Skip to content

TODOs.md

Latest commit

 

History

History
40 lines (33 loc) · 2.23 KB

TODOs.md

File metadata and controls

40 lines (33 loc) · 2.23 KB
Raw

Diffcore TODOs

Next: GitHub Integration

Replace GitHub's PR review tab entirely — do the full review in Diffcore, push results back.

Phase 1: GitHub Login + PR Fetching

  • GitHub OAuth login flow in Tauri app
  • Fetch a PR diff by URL — done without an API: the repository field and diffcore analyze --repo take PR/MR URLs across forges and resolve them from the provider's git refs. See docs/pr-url-providers.md.
  • Store GitHub token securely (keychain / 1Password). An API token would also close the gaps git refs cannot cover: Bitbucket Cloud (no PR refs at all) and fast-forward-merged PRs (no recoverable base branch).

Phase 2: Bidirectional Comment Sync

  • Push review comments from Diffcore back to GitHub as a PR review
  • Pull existing GitHub PR comments into Diffcore's review UI
  • Map Diffcore flow groups to GitHub's per-file comment model
  • Approve/request-changes/comment review actions from Diffcore

Phase 2.5: GitHub Comments API — Bidirectional Sync for Repos

Tag each comment with its origin: online (created on GitHub) vs diffcore (created in Diffcore).

  • Fetch all PR review comments via GitHub Comments API (GET /repos/{owner}/{repo}/pulls/{pull_number}/comments)
  • Tag incoming comments as online, tag Diffcore-authored comments as diffcore
  • On sync: pull new online comments into Diffcore, push new diffcore comments to GitHub
  • Conflict resolution: online comments are read-only in Diffcore; diffcore comments are editable in both but GitHub edits create a new online version
  • Display origin tag in Diffcore UI (badge/icon distinguishing online vs diffcore)
  • Support filtering/sorting by origin tag in the review panel
  • Webhook listener (or polling fallback) to detect new online comments in real-time

Phase 3: Structured Agent Feedback Loop

  • Reject a group with structured feedback
  • Generate a prompt for the agent to fix the rejected group
  • Track review iterations (attempt #1 → #2 → #3)

Future: Prompt Alignment

  • Ingest the agent prompt/task/issue that produced the PR
  • Score diff against intent: "here's what was asked, here's what was done, here's what's missing"
  • Flag unnecessary/unrelated changes that don't trace back to the task