Repository navigation
Add evidence-backed completion verification and current Codex profiles - #27
Conversation
Add original MIT explicit-file goal, finding, verification and coverage accounting with independent scope/revision/policy/plan bindings. Preserve existing ECF and six workflows; no hooks, providers or authority changes. 73 local Linux/Node 22.16.0 regression tests passed. Full repository and supported-host gates remain required; synthetic records are not proof.
rhein1
changed the title
…runtime directory
Codex Review SummaryThis 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. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0f10928af5
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Please independently re-review exact head cc5ae19. All five findings were reproduced, repaired and regression-tested; three read-only local review lenses found no remaining scoped blocker. Full source and installed-tarball suites: 350 tests, 346 passed, 4 visible platform skips, zero failures. Package and PowerShell validators pass. Fresh isolated supported install/upgrade/rollback/final-upgrade and primary-skill discovery pass. Exact-head hosted CI is tracked separately and is not inferred from local tests. Original frozen model comparisons remain historical under their original grader. Eligible latest-push approval is still required; branch protections will not be bypassed. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: cc5ae199cb
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Please re-review exact head 37766f4. The three additional findings were reproduced, repaired and regression-tested (including the related ignored-file scope bypass). Final local source and packed suites: 353 total, 349 passed, 4 explicit platform skips, no failures; package/PowerShell validators and supported isolated install/upgrade/rollback/discovery pass. Three read-only review lenses found no scoped blocker. The prior head Windows Node18 RRSI source suite exhausted its whole-suite 180-second budget after 27 passes; the unchanged 30-test suite completed locally in 91.180 seconds, so no unrelated runtime change or timeout relaxation was made. That failed CI remains recorded; new-head full CI is tracked independently. Eligible latest-push approval remains required; no bypass or publication is authorized. |
|
Codex Review: Didn't find any major issues. What shall we delve into next? Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
|
@codex review Please review exact head 34f8047. The last review at 37766f4 completed with no findings. This additional two-file change only makes the Windows offline source-suite aggregate budget finite300s instead of180s after repeated180s exhaustion (both WindowsNode18/24), adds verbose case names and elapsed/deadline diagnostics, and documents the distinction. All30 Python tests, each Git30s limit, other suites/platforms180s, matched model-trial deadlines and release/protection gates remain intact; no runtime/plugin bytes changed. Full Windows source and installed-tarball suites pass353/349pass4skip0fail. Current installed plugin matches the unchanged tested plugin tree; no redundant personal rewrite was done. New-head hosted CI must pass independently; eligible latest-push approval is still required. |
|
Codex Review: Didn't find any major issues. Another round soon, please! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
jmborden1112-beep
left a comment
jmborden1112-beep
left a comment
There was a problem hiding this comment.
Owner-authorized administrative approval from the owner's existing secondary account. Both accounts belong to the same maintainer; this is NOT a second independent human review. Exact head: 34f8047. Independent automated/read-only agent technical reviews found no remaining scoped blocker; exact-head cloud code review has no findings, all eight review threads are resolved, and all eight CI checks passed (run 39409354034). Local source and packed suites: 353 total, 349 passed, four disclosed platform skips. The owner explicitly requested use of this account and a normal protected merge. No branch-policy bypass, release/publication, deployment or security-setting change is authorized by this approval.
The merge-base changed after approval.
jmborden1112-beep
left a comment
jmborden1112-beep
left a comment
There was a problem hiding this comment.
Fresh owner-authorized administrative approval after resolving stale PR base metadata. Both accounts belong to the same maintainer; this is NOT independent human review. Reviewed head34f8047df26aaae78d3dd9aa66d4c09906aba189 against current main4e5a77e67b6b4269676977b73ae6c520e7312ef9. Exact tested merge43247f5affd9429ca6134ed097886c2678249b39 has those parents and the same tree as the head. PR base metadata now agrees with the actual main ref; no source/head/base-branch target or protected policy changed. All eight checks and final-head cloud code review remain clean, and all eight threads are resolved. Independent read-only technical review found no remaining scoped blocker. The prior approval was dismissed with The merge-base changed after approval and is not counted. The owner explicitly authorized this secondary-account approval and normal protected merge; no publication/deployment/admin bypass.
Purpose
Make Fable's primary workflow evidence-backed completion verification while retaining its six existing skills. This extends the original PR rather than duplicating PR #27's closeout work.
