Skip to content

requirements.md

Latest commit

 

History

History
96 lines (61 loc) · 2.67 KB

requirements.md

File metadata and controls

96 lines (61 loc) · 2.67 KB

PatchPro — Pod-Level Requirements (Sprint-0)

This document breaks down requirements for each pod to support our Sprint-0 goal: a comment-only vertical slice on demo PRs. Each pod should fill in its section with specific requirements, actions, and outputs.

1. Agent Core

Scope: Prompt scaffolding & guardrails for generating minimal diffs.

Requirements:

  • Define the prompt format (inputs, expected output style, minimal diff block).

  • Add guardrails:

    • Max number of lines changed.
    • Handle large files gracefully.
    • Fallback behavior if LLM output is empty or invalid.

  • CLI entrypoint: patchpro agent run --inputs <findings.json> --out patchpro.md.

  • Output spec: structured markdown snippet for sticky PR comment.

2. Analyzer/Rules

Scope: Normalizing static analysis findings (Ruff, Semgrep).

Requirements:

  • Pin versions of Ruff and Semgrep.
  • Define config baseline (e.g. .ruff.toml, semgrep.yml).
  • Ensure findings exported as JSON with consistent schema.
  • Write schema: schemas/findings.v1.json.
  • Normalize: deduplicate, unify file:line format, add severity labels.

3. CI/DevEx

Scope: GitHub Actions workflow orchestration.

Requirements:

  • Create workflow patchpro.yml triggered on pull_request.

  • Steps:

    1. Checkout repo.
    2. Run Ruff & Semgrep → export artifacts.
    3. Run PatchPro agent → generate patchpro.md.
    4. Post sticky PR comment with patchpro.md.

  • Permissions: minimal (contents: read, pull-requests: write).

  • Concurrency: ensure only 1 workflow per PR runs at a time.

  • Timeout: ≤ 5 min per job.

4. Eval/QA

Scope: Golden PRs and evaluation rubric.

Requirements:

  • Create 3–5 golden PRs in patchpro-demo-repo (common lint/fix cases).

  • Document expected PatchPro comment for each.

  • Define rubric (LLM-as-judge or human):

    • Did PatchPro detect the issue?
    • Did it propose the correct minimal diff?
    • Was the comment structured and clear?

  • Track pass/fail results per golden PR.

5. UI (Optional)

Scope: Playground for manual prompt/diff testing.

Requirements:

  • Minimal web/CLI playground to test prompts locally.
  • Input: sample findings JSON.
  • Output: markdown preview of suggested comment.
  • Keep styling consistent with CI sticky comment.

Next Steps

  • Each pod to fill in their section by [set date, e.g. Wed, Sept 3].
  • Once filled, we’ll open GitHub issues/branches mapped to these requirements.
  • Eval/QA will then test the first end-to-end vertical slice using golden PRs.

⚖️ This doc bridges high-level flow → actionable tasks. When filled, it becomes our shared reference for Sprint-0.