[dinakarjs]

Intermediate-parent symlink repair — 5 November 2026

Current head: 7215f44daf8db01250134f0898f84b6efd08be21. The changed-lock driver now checks every lexical path component before reading a lock. The real-Git regression replaces a tracked nested lock directory with a symlink and confirms the deleted lock is refused; an ordinary-directory edit remains accepted. Both subprocess modes are exercised.

Validation: 21 focused tooling tests pass normally and 21 under optimized Python in each port. Before the repair, the new negative control fails in both modes with PASS requirements/nested/dev.txt; the ordinary control passes. Whitespace checks pass. No runtime SDK code changed and no fresh full SDK suite is claimed. The standalone source checker is unchanged. Current-head hosted results and independent re-review remain separate. Draft and repository-specific adoption holds remain. Prepared with OpenAI Codex; DCO-signed.

---

Problem and change

Generated-lock changes can silently remove dependencies, replace hashes or lose the prior baseline through a rename. This proposed CI guard compares generated locks with the PR base, preserves quoted marker whitespace and lexical paths, and refuses symlink/non-regular lock paths. The driver discovers additions/deletions with rename detection disabled; repository-specific exclusions remain explicit.

The checker is byte-identical to WCM #249 at a5e523f54a8ac08e8c7adb8503de2a5c37a5f43f (blob 0b0961ce302e840daee13dc218695a463a4a1393). Rajnish and drk-ram have approved that source head. Each port still needs independent repository adoption/review; this is not an organization-wide policy mandate.

Validation — 3 November 2026

Head: 0359afe26fecae824004b596613af0295f247892.

• 20 guard tests pass normally and 20 under optimized Python.
• The actual changed-lock driver passes against current main; contribution whitespace checks pass.
• Current main is merged with both histories and DCO sign-off preserved. The published tree exactly matches the locally checked tree; checker/driver/tests are unchanged.
• Focused tooling validation only; no new full runtime-suite run is claimed. All returned current-head technical checks pass; the maintainer gate remains failed.

Review and adoption

@carloshvp: current-head re-review and the repository-specific adoption disposition remain requested. Earlier approvals were for superseded port revisions. A new formal reviewer request was attempted on 3 November but the connector returned HTTP 403; no successful reassignment is claimed. Draft retained until that disposition and normal merge gates are satisfied.

Prepared with OpenAI Codex; DCO-signed. No merge, release or change to repository protections.

[github-actions]

🟡 Contributor Check: MEDIUM

Check	Result
Profile	MEDIUM
Credential	LOW
Overall	MEDIUM

Automated check by AgenTrust Contributor Check.

[dinakarjs]

The new Generated lock delta job and hosted workflow lint pass. Ordinary CI stops at inherited certifi/filelock/setuptools audit findings (job 121471802182), covered by existing #795/#822. This draft does not alter those dependencies or bypass the audit; their normal review/merge remains separate.

[carloshvp]

Reviewed exact head 2e8fbb612cf5861f9a1d5869de772662743c86ec in an isolated checkout. Requesting one correction to the generated-lock driver before adoption.

The five committed unittest methods pass. An independent temporary-Git regression renames requirements/dev.txt to requirements/renamed.txt while dropping one of ten exact, hash-pinned dependencies. Git reports R090; the actual driver exits 0 and prints NEW requirements/renamed.txt: structure only; no prior dependency baseline. This reproduces both normally and under python -O.

git diff --name-only applies rename detection and emits only the destination. The old lock never reaches the deleted-lock guard, and the new pathname bypasses comparison with its predecessor, so both deletion and dependency loss escape the advertised review requirement. Please enumerate additions/deletions with rename detection disabled, or explicitly handle rename source/destination pairs and compare their dependency identities. Add a real-Git rename-plus-removal regression alongside the existing deletion and upgrade controls.

This is a defect in the proposed CI guard, not evidence of an installed dependency or production security bypass. The source WCM checker review and this draft's repository-specific acceptance remain separate.

Reviewed with OpenAI Codex.

[dinakarjs]

@carloshvp addressed the rename finding at 50edc08. The driver now passes --no-renames to git diff so the deleted source lock reaches the existing explicit-review guard, even when Git would classify the change as R090/R100. A real temporary-Git regression verifies rename classification and checks both unchanged renames and rename-plus-dependency-removal, normally and under python -O. All four controls fail on the old driver; all six unittest methods pass with the repair, including existing upgrade/deletion controls. Syntax/pyflakes lint and whitespace pass. The documentation makes the conservative rename policy explicit. No dependency pins or runtime behavior changed. Please re-review; draft and independent acceptance gates remain. Prepared with OpenAI Codex.

[carloshvp]

Reviewed current head 50edc08e4917bab141531d06164eaab7b68d0c58.

The --no-renames repair closes the previously reproduced rename/dependency-removal bypass. In an isolated checkout, all six unittest methods pass normally and under python -O; removing only --no-renames makes all four rename controls fail. DCO trailers are present where required.

One remaining CI gap needs fixing: the workflow selects only test_lock_delta.py, so the new test_lock_renames.py regression never runs in the advertised guard job. Its path is also absent from the trigger filter. Use a discovery pattern and path filter that include both test modules, and run that workflow command normally and under -O. Keep the PR draft until its separate acceptance requirements are satisfied. Existing repository checks and independent acceptance remain separate.

Prepared with OpenAI Codex.

