scan-issue-writer

name	scan-issue-writer
description	Run a named codebase maintenance scan and convert findings into 6-component, agent-ready GitHub issues. Use when a workflow or user says "run the <name> scan", "create issues from this analysis", or passes a prompts/scans/*.md file. Handles dedupe, priority labeling, and max-issue caps so the Ralph loop never starves and never bloats. Do NOT use for implementing fixes (use stay-green), reviewing PRs (use comprehensive-pr-review), or closing/triaging existing issues (use backlog-grooming).
metadata	{"author":"Geoff","version":"1.0.0"}

Scan Issue Writer

Turn scan findings into prioritized, deduplicated, agent-ready GitHub issues for the autonomous maintenance pipeline. The skill owns the process; each prompts/scans/<name>.md owns the domain of its scan.

Inputs

The invoking workflow (or user) supplies:

• scan_name — which scan to run; selects prompts/scans/<scan_name>.md.
• max_issues — hard cap on issues created this run (default 5).
• priority — P0 | P1 | P2 | P3, applied to every issue filed.

Instructions

Step 1 — Load the scan definition

Read prompts/scans/<scan_name>.md. It defines the tools to run, what counts as a finding, the severity rubric, and the title-slug prefix ([scan:<name>]).

Step 2 — Run the analysis (read-only)

Execute the scan's tool commands. Never modify code. Record the current SHA with git rev-parse HEAD — every issue must cite it in its Context section.

Step 3 — Rank and cap

Sort findings by the scan's severity rubric, highest first. Keep at most max_issues. Findings past the cut are NOT silently dropped: list them in a single run-summary comment (Step 6) so the next scheduled or hopper-dispatched run can pick them up. Never exceed the cap — the hopper controls volume.

Step 4 — Dedupe before every create

For each surviving finding, derive its title slug and search open issues:

gh issue list --search 'in:title "<slug>"' --state open --json number,title

• Exact match, finding still valid → add a comment with fresh evidence ("Still present at <SHA>; confidence …"). Do not create a duplicate.
• Exact match, evidence changed materially → gh issue edit the body.
• No match → proceed to Step 5.

Duplicate issues inflate apparent queue depth with fake work — this step is the single most important behavior for keeping the hopper honest. Other scans may run concurrently, so re-run the dedupe search immediately before each create.

Step 5 — Write the issue

Fill the canonical prompts/templates/scan-issue-body.md completely — all six components, no placeholders left. (It is the single source of truth; references/issue-body-template.md just points at it.) An issue missing any component gets needs-triage instead of agent-ready, and the grooming pass finishes it. Write the body to a file and create with:

gh issue create --title "[scan:<name>] <specific finding>" \
  --body-file /tmp/scan-issue.md \
  --label "<priority>,scan:<name>,agent-ready"

If any of the labels do not exist yet, create them first (see scripts/setup-scan-labels.sh for the canonical set), then retry.

Step 6 — Report

Append a run summary to $GITHUB_STEP_SUMMARY (or print it, when run locally): findings found / created / deduped / capped-and-deferred, each with its issue number. A clean scan reports zero created and that is a success.

Examples

Example 1 — Perf scan finds an unbatched embedding call

Title: [scan:perf] Unbatched generate_embedding() call in fragment loop Labels: P2,scan:perf,agent-ready. Body cites the per-fragment loop, file:line at the scanned SHA, and a generate_embedding → generate_embeddings batch before/after sketch.

Example 2 — Dedupe hit

[scan:dead-code] unused helper _legacy_link_summary in creek_mcp/tools/link.py is already open as #412 → add a comment: "Still present at <SHA>; vulture confidence 100%." No new issue is created.

Troubleshooting

Scan tool exits nonzero with no findings

Distinguish "tool crashed" from "clean scan." A genuine crash creates ONE P1 issue about the broken tooling (with the command and stderr). A clean scan creates zero issues — that is success, not failure.

More findings than max_issues

Never exceed the cap. Summarize the overflow in the Step 6 run summary so the next run picks it up. Raising throughput is the hopper's job (higher max_issues), not this skill's.