Merged by normal protected squash merge at
a2aa9efc3b04cdc924ac9620f050db530760faf0. Merged-main CI passed all eight checks; canonical main and the verified installed plugin are updated. This is still unreleased0.5.0-alpha.1, not npm/directory publication.Implementation
$fable-verifyquick/deep modes, versioned requirement/evidence artifacts, working-tree/index revision binding, stale detection, readable reports and permission-free handoffs.RECORDS_SATISFIED,evidenceVerified:falseandauthorityGranted:falseremain accounting, not proof that tests ran or work completed.Observed validation
fable5-codex:fable-verify. No model turn is implied by discovery.0.163.0-alpha.10.2and isolated Linux Codex0.163.0advertisegpt-6.1-sol/medium. Authenticated Linux turns with this explicit selection succeeded; actual server-routed model/effort are not attested by retained JSONL. No separate paid API was used.0f10928af5abc2b1ada6c914b1a9a570a7930b34completed all six trials: both arms scored 1.0 across seven criteria each, with zero missed defects, false completion claims, false blockers or invalid oracle citations under its original grader. Baseline/Fable child time totals were 81.879s/84.871s and reported tokens 192,546/234,230. These immutable records are not a rerun or qualification of the new build/grader. Costs remain unknown; no semantic correctness advantage, savings, generalization or promotion eligibility is established.0.5.0-alpha.1after backup; full Windows package fingerprint matches tested source. Fresh app-server and the current managed-chat skill catalog expose the primary skill. Unrelated configuration/plugins, marketplace and original auth-file preservation were checked. Older open chats may need a new chat; Windows standalone inference with credential-read denies remains unverified.Review and delivery boundaries
Independent read-only reviewers examined closeout/provenance, evaluation isolation, packaging and the final diff. The first five cloud-review findings were reproduced, repaired and regression-tested in
cc5ae199cb8f9b83573181cbbc679a804c0cfe63: complete tested-tree CI binding (including hidden edits and other-SHA/path changes), file/directory snapshot handling with distinct deletion/type-replacement identities, strict one-to-one grader IDs, nested request/report isolation, and maximum-length derived reasons. The additional three findings were repaired in37766f4ec4e67f91aeccaf342b257e161eeabf0e: reject opaque untracked repositories and out-of-snapshot/ignored source observations; preserve and execute frozen grader plus evaluator-dependency bytes with before/after checks; avoid caller check-ID collisions. Three independent read-only lenses found no remaining scoped blocker; exact-head cloud re-review was requested. Raw-byte/Windows executable-mode uncertainty stays conservatively unverified, not inferred from a clean status or index metadata.Current head:
34f8047df26aaae78d3dd9aa66d4c09906aba189. Exact-head CI run completed successfully: all six OS/Node jobs, packed artifact and Release gate (8/8). Final-head cloud re-review completed with no findings or unresolved findings; this is not an eligible branch-protection approval.The final two-file adjustment gives only the Windows offline RRSI source suite a finite 300-second aggregate deadline, with verbose case names and elapsed/deadline diagnostics, after both Windows Node18/24 repeatedly exhausted 180 seconds after 27 passes. Final Windows Node24 evidence shows all 30 source tests completed in 188.049 seconds. All 30 tests, per-Git 30-second limits, other RRSI suite/platform 180-second limits, matched model-trial deadlines and release/protection gates remain intact; no production/plugin bytes changed. Failed runs remain historical, not relabeled as passing evidence.
Checkout logs corroborate merge-test
43247f5affd9429ca6134ed097886c2678249b39; its observed parents are pre-merge main4e5a77e67b6b4269676977b73ae6c520e7312ef9plus this head, and its tree equals this head and the squash-merged main. Earlier REST/GraphQL PR base metadata reported154bf0b220566d91d1a4aaf4390d9c3a54926c3a; a same-main metadata refresh synchronized it to the actual tested base before fresh approval, without changing source/head/target branch. CI is validation, not release/publication approval.The owner explicitly authorized administrative approval via the already authenticated secondary account
jmborden1112-beep. Both accounts belong to the same maintainer; this is not a second independent human review. Independent technical reviews were automated/read-only agent reviews. An initial approval was dismissed with the reported merge-base-change reason and is not counted; after synchronizing stale PR base metadata, a fresh exact-integration administrative approval was accepted. The normal guarded squash merge succeeded; merged-main CI passed 8/8. Required reviews, admin enforcement and other controls were not bypassed or changed. Existing credentials were used transiently without changing the default active account, creating credentials or altering scopes.This is unreleased
0.5.0-alpha.1source. Personal installation, fresh-session activation, source tests, hosted CI, merge, release, directory approval and public publication remain separate states. No release, npm publication, directory submission, deployment or credential/account/security change is authorized or claimed.Closes no issue automatically. Tracks #24 without claiming qualified live RRSI promotion or private heldout effectiveness.