Skip to content

Pin the fork at the fail-closed CMDB loader (R13) - #41

Merged
djbclark merged 15 commits into
masterfrom
fork-pin-cmdb-fail-closed
Nov 14, 2026
Merged

djbclark merged 15 commits into
masterfrom
fork-pin-cmdb-fail-closed

Conversation

@djbclark

@djbclark djbclark commented Nov 3, 2026 •

Copy link
Copy Markdown
Collaborator

Operator ruling R13 (Q49). The fork's CMDB loader used to skip a bad entry and install the rest, so an interlock entry dropped at load left the services it guards actuated unguarded (docs/architecture/tendcf-validate-proposal-2026-11-03.md §13).

E. Fork pin and what it is tested against

This PR moves the pin in overlay/fork.json from b8abf28408b1b54e285d372a9bc5f2d850c960ad (the per-entry-skip fork, B-5b) to 6505f488a80ab3e18b089c7d8e32571c88128866, the tip of djbclark/core tendcf-integration. libntech is taken from core's gitlink at that commit (cc892efae2597ea99749880c5eed01544018a4e6) and fetched from the djbclark/libntech fork; the build step logs both shas.

What the new pin carries. R13 (operator ruling 2026-11-02): the fail-closed CMDB loader (a750b1dfe, dbf9dede6, 61065e551, 736becc07; merged d32cade13), which validates every entry first and installs nothing if any is invalid. Follow-up hardening, d32cade13 -> 6505f488a (6 commits; fork PR djbclark/core#98 and libntech fork PR djbclark/libntech#11): class insertion fails closed with rollback, the loader parses strictly (a repeated object member and non-whitespace after the document refuse the file), the libntech gitlink, and unit plus acceptance tests (06_host_specific_data 17 and 18; cmdb_test). Core CI on the pin: https://github.com/djbclark/core/actions/runs/43642211015 (green; unit 81/81, cmdb_test 24/24, 06_host_specific_data 18/18); libntech CI: https://github.com/djbclark/libntech/actions/runs/43640729887.

Limit of the strict parse. It is the CMDB loader's alone: readjson and def.json stay lenient, so PR 42's cross-check reads a lossy view of a file with a repeated member, a repeated vars or trailing input, and its independence from the loader is limited. The check reports a refusal when its lenient view still lists variables the loader did not install (or another feature it refuses on its own). When the lenient view is empty ({"vars":{},"vars":{}}, or {"vars":{}} plus non-whitespace after it) nothing installs and nothing actuates, but no refusal is reported: no actuation is not a refusal report, and the loader's error is the only trace. Docs and policy comments say so (proposal §13, register B-5b, policy header).

Tests (tests/policy/run.sh). Check 4b's malformed cases assert the policy refusal (reported once, nothing actuated). Block 9b has four rejection cases, one fixture each: a repeated member of vars, a repeated top-level vars whose later section drops the interlock, trailing non-whitespace after a valid object, and a whitespace-only default-namespace class name; on the fork each asserts the loader's reason and that nothing is installed (the first three also assert the cross-check's refusal and no actuation, as loader+policy integration assertions; class-space's refusal comes from the classes section on any loader). The existing-class case is a control: the variables load, the existing class is preserved, and the policy refuses the extra classes section. Two empty-view controls ({"vars":{},"vars":{}}, {"vars":{}} plus trailing input) assert no actuation and no cross-check report on any loader, and the loader's refusal on the fork.

Non-fork behaviour. Assertions that hold only with the strict loader (its log strings and the 9b integration assertions) go through one capability probe and print SKIP (loader not all-or-nothing): <check> elsewhere; a non-fork run is expected to pass with SKIP lines (derived, not run on the Mac). The Linux legs set TENDCF_REQUIRE_FORK_LOADER=1, so any SKIP there fails. AGENTS.md says so.

Rulings of 2026-11-14. (a) "Keep Mac, SKIPs": the Mac dev build (non-fork, lenient loader) stays a run.sh target with explicit SKIPs; Linux CI fails on any SKIP. (b) B-5b "Replace": the overlay will later carry the all-or-nothing loader instead of the per-entry skip; that is a separate later PR, and the overlay patches and manifest entry are unchanged here (the manifest note only records the ruling).

CI at beddad9: Linux https://github.com/frdminc/tendcf/actions/runs/43645480705 (4 legs, success; libntech cc892efae in every log; 0 SKIP lines; every case ok); Check https://github.com/frdminc/tendcf/actions/runs/43645480646 (success).

Not run: the SKIP path and the Mac dev build; the failure dump; no local run.sh or core build (Linux CI only).

Reviews

Cross-model review: the change was written by Sonnet 5.5 under an Opus 5.5 lead. Astra is gpt-6-astra at high effort; the adversary is a Claude security reviewer.

  1. Astra r2, at d90c78a: APPROVE, with 3 Low findings: B3, run.sh refusal asserts that a lenient loader cannot pass; C1, the refusal claim needed qualifying for an empty lenient view; C2, the body draft's pin history. All three are fixed in round 4.

