Skip to content

Clarify handoff completion reporting and improve pilot setup - #18

Merged
ArielSmoliar merged 3 commits into
mainfrom
fix/handoff-reporting-and-pilot-instructions
Oct 12, 2026
Merged

ArielSmoliar merged 3 commits into
mainfrom
fix/handoff-reporting-and-pilot-instructions

Conversation

@ArielSmoliar

Copy link
Copy Markdown
Owner

Summary

The pilot dry run performed the merged-but-not-installed task correctly, but its final response omitted the distinction between the earlier merge and installation just performed. Make the resume prompt explicitly request prior work, actions performed now, remaining work, and verification limits. Rebuild the inspector/plugin and bump the package to 0.1.2.

Add a copyable read-only check for the cached marketplace commit, since codex plugin marketplace list --json does not expose the ref. Pin the pilot to tested product commit bd55ab5bdd87b7366baa533b4db4d7650e377bd3, retain a separate packet revision, clarify baseline-first packet timing, and record the original dry-run failure alongside the focused rerun. The candidate PR must be reviewed and merged before distributing the updated packet.

Validation

  • npm test: 48 Python tests plus MCP, managed-install, and browser workflows passed.
  • npm run test:plugin: install, discovery, telemetry, upgrade, pinned rollback, and removal passed.
  • Pinned product CI: both matrix jobs passed on attempt 1.
  • Isolated real GitHub install confirmed version 0.1.2 and the exact candidate commit using the participant snippet; removal preserved unrelated synthetic configuration.
  • Focused scenario 01 rerun passed all four existing criteria. It reused the identical reviewed handoff and current request in a fresh workspace; only the resume prompt's reporting paragraph changed. The receiver explicitly reported prior merged state, new installation, preserved preferences, and untested application behavior.
  • Eight generator tests and clean-checkout kit generation passed; local documentation links and whitespace checked.

The focused rerun is one agent-reviewed resume-only observation, not six new passes or a reliability claim. Original model identity was not captured; this is not a controlled model comparison. No participants were contacted, real pilot results remain NOT RUN, and installed user copies are unchanged. Raw traces stay local; the committed report is sanitized.

@ArielSmoliar
ArielSmoliar merged commit dca2f6b into main Oct 12, 2026
4 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.

1 participant