs
- Issues: a comma-separated list of issue numbers, or
all open (default).
- Mode:
backlog (default) or execute.
Ranking Criteria
Score each issue on:
- User impact — how many users are affected, and how severely?
- Frequency — how often does the problem occur?
- Blocking dependency — does this block other planned work?
- Fix effort — estimated size relative to impact (prefer high-impact, low-effort)
- Risk — potential for regression or data loss if left unresolved
Present the ranked list as a table: issue reference, title, score rationale, estimated effort, suggested priority.
Execution Rules
- Issue comments are authoritative. When executing, read the full comment thread for each issue. Comments posted after the original body (review findings, blocker lists, updated acceptance criteria) supersede the original body. Pass updated scope to the Orchestrator as binding constraints.
- Execute one issue at a time. Do not start the next until the Orchestrator confirms the previous is closed.
- If an issue is blocked by another open issue, flag it and skip to the next unblocked one.
- Do not redefine ranking logic mid-run. If the user wants to re-rank, return to backlog mode first.
Output
Backlog mode: one row per issue with: reference, title, rank, score rationale, estimated effort.
Execute mode: after each issue closes, one row with: reference, terminal status, resolution summary, any issues recorded during the run.