  2. Adversary r2, at d90c78a: APPROVE, with 0 blocking and 3 Info findings. I1, the empty lenient view, is fixed in round 4 by wording and two controls. I2 and I3 are out of scope; I3, core .gitmodules, goes with the next pin move per the 2026-11-14 ruling.

  3. Astra round-4 delta, d90c78a..beddad9: APPROVE. B3, C1, C2 and I1 are fixed, with no new findings that need a change. On CI every assert moved to fork_only still runs. Astra read the Linux log at beddad9: 0 SKIP lines, and each moved assertion is ok on all four legs.

  4. Adversary round-4 delta, d90c78a..beddad9: APPROVE, with 0 blocking, 0 Low and 4 Info findings. Quoted verdict: "On CI the assertions moved into fork_only still run as plain checks and none can skip silently: TENDCF_REQUIRE_FORK_LOADER=1 is set on both Linux run.sh steps, and a mistyped POLICY value falls back to the stricter check. The two new empty-file tests hold on both kinds of loader and cannot hide a partial install. I1 is resolved." The Info findings are optional hardening for a follow-up PR:

    • I4: a narrower no-render/no-start assert would keep Mac coverage for dup-member, dup-vars and trailing.
    • I5: validate POLICY.
    • I6: a positive policy-ran control for the empty-view cases.
    • I7: two wording tightenings in policy/tendcf_services.cf.

🤖 Generated with Claude Code


Note

High Risk
Moves the production Linux CI pin to fail-closed CMDB loading behavior and tightens tests so policy enforcement depends on that loader; a wrong pin or skip path could hide the interlock fail-open class of bugs.

Overview
Bumps overlay/fork.json from the per-entry-skip build (b8abf284…) to 6505f488a, which loads host_specific.json all-or-nothing with strict JSON parse (operator ruling R13 / PR #41)—invalid or duplicated entries install nothing instead of leaving services unguarded when an interlock is dropped.

Policy CI gains a one-shot fork capability probe (check 4b), a fork_only / SKIP path for Mac and other non-fork builds, and TENDCF_REQUIRE_FORK_LOADER=1 on Linux workflow legs so CI fails if the loader is not the pinned strict variant. New checks 4b and 9b and loader fixtures exercise malformed CMDB, duplicate members, trailing junk, and empty lenient-read views against the cross-check and probe policy.

Docs and overlay metadata record that overlay B-5b still describes the old per-entry skip but is superseded_on_fork until a follow-up replaces the patch; comments in policy/tendcf_services.cf, projector.py, and architecture docs are aligned with the new pin and loader vs readjson limits.

.github/actions/build-fork logs resolved core and libntech commit SHAs in the job output for debugging submodule fetch.

Reviewed by Cursor Bugbot for commit beddad9. Bugbot is set up for automated code reviews on this repo. Configure here.

@cursor

cursor Bot commented Nov 3, 2026

Copy link
Copy Markdown

Bugbot couldn't run - usage limit reached

Bugbot 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: 8e5e9d8d-fe9f-4e87-939a-2c5037375964)

@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Nov 3, 2026 •

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-11-03T04:18:32.174445Z fe9bcc1 PR opened
🔒 Security Review ✅ Completed 2026-11-03T04:19:34.282721Z fe9bcc1 PR opened
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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@cursor

cursor Bot commented Nov 3, 2026

Copy link
Copy Markdown

Bugbot couldn't run - usage limit reached

Bugbot 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: 723f39eb-ec9e-4256-b800-68b83d54a548)

@cursor cursor Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

Fix All in Cursor

❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.

Reviewed by Cursor Bugbot for commit 83ddf57. Configure here.

Comment thread tests/policy/run.sh Resolved
djbclark and others added 8 commits November 14, 2026 14:45
…ck actuates no service

overlay/fork.json: core 0ea30a48f (tendcf-integration with cmdb-fail-closed
merged): LoadCMDBData() installs nothing from host_specific.json if any
entry is invalid (operator ruling R13). tests/policy/run.sh check 4b: a
file with one malformed interlock beside a present service processes no
service. upstream-register.md: note that B-5b is reversed on the fork.
…an index-key case

The fork now also rejects index-bracket keys, frame-dependent scopes, null
sections, non-string rule elements, duplicate variables and unknown
top-level keys. run.sh 4b runs a probe policy (fixtures/cmdb-probe.cf) that
asks isvariable() whether data:variables.tendcf_interlock and
...tendcf_service exist after each malformed load, with a loading file as
positive control, and a second fixture whose key 'tendcf_interlock[' the
variable parser reads as the interlock container.
….sh, sim-loader.cf, policy comment)

The pinned fork (d32cade13) installs nothing from a file with any invalid entry.
run.sh check 9's NB and check 10's header said the fork skips one entry; the
policy's CMDB cross-check comment and the sim-loader.cf header said the same.
State what each loader does, and why the simulated loader stays valid. Check 9's
malformed case now also asserts the valid sibling tendcf_service is not installed.
The 'did not load' report no longer says the loader 'skipped' the key.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ption 5); date the pre-#41 loader statements

