Skip to content

Add "Worker Bee" agent for autonomous CI-driven PR iteration#4048

Merged
raineorshine merged 3 commits into
cybersemics:mainfrom
fbmcipher:persistent-pusher
Apr 10, 2026
Merged

Add "Worker Bee" agent for autonomous CI-driven PR iteration#4048
raineorshine merged 3 commits into
cybersemics:mainfrom
fbmcipher:persistent-pusher

Conversation

@fbmcipher
Copy link
Copy Markdown
Collaborator

@fbmcipher fbmcipher commented Apr 7, 2026

Please see discussion #4047 for more details.

  • Introduces "Worker Bee", a general coding agent for Copilot that autonomously manages the full lifecycle of code changes – branching, committing, opening draft PRs, and iterating on CI failures until all checks pass.
  • This replaces the unsteered Copilot coding agent
  • Solves the core problem: Copilot's default agent doesn't check CI status and hallucinates about test results. PP uses GitHub MCP server tools to poll workflow runs, retrieve logs, and fix failures in a loop.
  • Agent persona ("confident, reliable, diligent engineer") is intentionally chosen to activate internal model vectors that reduce reward-hacking under pressure.
  • Handles edge cases: respects explicit requests for failing regression tests (verifying only the intended test fails), can invoke
    /puppeteer-update-snapshots for intentional UI changes, and escalates to the user when uncertain.
  • Safety cap of 5 CI fix-push cycles before stopping and escalating.
  • Tested on two prior issues (Test: Caret moves to incorrect thought from note #3973, Add failing test: gestures broken after dragging duplicate thought to Home #4045) where the default agent failed
  • WB produced correct results on both (see discussion thread for detailed results)

@fbmcipher fbmcipher changed the title Add "Persistent Pusher" agent for autonomous CI-driven PR iteration Add "Worker Bee" agent for autonomous CI-driven PR iteration Apr 9, 2026
@fbmcipher
Copy link
Copy Markdown
Collaborator Author

@raineorshine Agent renamed and ready to merge!

Re: the failing TDD tests. I noticed these new actions and wanted to discuss them with you yesterday at our AI session, but we ran out of time. I think let's hold off on a full discussion about these until the time feels right – but just wanted to flag that the tests seem a little flaky here?

I'm not 100% on what they're supposed to do, but I can see that they detect changed files. The TDD workflows fail because they detect changes to all these files, which I haven't changed in this PR:

package.json
packages/webview/package.json
src/actions/__tests__/swapParent.ts
src/actions/formatSelection.ts
src/actions/moveThought.ts
src/colors.config.ts
src/commands/newUncle.ts
src/commands/note.ts
src/commands/undo.ts
src/components/ColorPicker.tsx
src/components/Note.tsx
src/components/VirtualThought.tsx
src/e2e/puppeteer/__tests__/color.ts
src/selectors/getSortedRank.ts
src/util/compareThought.ts
src/util/noteValue.ts
yarn.lock

=== Unit test files ===
src/actions/__tests__/swapParent.ts

=== Puppeteer test files ===
src/e2e/puppeteer/__tests__/color.ts

=== iOS test files ===


=== Non-test application files ===
src/actions/formatSelection.ts
src/actions/moveThought.ts
src/colors.config.ts
src/commands/newUncle.ts
src/commands/note.ts
src/commands/undo.ts
src/components/ColorPicker.tsx
src/components/Note.tsx
src/components/VirtualThought.tsx
src/selectors/getSortedRank.ts
src/util/compareThought.ts
src/util/noteValue.ts

My fbmcipher:persistent-pusher branches off of an earlier commit in main, so maybe it's picking up the changes made in main since then, too.

@raineorshine
Copy link
Copy Markdown
Contributor

Thank you and sorry for the inconvenience! I'm still working out some issues with the TDD workflow. See my Discord message in the dev channel for full details.

@raineorshine raineorshine merged commit 43fe967 into cybersemics:main Apr 10, 2026
6 of 9 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants