Repository navigation
feat(rrsi): test cumulative host budgets and validate Memory review artifacts - #448
Conversation
…entation packet Add dependency-free offline contracts, synthetic replay, inert upstream Domain adapter, source pins, 36 tests, cross-platform CI, and staged Codex work order. No model execution, provider activation, installed-skill promotion, or deployment.
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
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: 8a7ffba2d4
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".
Preserve the reviewed 25-path source-only RRSI surface and take current main for all other files, including authentication, SkillSpector, durable handles, manifests, and hosted bundle.
jmborden1112-beep
left a comment
jmborden1112-beep
left a comment
There was a problem hiding this comment.
LGTM
Needs Fixing
None. The source-only RRSI host remains hard-off before upstream construction; synthetic budgets retain unknown/overrun holds, and the artifact-chain Memory importer remains quarantined without execution/promotion authority.
Requires Human Review
None beyond the repository gate fulfilled under the owner's second-account authorization. This is agent source review, not independent human sign-off.
Recommended Optional
None.
Create Follow-up Issue
None.
Workflow Trace: exact refreshed head3ba3ca1614946dfa8ead7e1a2ec02add41793ccf. /root/integrations_stack_review plus root verification; actual worker fallback gpt-6-luna medium. All25 PR-changed file blobs remain identical after the main refresh. Root ran61/61 offline tests with zero skips after reading and verifying the two exact pinned upstream ABI modules in an existing isolated checkout; only pure contract modules ran, not upstream package initialization/model/search. Initial unconfigured run skipped4 ABI tests; the configured repeat removed those gaps. All10 final-head hosted checks passed; integration counts and generated rename inventory checks pass. No live search/model calls, host activation, Memory writes, publication or spend.
Extend #436 with source-only host accounting/lifecycle contracts and a reviewed-Memory artifact importer. Budget reservations retain cumulative spent costs across retries and sessions; unknown consumption suspends the shared budget, overruns require reconciliation, and simulated resume binds scope, checkpoint and grant revision. Same-session transitions are serialized.
The importer verifies the real public stage/task/finalization/review/receipt chain pinned to Memory 3c2f17d. Its synthetic fixture was generated by the actual pinned Node bridge; Python verifies the original Node digests. Imported evidence stays quarantined and cannot authorize evaluation or promotion.
The branch includes #436 and shared prerequisites #444/#446. Current head f15935c includes the shared audit/inventory/hosted-bundle repairs and six reviewed RRSI corrections. The importer binds the stage-view projections to the complete reviewed artifact chain, rejects contradictory no-provider/no-spend evidence and truncated task sets, and retains observed consumption as a reconciliation lower bound. Concurrent sessions cannot reuse one reservation. The validation record distinguishes historical scaffold runs from the current 61-test suite. Exact-head hosted checks: 52 successful, 0 pending, 0 failed when refreshed on October 3, 2026; all four Ubuntu/Windows and Python 3.10/3.13 contract jobs pass 61/61 tests with no skips. Required independent approval remains outstanding.
HostOrchestrator remains hard-off before upstream Run construction or any authority callback, worktree, model or budget side effect. SyntheticHostSession tests illustrative control flow, not an authenticated host or actual optimizer. Durable host accounting, ECF/Risk Fork binding, representative evaluation, real isolation and owner promotion/rollback remain unqualified. No paid model, installed plugin, user Memory or production state is changed.
The shared hosted MCP repair updates its actual lock and embedded fast-uri implementation to 3.1.8, refreshes ip-address to 10.7.3, and regenerates integrity records and notices. A regression now rejects a lock that disagrees with the advertised, installed and attested dependency. All 16 hosted artifact tests, source/artifact verification, packaging and zero-vulnerability audits pass locally. The bundle remains private and publication-disabled.