Section 13 now says both options were ruled, each independent of the other,
without repeating the 'Implemented 2026-11-06' paragraph. Fact 2, the checked-
commit sentence, the expand.c citation and finding (1) described the pin of the
time as the pinned fork; they now name b8abf28408b1 as the earlier pin and say
the Linux legs no longer see a per-entry skip.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…t (manifest, register, overlay doc)

The B-5b entry still reproduces 8f0076b81, which the pinned fork reversed (R13).
Add a superseded_on_fork note to overlay.json, replace the register's 'see the PR'
pointer with the actual statement, and add a section 7 note. Whether to keep,
retire or replace the patch is a separate ruling because it changes what the
overlay reproduces.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…n moved to d32cade13, which differs only in cmdb.c/cmdb.h and tests

Checked with the GitHub compare API on djbclark/core (8 commits, 10 files).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@djbclark
djbclark force-pushed the fork-pin-cmdb-fail-closed branch from 072ceef to 867591e Compare November 14, 2026 19:49
@cursor

cursor Bot commented Nov 14, 2026

Copy link
Copy Markdown

Bugbot couldn't run - usage limit reached

Bugbot 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: ec57d00e-f3df-4483-82b3-8a7385b5bd43)

…ses the whole file), not a content difference

The pinned fork rejects a file in which 'variables' names a variable 'vars' already
names (CMDBPendingCheckNoDuplicates, 61065e551) and installs none of it, so the
cross-check sees the key missing, not a content difference (Linux CI 867591e, all
4 legs, one FAIL). Assert the loader's duplicate refusal, 'installing none', the
interlock reported not installed and no content-difference report; keep section
named, refusal once, no actuation. Same-id content differences stay tested in
check 10 (stand-in loader) and check 11. Fix the stale 'loader applies variables
after vars' text in run.sh, the policy comment, the proposal and projector.py.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@cursor

cursor Bot commented Nov 14, 2026

Copy link
Copy Markdown

Bugbot couldn't run - usage limit reached

Bugbot 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: dd6ade02-8bd4-4536-994a-da405774d3ea)

djbclark and others added 4 commits November 14, 2026 16:28
…losed); log core and libntech shas in the build; CI requires the fork loader

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…requires the fork loader); policy-level refusal in 4b; strict-parse and class cases 9b; fuller failure dump

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ed fork

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…he 6505f488a source; strict parse is the CMDB loader's alone

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@cursor

cursor Bot commented Nov 14, 2026

Copy link
Copy Markdown

Bugbot couldn't run - usage limit reached

Bugbot 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: fed77210-8a44-4530-a071-f964ae412124)

djbclark and others added 2 commits November 14, 2026 16:56
…, repeated vars, trailing input) as fork_only; empty-view controls; header and AGENTS.md say which assertions are integration and what a non-fork run does

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…on, no report); non-whitespace after the document; record the 2026-11-14 Replace ruling for B-5b

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@cursor

cursor Bot commented Nov 14, 2026

Copy link
Copy Markdown

Bugbot couldn't run - usage limit reached

Bugbot 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: d76b8a3b-bd2e-4ecd-bfad-20bb313a72db)

@djbclark
djbclark merged commit daa4530 into master Nov 14, 2026
8 checks passed
djbclark added a commit that referenced this pull request Nov 14, 2026
Corrects the rulings record against the three reviews of PR 62:
schema facts (v6 and contract_version const 2 are implemented by PR 52;
later changes take the next unused version, grouping not ruled), stale
merge/pending statuses (PR 52, #42, #41 merged; WarpBuild run
succeeded), the spike lane's rails restored word for word with the
workflow-permission hardening marked as hardening, the light review tier
read conservatively (no waiver of routing or admission rules), the Linux
design-work freeze removed (later in the order, not paused), option-number
notes, and open items marked 'open: awaiting operator ruling' (Android
log/index ownership, root enforcement timing, reviewer eligibility,
schema grouping) with a new open-items list in 14. The rulings file is
now committed beside 14 as rulings-record-2026-11-14.md and local paths
are stripped from public docs. No new decisions.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
djbclark added a commit that referenced this pull request Nov 15, 2026
Corrects the rulings record against the three reviews of PR 62:
schema facts (v6 and contract_version const 2 are implemented by PR 52;
later changes take the next unused version, grouping not ruled), stale
merge/pending statuses (PR 52, #42, #41 merged; WarpBuild run
succeeded), the spike lane's rails restored word for word with the
workflow-permission hardening marked as hardening, the light review tier
read conservatively (no waiver of routing or admission rules), the Linux
design-work freeze removed (later in the order, not paused), option-number
notes, and open items marked 'open: awaiting operator ruling' (Android
log/index ownership, root enforcement timing, reviewer eligibility,
schema grouping) with a new open-items list in 14. The rulings file is
now committed beside 14 as rulings-record-2026-11-14.md and local paths
are stripped from public docs. No new decisions.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant