prpPlanOnly

Prompt file imported from pankaj28843/article-extractor (.github/prompts/prpPlanOnly.prompt.md). Copyright stays with the author.

TechDocs Research

Use #techdocs to ground every assertion. Prioritize python, pytest, fastapi, playwright, and docker. Run list_tenants to discover additional documentation sources. Reference .github/instructions/techdocs.instructions.md for detailed strategies.

Mission

• Draft or update a PRP aligned with .github/instructions/PRP-README.md.
• Stay in planning mode: no code edits, migrations, config changes, or tests until stakeholders approve.
• Apply AI-DLC planning rules so each step is explicit before execution.
• Embed the “Do next steps” operating contract in every PRP plan so future execution is unambiguous even if the plan is used only once.

Non-Negotiable: “Do next steps” contract must be included in every plan

Every plan you create/update MUST contain a section titled:

• ## Plan Operator Contract (“Do next steps”)

That section must explicitly encode the following behavior so the user never has to repeat it:

• When the user says “Do next steps” (or “yes do the next steps”), the agent must:
  1. Re-open and re-check the plan file first.
  2. Interpret “next” as the next incomplete checkbox items from the earliest incomplete phase in Implementation Blueprint.
  3. If uncertain, re-check ## What else remains? and proceed from the topmost unfinished item.
  4. After each work chunk, update the plan (Status Snapshot + blockers + checklist + What else remains).
  5. If all plan items are complete: redeploy + full validation loop; record results.
  6. If plan items remain: implement what remains immediately; stop only when done or when a blocker is recorded.

Also require the plan to include:

• ## Status Snapshot (YYYY-MM-DD) (newest first)
• ## Plan Watchers (status cadence + blockers)
• ## What else remains? (single source of truth list, kept in sync with checkboxes)

Required Sections (in the plan output)

1. Plan Operator Contract (“Do next steps”) (required; see above)
2. Status Snapshot (YYYY-MM-DD) (required; newest first)
3. Goal / Why / Success Metrics (tie back to measurable outcomes)
4. Current State (existing modules in src/article_extractor/, dependencies, outstanding gaps, references to specific files/lines)
5. Implementation Blueprint (phased work packages mapped to files + TechDocs evidence)
   • MUST be a checkbox-driven checklist so “next steps” is deterministic
6. Context & Anti-Patterns (cite software-engineering-principles.instructions.md: deep modules, no backward compat, green tests always)
7. Validation Loop (commands per phase):
   • uv run ruff format .
   • uv run ruff check --fix .
   • timeout 60 uv run pytest tests/ -v
   • uv run article-extractor --help
   • uv run scripts/debug_docker_deployment.py (if Docker/Playwright involved)
8. Open Questions & Risks (blockers, missing context, required approvals)
9. Plan Watchers (status cadence, blockers, decision log if needed)
10. What else remains? (required; synced with Implementation Blueprint)

Process

• Gather facts from repo files (src/article_extractor/, Dockerfile, scripts/) and existing docs before drafting conclusions.
• Use TechDocs citations (URL + snippet) for every pattern, architecture, or tooling claim.
• Keep bullets crisp; prefer ASCII tables for evidence matrices or decision summaries.
• Ensure every Implementation Blueprint step is specific:
  • explicit files
  • explicit intent
  • validation commands
  • exit criteria (“done when…”)
• End with a readiness statement: "Ready to implement" or "Need clarification on X" (but do not block on minor ambiguities; prefer clearly labeled assumptions).

Output

• Save/update the plan under: .github/ai-agent-plans/{date}-{slug}-plan.md.
• Final response must:
  • recap key updates,
  • link to the plan file,
  • list unresolved questions or approvals needed before coding,
  • and explicitly confirm the plan includes the Plan Operator Contract, Status Snapshot, and What else remains? sections.