Repository navigation
Clarify specialist skill access and refusal delivery - #7
Merged
Merged
Conversation
added 6 commits
November 12, 2026 23:43
added 4 commits
November 13, 2026 00:02
hudsonaikins
marked this pull request as ready for review
November 13, 2026 05:05
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
hudsonaikins
changed the base branch from
codex/analysis-eval-foundation
to
main
November 13, 2026 05:09
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The specialist was probing virtual /skills paths inside the VM shell, causing avoidable file errors before recovering. Clarify the instruction filesystem versus VM tools, and require analytical completion summaries and useful missing-input refusals without substitute reports or temporary-delivery promises. Sandbox permissions, model selection and export enforcement are unchanged.
The fresh live paired experiment used GPT-6.1 Sol at medium reasoning through the existing ChatGPT subscription on SmolVM: four cases per arm, one trial each. Both baseline 9b75233 and candidate 3d0bb71 passed 4/4, including a separately recorded regrade with PR #6's corrected evaluator. Each exported three verified bundles and one refusal. Virtual-skill shell mistakes fell from 4/4 baseline cases to 0/4 candidate cases. Both refusal trials already behaved correctly, so no refusal-quality improvement is claimed.
Baseline: 41 calls, 401,583 input / 10,325 output tokens, 372.663s summed latency. Candidate: 34 calls, 387,065 / 11,650 tokens, 408.103s. Total 75 calls; complete subscription usage; no paid API fallback, retry or credit purchase. Latency is descriptive, not a controlled speed comparison. Candidate energy recovered one incorrect workbook-sheet-name assertion before successful independent verification.
Baseline experiment · Candidate experiment. LangSmith read-back verified all eight run records and 26 additional versioned/behavioral feedback records; original feedback is preserved.
Validation: actual-graph skill-file/refusal regression passes; the initial guidance branch passed 188 tests with 15 optional skips. The underlying package's revised checks also pass, including six real source stores. Updated README and results preserve historical failures and state the limited scope. No production reliability, unfamiliar-domain or v0.0.1 release acceptance is claimed.
Builds on merged PR #6 and now targets main. Synchronization commit 648d41a preserves exactly the tested tree at 664e956; only Git ancestry changed. Full plan, identities, results and limitations are in docs/results-analysis-sandbox.md. Tracks INT-451 and INT-454.