Repository navigation
Move core pin to tendcf-integration 4443504; adversary I4-I7 - #61
Conversation
…t seam under TEST_UNIT_TEST; .gitmodules names the libntech fork); re-cite cmdb.c lines at the new pin Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…_start on any loader), I5 (validate POLICY), I6 (positive probe control for the empty-view cases), I7 (wording, and the {vars:{real},vars:{}} case named)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: 74cc0e48-a851-4a23-aa4e-ca19f794b8fa) |
Codex Review Summary: ✅ CompletedThis 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. |
…valuated the policy (PR 61 review: Astra r1, adversary r1) entry.cf now runs tests/policy/fixtures/test-done.cf after tendcf_interlocks and tendcf_services. strict_case (all four cases, any loader) and the two empty-view cases require its report in the same output file as the negative greps; the separate probe is kept and its comments reworded to what it proves (what the data loaded as). Non-fork strict cases also assert the line a lenient loader's data makes the policy print (derived, not run). No new output files, so the failure dump is unchanged. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: 07741cb2-1c76-4c6d-ace7-86286b8a2c97) |
What changed
Moves tendcf's core fork pin (
overlay/fork.json) from6505f488a80ab3e18b089c7d8e32571c88128866to4443504fdeb7ad0be247b26ee8ebdf8ee2ee5a28, the tip of djbclark/coretendcf-integration(tagtendcf-pin-4443504). libntech stays atcc892efae2597ea99749880c5eed01544018a4e6(tagtendcf-pin-cc892ef).Delta 6505f488a..4443504 (verified with
gh api, 4 commits ahead):libpromises/cmdb.c,libpromises/cmdb.h,tests/unit/cmdb_test.c,.gitmodules. It is fork PR 99: the CMDB insertion-failure test seam moves underTEST_UNIT_TEST(no writable exported pointer in the production library), and core's.gitmodulesnames https://github.com/djbclark/libntech. No loader behaviour changes (the cmdb.c diff is the seam only), so the run.sh expectations are unchanged.Core CI gate on the pin: https://github.com/djbclark/core/actions/runs/43648410082 (unit 81 PASS / 0 FAIL, cmdb_test PASS, 06_host_specific_data 18/18, def.json 26 + 1 known soft CFE-2956, libntech cc892ef).
Every place tendcf recorded the old pin or cited cmdb.c lines at it is updated (
overlay/fork.json,overlay/overlay.jsonnote, register B-5b note, proposal section 13, policy comment, run.sh comments). The cmdb.c line citations in run.sh moved by +14 (ReadJsonFileMode 73-96, the strict call at 1185-1191, CheckCMDBClassName 790, the class-exists skip 1109-1116), re-read from the file at 4443504.Operator ruling B11 round 2, item 4 (.gitmodules: bundle with next pin move)
Satisfied on the core side by 8d8dc94 being in 4443504; this PR moves tendcf's pin onto it. tendcf carries no submodule url override or
git config submodule.libntech.urlworkaround:.github/actions/build-forknever reads.gitmodules, it fetches libntech at core's gitlink fromlibntech.repoinoverlay/fork.json. So nothing is redundant and nothing is removed; only the action's comment is updated to say.gitmodulesnow names the fork too and the action does not depend on it.Adversary r4-delta Info items
I4: done. A new plain
checkno_render_start(noCreated file, no launchctl, no systemctl) runs for every strict_case, so dup-member, dup-vars and trailing still assert a safety property on a non-fork run; the full refusal assertions stayfork_only. Mac behaviour is derived from the parser behaviour, not run.I5: done.
strict_casevalidates POLICY (any|fork) and exits 2 on anything else.I6: done. Each empty-view case now runs the probe policy over the same file and asserts both "absent" lines on any loader, so the negative greps cannot pass because cf-agent did not evaluate policy.
I7: done. The policy comment now says "if its lenient view lists any
varskey"; both the policy comment and proposal section 13 name{"vars":{<the real configuration>},"vars":{}}as the operationally relevant member of the empty-view class and say the examples are not the whole class.Not run
No local build or run.sh (Linux CI only). The non-fork (Mac) outcomes are derived, not run.
