Repository navigation
Bound native autopilot observation to two opportunities - #151
Open
dfrysinger wants to merge 1 commit into
Open
dfrysinger wants to merge 1 commit into
dfrysinger wants to merge 1 commit into
Conversation
Add an optional two-opportunity boundary to the existing native evaluator. Preserve the initiating send, require correlated final responses and turn ends, pass native continuations unchanged, and stop at the selected boundary without changing the default idle behavior. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: cf89240b-a65d-414a-9123-5c1dc9c40c7d
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Adds an opt-in
native_opportunity_limit: 2to the existing capped native continuation evaluator. The default observation path stays unchanged.The evaluator sends the initiating request once, passes the next native prompt through unchanged, and counts ordinary completions only when the final answer has a matching turn end. It stops at the selected boundary and holds further prompts during shutdown. Genuine task completion, protected input, errors and the original deadline remain separate exits.
This includes the capped source-bound restoration prerequisites used by the continuation path. It changes no target-skill instructions, permissions or time/credit limits.
Verification covers 38 focused owning and legacy checks, including missing/mismatched turn ends and non-final text followed by idle. A native setup performed file, shell and SQL work across two opportunities, then stopped before a third with the objective still active and clean shutdown. This is evaluator setup evidence, not a skill-improvement or reliability claim.