[dinakarjs]

@carloshvp addressed the CI discovery gap at 96f164a. The workflow and path filter both use test_lock_*.py, which includes the delta and rename modules; its regression step runs in normal and optimized (-O) modes. Both exact commands pass six unittest methods locally on Linux/Python 3.12.14, and whitespace checks pass. The published tree is the tested parent tree plus only this workflow edit. Draft retained pending independent source-checker/port acceptance and normal repository checks. Please re-review the updated head. Prepared with OpenAI Codex; DCO-signed.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[dinakarjs]

Hosted validation follow-up at 96f164a: Generated lock delta and all returned applicable technical workflows now succeed. Primary run: https://github.com/agentrust-io/cmcp/actions/runs/40918823720. The CI discovery repair is now verified by the hosted guard job as well as the local normal/-O controls. Draft status and independent source-checker/port acceptance remain unchanged. No merge or publishing workflow was performed. Prepared with OpenAI Codex.

[dinakarjs]

Closure follow-up to @carloshvp: the four current draft port heads still contain the published CI-discovery repair (cmcp #824 96f164a, agent-manifest #557 0ae45b05, ca2a #294 45edfb0c, integrations #394 40d52e23), and their hosted follow-ups already verified both regression modules. Please disposition the remaining current-head review requests/threads when you can. No new duplicate implementation or automatic adoption is proposed; WCM #249 remains the separate source-checker acceptance decision. I cannot dismiss another reviewer's changes request on their behalf. Prepared with OpenAI Codex.

[carloshvp]

Reviewed current head 96f164a3b48f155d840a0589839b54caa6d92bbe.

The CI discovery and trigger filters now include both test_lock_delta.py and test_lock_renames.py, and the workflow runs the same discovery command normally and under -O. I independently ran both commands: all six unittest methods pass in each mode, including the real-Git rename/dependency-loss controls. The four ports contain identical checker/driver code; sign-offs and diff whitespace checks pass.

This resolves my previous CI-discovery change request. Approving this bounded port implementation; keep the draft status until the separate source-checker/adoption decision and repository acceptance requirements are satisfied. This review does not adopt a cross-project policy or approve WCM #249 on its behalf.

Reviewed with OpenAI Codex.

[dinakarjs]

@carloshvp requesting re-review of the subsequent hash/inventory corrections at 95171ef90a4f6a74399d2a8f66d8352a4178e1f2. Your prior CI-discovery review applies to an older revision; this changed head still needs independent disposition.

I verified the current Git trees: tools/check_lock_delta.py is identical to WCM #249 source head 9fcf6bd6684dbd87473d1b0c670eaeb154749a75 (blob 4c702425da6ed1c554bcdcbd78245b2d9e9ee62d). The four ports also have identical driver, workflow and four regression-module blobs. Their repository exclusions remain separate.

The current-head guard job logs confirm that test_lock_*.py discovers 18 tests and passes in both normal and optimized Python, followed by successful execution of check_changed_locks.py against the PR base. This verifies the hosted discovery/execution path, not just a workflow-level green result. All other returned current-head technical workflows also succeed; the remaining returned failure is the independent maintainer/policy gate.

WCM #249 source acceptance has been requested explicitly from Imran/Rajnish and remains pending. Please review this port's updated implementation and identify remaining changes; keep draft/adoption holds until source acceptance and this repository's requirements are met. No new local suite run, source acceptance or merge is claimed by this status update. Prepared with OpenAI Codex.

[dinakarjs]

@carloshvp requesting bounded re-review of current head 146b5a0dbfefdec9341f0c1056aa805882238ada.

The subsequent WCM #249 repair is carried here: quoted marker whitespace is preserved, lexical baselines are retained, and symlink lock paths are refused by the checker and changed-lock driver. Recorded local guard discovery passes 20 tests normally and under optimized Python. All returned current-head technical workflows pass.

Please review this updated port and identify any remaining correction. Draft status is retained pending independent acceptance of WCM #249 at a5e523f54a8ac08e8c7adb8503de2a5c37a5f43f and this repository's own acceptance. This request does not adopt a cross-project policy or imply that prior-head approval covers the new source. Validation details and scope are in the description; no new local rerun is claimed by this follow-up. Prepared with OpenAI Codex.

[carloshvp]

Reviewed commit 0359afe26fecae824004b596613af0295f247892.

The current 20 regression tests pass normally and under -O, and the four ports carry identical checker/driver code. I found one remaining symlink bypass in the changed-lock driver.

In a real Git fixture, replace a tracked requirements/nested/dev.txt with a symlink requirements/nested -> ../elsewhere, with elsewhere/dev.txt containing the same pins. Git reports the original lock as deleted, but the driver returns exit 0 and PASS requirements/nested/dev.txt: changed packages []. It follows the intermediate symlink because it checks only the final path and the top-level requirements directory.

Check every lexical path component before reading, as check_lock_delta.py already does, or reject a deleted Git-tree entry before consulting the filesystem. Add a real-Git intermediate-directory symlink case with an ordinary-directory valid control in both normal and optimized modes. Keep the separate source-adoption/draft hold until the repaired source and this port are accepted.

[carloshvp]

[P2] Reject symlinked intermediate directories before reading a lock. Replacing requirements/nested/ with ../elsewhere makes a deleted tracked nested/dev.txt pass as an unchanged lock. path.is_symlink() and the top-level directory check miss that parent. Walk all lexical components (as the standalone checker does), and cover this deletion substitution with a real-Git regression.