Add deterministic vLLM-Omni release drift audit - #30
Merged
Conversation
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
zuiho-kai
force-pushed
the
agent/issue-28-release-drift-audit
branch
from
July 31, 2026 03:40
b96bc59 to
087ac38
Compare
zuiho-kai
marked this pull request as ready for review
July 31, 2026 03:47
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
zuiho-kai
added a commit
that referenced
this pull request
Jul 31, 2026
## Summary Add three deliberate update paths: 1. `/imupdate <local-git-path> [target]` — derive the old SHA from the baseline, default the new SHA to checkout `HEAD`, run deterministic audit, let the host model update evidence-backed facts, then enforce. 2. `/imupdate <repository-name-or-alias-or-url> [target]` — let the host model resolve the canonical repository and version, prefer a configured or temporary checkout, then enter the same machine-audit path. If machine audit is unavailable, report model-only analysis and never claim `CLEAN`. 3. `python tools/audit_vllm_omni_release.py --from ... --to ... --repo ...` — keep the original pure machine-rule interface with no LLM and no file edits. Known aliases `vllm-omni`, `vllmomni`, and `vllm omni` resolve to `vllm-project/vllm-omni`. Target selection refuses silent downgrades when GitHub's latest stable release is older than an already-audited prerelease. The portable installer introduced by #29 now installs all bundled skills for Codex, Claude Code, and Cursor instead of hard-coding only `imreview`. ## Validation - focused installer/release/adapter/routing/knowledge/skill/capability tests — 59 passed - `quick_validate.py plugins/infermatrix-copilot/skills/imupdate` — valid - Ruff check and format check passed - knowledge tree — 0 errors; wiki lint — 0 errors, 0 warnings - plugin manifests parse as valid JSON - real local-path resolution: baseline `807db6ef` to `D:\omni\vllm-omni` checkout `HEAD` `0c88f8a0`, 414 changed paths and 15 drift issues reported - real alias identity check: `vllmomni` maps to authoritative `vllm-project/vllm-omni` metadata and the local checkout remote matches `https://github.com/vllm-project/vllm-omni.git` - GitHub Actions release audit — passed This is a UX follow-up to the already-merged #30.
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
1 participant
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.
Summary
/imupdate <local-upstream-repo> [target-tag-or-sha]as the user-facing Codex/Claude/Cursor skill; it derives the old revision from the audited baseline and defaults the target to the upstream checkout'sHEADtools/audit_vllm_omni_release.pycommand that resolves immutable SHAs, compares Git path changes, registry entries, pipeline keys, deploy YAML, knowledge sources, Direct owner routing, and adapter module coveragereport-onlyandenforcemodesv0.26.0rc1(807db6ef), including Boogu Image, LingBot Video, the Voxtral XPU deploy migration, and renamed CI documentation sourcesWhy
The existing knowledge validators prove that Markdown structure is valid, but they cannot prove that model catalogs, source paths, routing, or adapter coverage still match an upstream vLLM-Omni release. Several independently maintained pins had already drifted.
The low-level audit reads upstream Git objects and parses registry dicts with Python AST; it never imports upstream code or rewrites knowledge rules. The
imupdateskill wraps that evidence collector, lets the host model make bounded knowledge edits, and then reruns enforce mode plus validators.Validation
py -3.13 -m pytest test/test_imupdate_skill.py test/test_release_audit.py test/test_adapters.py test/test_routing.py test/test_knowledge_source.py test/test_skills_scope.py test/test_capabilities.py -q— 57 passedpy -3.13 knowledge/tools/check_knowledge_tree.py— 0 errorspy -3.13 knowledge/tools/check_wiki_lint.py— 0 errors, 0 warningsquick_validate.py plugin/skills/imupdate— validinstall-mcp.ps1; Ruff check and format check passed for the new Python test5d44868etov0.26.0rc1: 679 changed paths, AR+3, Diffusion+2, deploy YAML+6, 0 unexplained issues,RESULT: CLEAN/imupdateinput resolution againstD:\omni\vllm-omni: baseline807db6efto checkoutHEAD0c88f8a0, 414 changed paths and 15 drift issues reported inreport-onlymodeFull-suite validation on Windows remains blocked by existing platform assumptions outside this diff (POSIX path separators/permissions and shell
echoexecution); the focused #28 surface is green.Closes #28