Repository navigation
chore(claude): allow-list the two landing REST calls in settings.json - #19047
Merged
Merged
Conversation
The repository's committed `.claude/settings.json` allow-listed every REST
write a seat makes through `curl` — labels, `PATCH .../pulls/*`, the reads —
except the two calls the landing act is made of. No `allow` rule matched a
`POST` or a `PUT` on `.../pulls/*` at all, so both fell through to the
harness's auto-mode classifier and were denied non-deterministically: green,
accepted, non-governed pull requests sat in draft for hours until a human
landed by hand what the rules say no human needs to look at.
Two rules, in the spelling the file's existing rules use:
Bash(curl -sS -X POST .../objectstack/pulls/*/ccr/ready_for_review *)
Bash(curl -sS -X PUT .../objectstack/pulls/*/ccr/auto_merge *)
Measured on this tree: `permissions.allow` 47 -> 49, `permissions.deny`
17 -> 17, key order unchanged, the file still parses.
⛔ Not a widening of who may land. The act stays behind the seat's own
pre-checks — record on head, `--pair` 0, checks green, the path face not
governed, the size face — and behind Prime Directive #14 for any diff that
touches a governed surface. These rules only stop the harness from
second-guessing a call the repository has already decided is the seat's.
No gate reads `permissions.allow`: `scripts/pm/check-settings-deny-roster.mjs`
pins `permissions.deny` against CONTENT_WRITE_TOOLS and states in its own
header that it does not assert that allow and deny agree; a sweep of
`scripts/`, `.github/` and `.claude/` finds no other reader. So the two added
rules move no gate except JSON validity.
Co-authored-by: Claude <noreply@anthropic.com>
hotlong
added
the
skip-changeset
This was referenced Sep 18, 2026
os-zhuang
approved these changes
Sep 18, 2026
os-zhuang
marked this pull request as ready for review
September 18, 2026 14:20
os-zhuang
deleted the
claude/issue-19014-settings-allow-landing-calls
branch
September 18, 2026 14:20
This was referenced Sep 18, 2026
Merged
os-zhuang
added a commit
to objectstack-ai/objectui
that referenced
this pull request
Sep 18, 2026
… (objectui#9862) (#9900) Fixes #9862 ## What this lands Two `permissions.allow` rules in `.claude/settings.json`, appended as the last entries of the array, in the spelling the framework repository's rules use and matching the command spelling the seats type for the landing act: ```text Bash(curl -sS -X POST https://api.github.com/repos/objectstack-ai/objectui/pulls/*/ccr/ready_for_review *) Bash(curl -sS -X PUT https://api.github.com/repos/objectstack-ai/objectui/pulls/*/ccr/auto_merge *) ``` Nothing else in the file moves: `permissions.allow` 10 → 12, `permissions.deny` 14 → 14, key order unchanged, three insertions and one deletion (the anchor line gains its trailing comma). ## Why Measured on the base of this branch: no `allow` rule matched a `POST` or a `PUT` on `…/pulls/*`, and none named `ccr/`. So `POST …/pulls/{n}/ccr/ready_for_review` and `PUT …/pulls/{n}/ccr/auto_merge` fell through to the harness's auto-mode classifier, which objectstack#18469 measured as non-deterministic on byte-identical calls, and which denies them today. The cost is the one seat 3 recorded in objectui#9800: ACCEPTED, green pull requests that no seat can land. This is the twin of objectstack#19014; the framework-side pair is in objectstack-ai/objectstack#19047. ## ⛔ What this is not Not a widening of who may land. The act stays behind the seat's own pre-checks — record on head, `--pair` 0, checks green, the path face not governed — and behind the governed-surface rule for any diff that touches one. These two rules only stop the harness from second-guessing a call the repository has already decided is the seat's. ## Measurements - **No gate reads either list.** A sweep of `scripts/` and `.github/` for `permissions.allow` or `permissions.deny` returns no reader, so the two added rules move no gate except JSON validity. - **JSON validity, asserted directly.** The file parses, and the parsed document reports 12 allow members and 14 deny members with the two new rules last. - **No changeset is owed, and that is measured rather than asserted.** `node scripts/check-changeset-presence.mjs` on this branch: "1 file(s) changed, 0 of them published source of a package the release covers, 0 of them a manifest whose published contract moved … so no changeset is owed." ## Landing `.claude/**` is a governed surface, so this pull request is left in draft and awaits one authorized APPROVED review. ⛔ Until it appears no seat flips it ready, joins the merge queue, arms auto-merge, or merges it — and ⛔ no seat submits that approval itself, under any account. --- _Generated by [Claude Code](https://claude.ai/code)_ Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: os-zhuang <jack@objectstack.ai>
This was referenced Sep 18, 2026
Merged
This was referenced Sep 21, 2026
os-project-manager
added a commit
that referenced
this pull request
Sep 21, 2026
…ings.json The objectui spellings of the two landing-endpoint allow rules that already exist for objectstack since #19047. Written by the director seat itself under the maintainer's explicit per-PR authorization, verbatim: 「19362 你可以直接开发，不派dev」 — the one charter exception to 「PM 永不写代码」 (a `.claude/` internal-tooling PR with the authorization quoted; reviewed and merged by the maintainer, never self-merged). Every dev-seat attempt to edit this file was refused by the session classifier (Self-Modification, 4 refusals on 2 cards); `deny` is untouched. The card relation is declared in the PR body. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012GcsUbuqFGBibkEDMRC1eE
