Fix P0 Issues

Fix P0 issues

Contract

Inputs

• The repository root, resolved via git rev-parse --show-toplevel.
• The most recent weekly product briefing under reports/weekly_product_briefings/*.md.
• Optional: GitHub issue numbers embedded in briefing P0 sections as #NNNN.

Procedure

1. Resolve the repository root. Find the most recent briefing by sorting filenames, which contain YYYY-MM-DD dates. Do not use ls -t because filesystem modification times are unreliable in freshly cloned repos.

   REPO_ROOT=$(git rev-parse --show-toplevel)
   ls "$REPO_ROOT/reports/weekly_product_briefings/"*.md 2>/dev/null | sort | tail -1
   

   Done when: the most recent briefing is identified by path.
2. Parse the briefing for sections starting with #### P0:. For each P0, extract the title (text after #### P0:), GitHub issue numbers (#NNNN patterns), the problem description paragraph(s), and any engineering-alignment notes. Done when: every P0 section is parsed with title, issue numbers, description, and alignment notes.
3. When a P0 includes issue numbers, fetch the GitHub issue details (gh issue view <issue_number> --repo <org>/<repo>). Then search the codebase using keywords from the issue. Document relevant file paths, related code or tests, and dependencies. Done when: every P0 with issue numbers has its issue fetched and codebase searched.
4. Triage each P0 into one category and record the decision:
   • Inactionable Complaint: no specific description, reproduction steps, or fix indication. Skip; note in the report.
   • Bug Fix: clearly scoped, has reproduction steps or a clear error, narrow scope, identifiable root cause. Decision: Spawn Agent.
   • Feature Request: assess scope (Small 1-2 files, Medium 3-10 files, Large 10+ files or architectural), complexity, and confidence 1-5. Confidence 5 or 4 → Spawn Agent (4 with caveats); 3 or lower → Skip and note. Skip any P0 the briefing notes is already being addressed on an active branch. Do not spawn more than 3 agents at once. Done when: every P0 has a category and decision recorded.
5. For each P0 marked Spawn Agent, construct a prompt containing the issue number and title, problem description, fetched issue details, relevant file paths, and task (investigate, implement fix, add or update tests, create a PR with a clear description). Present the full spawn plan to the user, including P0 titles, issue numbers, categories, confidence, decisions, and constructed prompts. Obtain explicit confirmation before proceeding. Do not spawn any agent without confirmation. Done when: the spawn plan is presented and explicit confirmation is obtained or refused.
6. For each confirmed P0, spawn an ODIN cloud coding agent:

   odin agent run-cloud \
     --environment <ODIN_ENVIRONMENT_ID> \
     --prompt "<constructed prompt>"
   

   Use exactly odin to spawn cloud agents. If odin is not available, stop and report the error rather than substituting another binary. Done when: every confirmed P0 has a cloud agent spawned (or odin is unavailable and the skill stops).
7. Record each spawned agent's run ID and emit the report described in Output. Done when: every spawned agent's run ID is recorded and the report is emitted.

Failure and recovery

• No briefing found: stop and report that no weekly product briefing exists under reports/weekly_product_briefings/; do not spawn agents.
• odin unavailable: stop and report the error; do not substitute another binary or spawn method.
• GitHub issue fetch fails: continue with briefing-only context for that P0; note the fetch failure in the report.
• User declines confirmation: do not spawn any agent; report the declined plan as the terminal result.
• Partial-result rule: agents already spawned remain spawned; report their run IDs. No rollback of remote runs is performed by this skill.

Output

Use this report form:

=== P0 Fix Agents Spawned ===

P0: <Title>
- GitHub Issues: #NNNN, #NNNN
- Category: <Bug Fix | Feature Request | Inactionable Complaint>
- Confidence: <1-5> (Feature Requests only)
- Decision: <Spawn Agent | Skip>
- Run ID: <run-id>
- Monitor: odin run get <run-id>

...

Skipped:
- <Title>: <reason>

To check all runs:
odin run list --output-format text

Skipped P0s (inactionable, low-confidence, or already addressed) are listed with their reason. No agent runs without user confirmation.