Repository navigation
docs: record the 2026-11-13/14 goal-review rulings - #62
Conversation
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: 30d181bd-5630-4cb9-8632-e8f9f4c48f53) |
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. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5fd6ec39b4
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| - Design rulings (a `schema_version` bump, the privilege or enforcement | ||
| scope on a host, a deploy plan, the unit-writer or consent-diff design) | ||
| get two independent frontier reviews, astra and Fable. Argon stands in | ||
| scope on a host, a deploy plan, the unit-writer or consent-diff design), | ||
| and root, trust-root, key and handshake code (O2, above), get two | ||
| independent frontier reviews, astra and Fable. Argon stands in |
There was a problem hiding this comment.
Apply O2's narrowed definition of a design ruling
For deploy-plan, unit-writer, or consent-diff documentation changes that do not change a schema or trust, this retains the old two-frontier-review requirement even though the adopted O2 explicitly narrows “design ruling” to schema bumps and trust changes (docs/review-2026-11-13/13-premortem-and-ideas.md:192-196). Consequently, these changes still require both frontier pools and can remain blocked on quota, rather than receiving the lighter review the ruling selected. Narrow this definition to match O2 while retaining the separate two-review requirement for root, trust-root, key, and handshake code.
AGENTS.md reference: AGENTS.md:L68-L69
Useful? React with 👍 / 👎.
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: 77481b4a-36c1-46d4-8d4d-9212f3eb1824) |
…em format New files only; no original document is changed. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Thirteen response and research documents (01-13) measured against the goal agreed 2026-11-13, the six independent check files that tried to refute them, and a synthesis listing the rulings needed, confirmed defects and a merge map. No original document is changed. Reviewers and checkers are all Claude Opus; this is not the cross-model frontier review, and nothing was built or executed. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
A later pass re-fetched 35 source groups the first review could not read: 15 confirmed, 18 weakened, 1 refuted (the paper's Promise Theory purpose claim), 1 still unfetched. Corrections R1-R21 are applied to the response documents and the synthesis; two items are added to the sources check (P-§3.1-s28, P-§3.2-s29). No original document is changed. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Writes every operator ruling of the 2026-11-14 goal-review walk (B1-B13, riders, sub-rulings, smaller rulings, PR 52 items, B11 round 2) into the normative docs: AGENTS.md rule 3 (spike lane, tiered review, docs freeze), the map and decision register (D47-D50 new; D6, D8, D16, D45, D46 amended), change-process and the upstream register, and the review proposals they settle. Also repairs the 21 dangling references xref_lint found in the review docs. Ruled items not yet built say so. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
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>
Records the four operator rulings given on the PR 62 review questions (root-enforcement timing: exception for the VM; light review: cheaper reviewer allowed; Android log/index: agent keeps log role; schema grouping: decide at design review), replacing the round-1 open notes, appends them to the in-repo rulings record (lines 57-60), and marks the PR 61 pin move merged (401e0f6). Fable B2, B4 and B5 stay open. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
9b3da02 to
ae841bc
Compare
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: 7fa8fd29-a472-43df-b431-f020eb98bced) |
Q5 "1. Yes, keep it" (the G4 sentence in the goal text is the ruled wording), Q6 "1. Keep both" (map section 0 item 3 and the section 8 log-archive paragraph stay), Q7 "1. Defer to design review" (the AI reviewer's up-to-30 stays open). Appended as lines 61-63 of the rulings record; 14's open-items list now holds Q7 only. 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: 2a475e1e-783f-4432-b39d-29dae9d3022d) |
… L64) Records the operator ruling on the round-2 review (L64) and corrects the round-2 review findings: AGENTS.md rule 3 restated (ruled O2 tiers and Q2 in their own cited sentences, the lead notes labelled lead practice, the not-light list widened, Opus and Sonnet one family), Q7 marked as deferred to the schema design review everywhere without choosing per reviewer or total, the Android log notes after Q3 closed in the map, one wording for the standing-approval reading, the Q1 exception worded narrowly, plugin-layout.md annotated with the L33 deferral, the rulings index and Lives-in pointers completed, PR 61 timestamp aligned. No new decisions. 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: bfa8cd75-bf56-44df-9570-49b8544bd515) |
Use one strong-form wording for the standing-approval reading in six places, labelled a reading of the ruling (L5, B2); scope the AGENTS.md rule 3 lead-practice sentence to readings that only tighten; keep only L64's own text under Ruled in 14; align the Q7 wording; drop the 'later ruling revives it' phrase in plugin-layout; add the D46 and plugin-layout 'Lives in' pointers for Smaller 5. No change to the rulings record. 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: 5f52d5d3-ba84-4427-9040-6c85165a7328) |
Reword the standing-approval reading in all six places: it keeps B1's other content checks and the rest of DC-2, its ceremony class is still checked, and signer and algorithm depend on X-trust-06 being adopted. Say the two readings are lead practice on the lead's authority, point the map's DC-2 sentence at the reading, correct the change-process pointers, and list the open points as a schema design review agenda item. No change to the rulings record. 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: e5c8c284-58d8-437f-9a54-fb5627f15bbc) |
Add an interim default to the standing-approval reading in all six places: until the operator rules, a standing approval covers ordinary changes only. Rewrite the schema design review agenda note in 14 so it lists the questions the review proposes to the operator and the checks that stay. No change to the rulings record. 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: 9b89fd31-2791-45a4-8463-e2b3f329d393) |
Summary
Writes every operator ruling of the 2026-11-14 goal-review walk (rulings file lines 4-56) into the repo's normative docs, each where a reader of that section would look, quoting the chosen option exactly and citing "operator ruling, 2026-11-14" plus the item id. Docs only: no code, no builds.
AGENTS.md: standing rule 3 now carries B10 (spike lane O1, tiered review O2, docs freeze O4; O6 not chosen); the B11 round 2 "Keep Mac, SKIPs" text, PR 52 gate 2 (WarpBuild), the PR 52 follow-up issues, the ShizukuTendCF dev-mode/dev-build rule and the AGPL-3.0-only review TUI.docs/architecture/architecture-DEFINITIVE-v3.md(the map): D47-D50 added as ruled rows; D6, D8, D16, D45, D46 and R5 amended; section 0 maxim; new section 9.12 (safety score); Android, macOS and build-order sections.change-process.md(Nix on the release path, build-time only; the person's-own-fork sentence),upstream-register.mdandupstream-overlay.md(B11 and round 2).docs/review-2026-11-13/: each ruled proposal section gets a "Ruled" paragraph and its proposed text rewritten (for example M-§8-01 for "make that both", M-§9-03 caps table, G-§7-01 floor, X-safety-06 with the arXiv citation); the goal definition gets the B13 edits; 14 §B has a ruling index.bin/xref_lint.pyfound in the review docs (it now reports 0 findings).Ruled but not yet implemented says so in the text (for example B-5b "1. Replace": "ruled; implementation pending in a follow-up PR", overlay phase B).
arXiv 2404.13076 was fetched before it was cited (Panickssery, Bowman, Feng, "LLM Evaluators Recognize and Favor Their Own Generations", 15 Apr 2024, https://arxiv.org/abs/2404.13076) and is cited only for self-preference bias.
Judgment calls to confirm
schema_version6 andcontract_versionconst 2, B3 sub 1 profile field on entries (rides the agent-id bump), B8 sub 1 score values fixed in the schema. This PR only records them.Coverage table (rulings file line -> where it now lives)
Zero unmapped lines (4-56); none "not applicable".
