4 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
Hermes AgentandClaude Opus 5	845e3143ab	ci: enumerate the issue-close regression on the sanitization surface ... check-test-enumeration.sh failed the previous push: the new test existed on disk but was on no CI surface and not signed in the exclusions file. That guard is correct and caught exactly what it exists to catch -- a test that would never have run. Registered on the sanitization step rather than the exclusions file, because this test is hermetic: it mocks tea and curl onto PATH and sandboxes a throwaway git repo, so it resolves no real credentials. The tools/git tests in the exclusions file are there precisely because they do. Guard now: population 51, enumerated 32, excluded 19, all surfaces present. Refs #1081 Co-Authored-By: Claude Opus 5 <[email protected]> Claude-Session: https://claude.ai/code/session_01Amf1Neca162odgcbCWMk1y	2026-08-06 17:40:07 -05:00
Hermes AgentandClaude Opus 5	51874794c2	fix(tools/git): use tea comment, keep one principal, add a red-first regression ... Addresses both blockers from review 127 (rev-security-02) and corrects the severity claim in the original report. Blocker 1 -- mixed principals. Routing the comment through the token- authenticated gitea_issue_comment_api() attributed the comment to the token holder while the close still used --login $GITEA_LOGIN_NAME: two principals for one operation. `tea comment` accepts the same --repo/--login flags, so the tea branch now uses it and both calls carry the same principal. The no-login branch keeps the API helper for both, also a single principal. Blocker 2 -- no regression test. Adds test-issue-close-fail-closed.sh on the existing mocked-tea/sandboxed-git harness pattern. Asserts: a failed comment does not close the issue and exits non-zero; a successful comment does close it; the subcommand is top-level `tea comment`, never `tea issue comment`; and the comment and close carry the same --login. GREEN on this branch, RED on main. Severity correction. The original report said the issue closes anyway and the audit trail is silently lost. It does not: set -e at line 5 aborts the script when the comment fails, so the close is never reached. The real defect is that issue-close.sh -c cannot succeed at all where a tea login resolves -- loud, not silent. The explicit || guard is retained deliberately: a fail-closed property that depends on set -e disappears the moment anyone adds `|| true` or wraps the call in a conditional. Posted as a comment on #1081 and #1085. Refs #1081 Co-Authored-By: Claude Opus 5 <[email protected]> Claude-Session: https://claude.ai/code/session_01Amf1Neca162odgcbCWMk1y	2026-08-06 17:28:31 -05:00
Hermes AgentandClaude Opus 5	23483a773d	fix(tools/git): issue-close.sh silently dropped the closing comment ... `tea issue comment` is not a subcommand -- tea 0.11.x exposes only list/create/edit/reopen/close under `tea issue`, and comments are the top-level `tea comment`. The call therefore always failed. Its result was never checked, so the script went on to `tea issue close`, which IS valid: the issue closed and the record of why it closed was silently lost. Route the comment through the existing gitea_issue_comment_api() helper on both branches. It is login-independent and was already the mechanism used by the no-login fallback. Both call sites now fail closed: if the comment cannot be posted, the issue is not closed. Verified behaviourally against the live provider, both directions: negative -- comment cannot post => "NOT closing (fail closed)", exit 1, issue left open positive -- comment posts and issue closes => state=closed, comments=1, exit 0 Refs #1081 Co-Authored-By: Claude Opus 5 <[email protected]> Claude-Session: https://claude.ai/code/session_01Amf1Neca162odgcbCWMk1y	2026-08-06 15:35:12 -05:00
Hermes Agent	880c28b191	docs(glpi-skills): genericize operator-specific content per review	2026-07-20 19:45:50 -05:00