Skip to content

Acknowledge draft commands and restore model execution - #495

Merged
robgruen merged 1 commit into
mainfrom
dev/robgruen/draft-eyes-ack
Oct 30, 2026
Merged

robgruen merged 1 commit into
mainfrom
dev/robgruen/draft-eyes-ack

Conversation

@robgruen

Copy link
Copy Markdown
Contributor

Add native eyes reactions to owner draft-command comments and fix the premature completion signal exposed by the first live draft attempt.

Changes

  • Enable on.reaction: eyes for Issue and Discussion comment triggers. Keep status-comment: false, so acknowledgment does not create a public status comment or publish the proposed response.
  • Keep reaction writes in the trusted activation job; the model, detector, and sealing jobs remain read-only apart from their existing inference permission. The pinned compiler includes pull-requests: write for issue_comment reactions, but TypeChat's owner-only guard still rejects PR comments before activation.
  • Remove the synthetic pre-agent noop introduced in #494. The runtime interprets that output as completed work and skips model execution. Restore the normal no-op allowance and ask the model to emit completion only after writing the candidate file.
  • Preserve the executed-scan, final-verdict, exact-byte binding, owner approval, and publication-provenance checks. No budgets or publisher permissions are changed.
  • Update owner guidance and regression tests.

Live failure evidence

The owner-requested run 41136773812 for #75 logged noop message found in safe-outputs and skipping agent (work is already complete or no work needed) during model startup. Execution evidence remained not_started; no candidate or sealed draft was produced. Detection and sealing failed closed, and no response was published.

Validation

  • 126 automation tests passed with Node 24.22.1 (npm test from tools).
  • Both workflows strictly compile with gh-aw v0.93.5, with zero warnings and no repeat-compilation drift.
  • Tests cover native acknowledgment configuration, owner/PR guards, model read-only permissions, absence of premature completion, and the compiled Bash detection guard for completion, incomplete, and missing signals.
  • The backlog workflow, runtime/compiler pins, and .gitattributes are unchanged; whitespace checks pass.

Rollout

No live retry or bot publication was performed. After human review, merge, and post-merge CI, create a new /typechat draft comment for verification; rerunning the old failed run reuses its old workflow revision. The eyes reaction is acknowledgment only—not confirmation of a successful draft or acceptance of work.

Use native eyes reactions on owner command comments without public status comments or model write permissions. Remove the synthetic pre-agent noop that made the first live draft skip model execution, and request completion only after the candidate file exists.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot AI balanced review requested due to automatic review settings October 30, 2026 23:29

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🔵 Needs a closer look

The fix depends on inaccessible pinned gh-aw runtime behavior, and fresh owner-led verification of model execution, detection, and sealing remains outstanding.

No open issues 🎉

What changed in this PR

This PR restores model execution in TypeChat’s owner-requested drafting workflow and adds acknowledgment reactions without publishing status comments or changing draft approval requirements.

Changes:

  • Add native eyes reactions, with reaction-write permissions confined to trusted activation.
  • Remove the premature completion signal and instruct the model to report completion only after writing the candidate.
  • Update regression coverage and owner guidance while preserving scanning, sealing, and publication safeguards.
File Description
tools/​tests/​discussion-workflows.test.mjs Checks acknowledgment, permission boundaries, and completion-signal handling.
OWNER-CHEATSHEET.md Explains acknowledgment reactions and where to check progress.
AUTOMATION.md Documents reaction permissions, the startup failure, and fresh live-verification requirements.
.github/​workflows/​discussion-draft.md Enables reactions and moves completion signaling after candidate creation.
.github/​workflows/​discussion-draft.lock.yml Regenerates the workflow with reaction support and corrected completion configuration.

Review effort: Balanced

@robgruen
robgruen merged commit 6df1633 into main Oct 30, 2026
14 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