Fix round 1 (head 9b3da02)
Corrections from the three reviews of the first head; no new decisions, and every "open" item below is left for the operator.
contract_versionconst 2 (PR 52). C-policy-01 is recorded as implemented there; the profile/agent-id and standing-modes changes take "the next unused schema version (v7 or later)", grouping not ruled. The Stage 0 "one bump" allocation is gone.schema/*.jsonuntouched.permissions: contents: readand no secrets" is added and marked as a lead/reviewer hardening, not part of the ruling.cf-execdcounts as root code.docs/review-2026-11-13/rulings-record-2026-11-14.mdand local paths are stripped from public docs; X-safety-06's task and model scope restored; stale arXiv to-do removed; step-to-stage mapping sentence; standing approval marked as a reading of line 5 (only the nonce is dropped); dev build's distinct applicationId and signing key noted as a recommended follow-up, not ruled.Fix round 2 (head 9edc7da)
Rebased onto master 401e0f6 (PR 61 merged 2026-11-15: pin 4443504); no conflicts. Records the four operator rulings given on the review questions, quoted exactly, and appends them as lines 57-60 of
docs/review-2026-11-13/rulings-record-2026-11-14.md:Q1, root-enforcement timing: "2. Exception for the VM": an explicit, ruled exception like pastebin's; the VM may enforce early, canary only, operator-authored data (stage table stage 1 and D45 timing).
Q2, light review: "2. Cheaper reviewer allowed": any model outside the author's family not on the weaker-tier ban list may do a light review; if it is a frontier model the frontier checks still apply (AGENTS.md rule 3). The conservative limits from round 1 stay and are labelled as lead/reviewer readings.
Q3, Android log/index: "1. Agent keeps log role": "only" is about decisions; the thin Kotlin agent still writes the device JSONL and index as a support role; the map and D18 stand.
Q4, schema grouping: "1. Decide at design review": docs say "the next unused version (v7 or later)", grouping not ruled; the schema design review (two frontier reviews) decides.
Q5, G4 goal sentence: "1. Yes, keep it": the current
00-goal-definition.mdG4 sentence (wording 3 plus the highlighting clause) is the ruled wording.Q6, reviewer-proposed map text: "1. Keep both": map section 0 item 3 (M-§0-01) and the section 8 log-archive paragraph (M-§8-01) stay, in scope of recording rulings.
Q7, "AI reviewer subtracts up to 30": "1. Defer to design review": stays open; the schema design review decides.
Open items awaiting operator ruling (2026-11-14)
Still open (listed in
14-synthesis-and-rulings.md): only Q7, whether the 30-point maximum is per AI reviewer or the total for all AI reviewers; the schema design review decides.Fix round 3 (head cb84942)
Round-2 reviews: Astra (gpt-6-astra, high) changes-requested; Fable (claude-fable-5-2, high) changes-requested; adversary findings 1 medium, 4 low. This round fixes each once; no new decisions.
rulings-record-2026-11-14.md(now 64 lines) and recorded in14. AGENTS.md rule 3 now states the ruled O2 tiers and Q2 in their own cited sentences; the lead notes (light-docs limits, spikepermissions: contents: readand no secrets, one frontier review for compiler/CI/tooling) are labelled "in force as lead practice until djbclark rules otherwise"; the not-light list is widened per L64; Opus and Sonnet are stated as one family.plugin-layout.mdannotated with the Smaller 5 (L33) deferral of out-of-tree discovery; rulings index and "Lives in" pointers completed; PR 61 timestamp aligned; B8 option-number note.Fix round 4 (head de1b8a0)
Round-3 reviews: Fable (claude-fable-5-2, high) approve with 7 low/nit findings; Gemini 4 Argon (high) approve with 2 nit findings (Astra's codex pool is below the tendcf floor until 2026-11-18, so it did not review this round); adversary findings 1 low and 3 info. This round fixes each once; prose only, no new decisions, and the rulings record is unchanged.
