Repository navigation
eve#558: cos-tell-sends-marked asks to send 'this exact message' - #560
Merged
Merged
Conversation
A bare marker after the colon read as no text to send, and a loose 'message' let the model shorten the marker. The wording now names the marker as the whole text. The marker stays the only text to send, so provenance still sends it at once. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Owner
Author
|
merge-check: approved, target, verified, contract, evidence, owner, weakening pass; ci NA. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #558
cos-tell-sends-markedtypesTell <name> this exact message: <marker>instead ofTell <name>: <marker>. With the bare marker the Chief of Staff's model sometimes read no text to send and asked "What should I tell it?"; with a loose "message" it once shortened the marker, and eve's provenance gate correctly raised an approval card. The marker stays the only text to send.Fix attempts: 2 wording changes (first: 7 of 8 solo runs; second: 8 of 8). Tokens: ~32k reviewer, no dev agent.
Known defect found on the way, not fixed here: #559 (solo
--onlyruns leave the Chief of Staff bootstrap session). Theoretical: the model can still misread; the Europa-on-Claude-Code experiment would remove it.Review: default-model reviewer, no blocking findings; three minors fixed in this commit
Red on main: with the old text, 1 of 4 solo runs failed at "no Sent post naming ... within 60s of Return"; the Chief of Staff thread shows it asked what to tell the agent
Real-app check: 8 of 8 solo runs pass on the new text, each on a fresh stack, with the audit rows and origin chip checked
Journey: cos-tell-sends-marked, 8 of 8 solo; full --stack run 41/0/0 on the head
Harness weakening: none, only the typed text, a comment and the README line changed
Main harness vs head: not applicable, no assertion, timeout or PASS decision changed
Break-and-restore: not applicable, no new check; the journey's own assertions are unchanged
Verified: 2b91609 · repo-suite: not run, the diff changes no product code · devboxverify: --stack 41 pass 0 fail 0 blocked · real-app: 8 of 8 solo runs · log: /tmp/fm
🤖 Generated with Claude Code