🤖 Generated with Claude Code
Fix round 1 (head f32cc46)
Review finding (both reviewers): the positive controls did not prove that the invocation under assertion evaluated the tendcf policy. The empty-view probe is a second
cf-agentprocess runningcmdb-probe.cfonly, and on a non-fork loaderno_render_startwas the only assertion left on a strict case.Fix:
entry.cfnow also runs a test-only bundle,tendcf_test_done(tests/policy/fixtures/test-done.cf, not a shippedpolicy/file), aftertendcf_interlocksandtendcf_services. The four strict cases (any loader) and both empty-view cases require its report in the same output file as their negative assertions.cf-agentgoes on to evaluate policy after the strict loader refuses a file (generic_agent.c:1234at 4443504 only logsFailed to load CMDB dataat ERR), so the marker applies on the fork too. Non-fork strict cases also assert the line a lenient loader's data makes the policy print (the failing interlock's refusal for a repeated member and trailing input, the runit flavour refusal for a repeatedvars): derived from the policy source, not run on a lenient build. The separate probe stays, with its comments and check names reworded to what it proves (what the data loaded as). No new output files, so the failure dump is unchanged.CI at the new head: Linux https://github.com/frdminc/tendcf/actions/runs/43651117422 (4 legs success; 24 marker checks ok = 6 cases x 4 legs; 0 SKIP, 0 FAIL,
all policy tests passedx4), Check https://github.com/frdminc/tendcf/actions/runs/43651117455 (success).Note
Low Risk
Pin move documents a test-seam commit without loader behavior changes; risk is mainly CI/test semantics and build reproducibility at the new revision.
Overview
Bumps the pinned CFEngine core fork in
overlay/fork.jsonfrom6505f488ato4443504(djbclark/coretendcf-integration). That delta is fork PR 99: CMDB unit-test seam underTEST_UNIT_TESTand.gitmodulesnaming the libntech fork—no CMDB loader behavior change.Documentation and inline references (
overlay/overlay.json, architecture docs,policy/tendcf_services.cf,run.shcomments,build-forkaction comment) are updated to the new pin and revisedcmdb.cline citations.Policy integration tests gain a test-only
tendcf_test_donemarker bundle sostrict_caseand empty-view cases require proof the samecf-agentrun finished evaluating tendcf bundles. New assertions includeno_render_start,lenient_onlyexpectations on non-fork loaders,POLICYvalidation instrict_case, and probe checks on empty-view fixtures.Reviewed by Cursor Bugbot for commit f32cc46. Bugbot is set up for automated code reviews on this repo. Configure here.
Reviews (prose, at head f32cc46)
Cross-model frontier review, gpt-6-astra (high): round 1 at c4f8d7b requested one change (the positive controls did not prove the asserted cf-agent invocation evaluated the policy); round 2 at f32cc46 approves. It found the round-1 finding closed, the generic_agent.c:1234 reasoning correct at core 4443504, the lenient-loader strings matching the fixtures and policy reports, and CI at this head green with 24 completion-marker checks, 0 SKIP and 0 FAIL. One comment-only Low remains (run.sh:957-959 still calls the negative greps "the only assertions left" on a non-fork run); it is carried to the next tendcf test PR.
Security review (adversary sub-agent): round 1 at c4f8d7b found one Low (the same positive-control gap) and confirmed the pin is the tip of djbclark/core tendcf-integration, the build fetches only the pinned 40-character commit and libntech by core's gitlink, and the TEST_UNIT_TEST seam cannot reach a release build. The round-2 delta at f32cc46 found nothing: the marker closes the gap on fork and non-fork loaders, the test-only fixture cannot reach shipped policy or deploy paths, and there is no quoting issue. Informational notes carried forward: anchor the marker grep to
R: tendcf-test-done:, and the pre-existingbash -c "! grep -q ... '$out'"quoting.