14-synthesis-and-rulings.md(Fable A1, A3, A4, adversary C3, Argon 1, 2): "Ruled:" keeps only L64's own open list and the files follow as the lead's current examples; the Q7 text quotes 09 as it stands ("up to 30, only for findings citing a key path") with the aggregation still open; the section 9.12 pointer and the "Where" list agree; Smaller 5 "Lives in" adds the map D46,plugin-layout.mdandchange-process.md.plugin-layout.md(Fable A2, A7): drops "and a later ruling revives it" (L33 sets two conditions only); says PRs Pin the fork at the fail-closed CMDB loader (R13) #41 and policy: cross-check the loaded CMDB against host_specific.json; refuse to actuate on any difference (Q10) #42 are merged.Fix round 5 (head d96d13a)
Round-4 reviews: Fable (claude-fable-5-2, high) approve with 1 low and 2 nit findings; adversary no findings above info (3 info); the Argon r4 review is folded into this round. This round fixes each once; prose only, no new decisions, and the rulings record is unchanged.
14-synthesis-and-rulings.md(Fable A2, adversary info 2): thechange-process.mdpointers are "section 3 property 1, section 6 gate 2"; a schema design review agenda note lists the ceremony-class, persisted-rejects and X-trust-06 points as open.Fix round 6 (head 7392291)
Round-5 reviews: Fable (claude-fable-5-2, high) changes-requested (1 medium, 1 nit); Gemini 4 Argon (high) approve with 1 nit; adversary 2 low and 2 info. This round fixes each once; prose only, no new rulings, and the rulings record is unchanged.
ordinarychanges only; whether one may coverprivilegedorbaselinechanges is for the schema design review to propose and the operator to rule. Everything else in the reading is unchanged.14-synthesis-and-rulings.mdis rewritten (still a lead note, not a ruling). The review proposes to the operator: privileged or baseline coverage, how persisted rejects apply, how the single-use counter applies to a record that authorises more than one change, which signers a standing approval may name outside developer mode, and X-trust-06. The ceremony-class check, persisted rejects and the counter stay; dropping a check needs an operator ruling.Round-6 reviews (final, head 7392291): Fable (claude-fable-5-2, high) approve with 2 nits; Gemini 4 Argon (high) approve with no findings; adversary security review: no findings above info (1 info). All round-5 findings are resolved. Follow-ups after merge (not blocking): the interim "ordinary only" default should also name the hold window (Fable nit 1, adversary info); the reading's label could drop "round-3" (Fable nit 2). Astra's codex pool is below the tendcf floor until 2026-11-18, so it did not review rounds 3-6.
🤖 Generated with Claude Code
Note
High Risk
This PR redefines review gates in AGENTS.md and trust/consent/platform enforcement text across the architecture map and change-process docs, which directly governs how privileged work is reviewed and merged even though no runtime code changes.
Overview
This PR writes the 2026-11-14 goal-review walk into normative docs—mostly prose, no code or schema changes—so decisions live where implementers and reviewers already look, with
operator ruling, 2026-11-14citations and exact chosen options.AGENTS.mdstanding rule 3 is reworked for B10: a spike lane (unmergedspike/*, safety rails, GitHubcontents: read), tiered cross-model review (light for docs/spike notes, one frontier for device code, two for root/trust/key/handshake and design rulings), and a docs freeze until spikes run for state-1 work. It also records WarpBuild macOS check 8, PR 52 follow-up issues, ShizukuTendCF dev-mode/dev-build rules, AGPL-3.0-only review TUI licensing, and Mac SKIP vs Linux require fork loader policy test gating.architecture-DEFINITIVE-v3.mdabsorbs D47–D50 (adoption aim, platform order, Nix-first authoring, peer handshake/content checks), adds §9.12 safety score, standing approval modes, Android/macOS state-1 shape, fleet log archive indexes, a Stage 0–16 build order table, and widespread amendments to D6/D8/D14–D18/D45–D46.change-process.md,plugin-layout.md(defer out-of-tree plugins), and upstream overlay/register docs pick up B11/B-5b and fork-pin narrative.The
docs/review-2026-11-13/tree gains or extends goal-definition, fork-management, and ruling cross-references; several items stay ruled but implementation pending (e.g. overlay B-5b replace, schema v7+ grouping).Reviewed by Cursor Bugbot for commit 7392291. Bugbot is set up for automated code reviews on this repo. Configure here.