This was referenced Sep 23, 2026
akarma-synetal
pushed a commit
to akarma-synetal/framework
that referenced
this pull request
Sep 28, 2026
…ings.json (ruling objectstack-ai#204-4 A) (objectstack-ai#19484) Fixes objectstack-ai#19362 Clause-②: no ## Ruling and authorization, verbatim Batch objectstack-ai#204 item 4, letter A (maintainer 「204 同意」; record 5754492446 on objectstack-ai#19362): the two landing endpoints are allow-listed by name for objectui, as they already are for objectstack since objectstack-ai#19047. Written by the director seat itself under the maintainer's explicit per-PR authorization in chat (2026-09-21): 「19362 你可以直接开发，不派dev」 — the one charter exception to 「PM 永不写代码」 (a `.claude/` internal-tooling PR with the authorization quoted). Per that exception the PR is ⛔ not self-reviewed and ⛔ not self-merged: the maintainer's own read and hand-merge is the review (Tier S by the register, `.claude/**`). ## What lands Two `permissions.allow` entries in `.claude/settings.json`, placed right after their objectstack twins: ``` Bash(curl -sS -X POST https://api.github.com/repos/objectstack-ai/objectui/pulls/*/ccr/ready_for_review *) Bash(curl -sS -X PUT https://api.github.com/repos/objectstack-ai/objectui/pulls/*/ccr/auto_merge *) ``` `deny` is byte-identical; key order and formatting unchanged; the file parses (55 allow entries, 2 matching `objectui/pulls/*/ccr`). Nothing else changes: an APPROVED review stays forbidden for every seat account, governed surfaces stay the maintainer's, the queue still runs every gate — these two calls only take a reviewed non-governed objectui PR out of draft and hand it to the queue. ## Why the seat wrote it Two dev rounds (this card's and PR objectstack-ai#19479's) were refused by the session permission classifier on every edit to this file (`[Self-Modification]`, four refusals), so the dev lane is not a channel for it; the maintainer authorized the seat to write the two lines directly. `Check Changeset` needs `skip-changeset` (no package touched) — applied by the seat. ## Verification `node -e 'JSON.parse(...)'` exit 0 · `node scripts/pm/check-settings-deny-roster.mjs` exit 0 (17 content-write tools declared = enforced, unchanged) · the diff is exactly +2 lines. --- _Generated by [Claude Code](https://claude.ai/code/session_012GcsUbuqFGBibkEDMRC1eE)_ Co-authored-by: Claude <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #19014
Clause-②: no
skip-changeset—.claude/settings.jsonis not published source of any released package and moves no published contract field, so this PR declares no release of its own.What this lands
Two
permissions.allowrules in.claude/settings.json, appended as the last entries of the array, in the spelling its existing rules use and matching the command spelling the seats type for the landing act:Nothing else in the file moves:
permissions.allow47 → 49,permissions.deny17 → 17, key order unchanged, three insertions and one deletion (the anchor line gains its trailing comma).Why
Every REST write a seat makes through
curlwas already allow-listed by shape —POST …/issues/*/labels,PATCH …/pulls/*, the reads — except the two calls the landing act is made of. Noallowrule matched aPOSTor aPUTon…/pulls/*at all, so both fell through to the harness's auto-mode classifier, which #18469 measured as non-deterministic and which denied them repeatedly. The visible cost the card records: seven ACCEPTED, green, non-governed PRs sat in draft for hours until a human landed by hand what the rules say no human needs to look at.⛔ What this is not
Not a widening of who may land. The act stays behind the seat's own pre-checks — record on head,
--pair0, checks green, the path face NOT governed, and the size face of #19012 — and behind Prime Directive #14 for any diff touching a governed surface. These two rules only stop the harness from second-guessing a call the repository has already decided is the seat's.The structural pin the card asked about — measured, and there is none
The card asked for "the self-test / structural pin the settings file already has for its rule set, if one exists (measure: grep the tree for a test that reads
permissions.allow)". Measured on this branch:scripts/pm/check-settings-deny-roster.mjspinspermissions.denyequal toCONTENT_WRITE_TOOLSand states under its own "What is deliberately NOT asserted" heading that it does not assert that allow and deny agree. It reads the allow list only in one self-test fixture, to prove that a document with nodenykey reads as absent rather than malformed.scripts/,.github/and.claude/for any other reader ofpermissions.allowreturns none.So no pin exists to extend, and the two added rules move no gate except JSON validity — which every reader of the file needs, and which is asserted below.
Gates
node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack .claude/settings.jsonon this branch derives 15 commands. All 15 ran; 15 pass.One finding was printed and is not this diff's:
check-agent-test-spellingflags a line in.claude/settings.local.json, a file that is untracked here and ignored by the global ignore file, so it exists on one workstation and never reaches CI. Nothing in this PR touches it.JSON validity is asserted directly: the file parses, and the parsed document reports 49 allow members and 17 deny members with the two new rules last.
Landing
.claude/**is a governed surface (Prime Directive #14), so this PR is left in draft and awaits the maintainer's word. ⛔ No seat flips it ready, queues it, arms auto-merge on it, or approves it.Generated by Claude Code