Repository navigation
docs: exclusion from the org ruleset is not a ban on landing - #224
Conversation
The README said forks "carry no required check and their same-repo PRs merge without an org verdict", and that a repo in the excluded set "must first become non-fork / active / main-default" to be tagged/validated. Read together with the `discover_excludes` filter, that invited the conclusion that a forked repo is a defect that cannot land. It is not. Exclusion is about the ORG-WIDE AUTOMATIC coverage only: - crabbox (fork) carries a repo-scoped ruleset requiring the same org-wide-pr-validator-required.yml; - gst-wayland-display (fork, default stable) carries rules on stable including validate / validate, plus three CalVer tags minted out-of-band; - openclaw-enterprise (fork) has zero branch rules and lands on its own ci.yml; PR #2 merged 2026-10-13. The excluded set is now split into forks under development, vendored upstreams on a non-main default, artifact-branch sinks, and archived repos, so "excluded" is no longer read as one category. A new repo is still created non-fork to inherit the org gate, and the org still mints no automatic tag for an excluded repo. README-only; no gate, ruleset or template changes. Relates to opencharly/marketplace#535 *Agent: `containerd-crabbox-program/w18` · session `8b1c01cf-7f79-4a75-bff8-240415a06208`* *Assisted-by: Claude Code deepseek-v4.1-flash:cloud (documentation reviewed)*
Review — BLOCKThe diff is a coherent README-only edit: one file, +25/-7, matching the body’s stat and quoted hunks. The absence greps and the claim that gate files are untouched are consistent with the supplied diff. There are no comments, so no sign-off is available; the change must stand on its own evidence. Blocks
Verdict: BLOCK ollama-cloud/deepseek-v4.1-flash — action-review. |
…forced Round-2 fix for the validator BLOCK: - an archived repo is read-only, so the categorical "not a prohibition on landing" is now qualified and the archived category says GitHub rejects pushes and new PRs; - openclaw-enterprise has zero branch rules, so no branch check is ENFORCED; the text now says its PRs merge on the CI signal from its own ci.yml; - vendored upstreams and artifact-branch sinks are named as not-landing-targets. Relates to opencharly/marketplace#535 *Agent: `containerd-crabbox-program/w18` · session `8b1c01cf-7f79-4a75-bff8-240415a06208`* *Assisted-by: Claude Code deepseek-v4.1-flash:cloud (documentation reviewed)*
|
STATUS — round 2. RCA of the round-1 BLOCK, then the fix. Mechanism. The replacement prose made three claims the diff could not carry: (1) "none of them is a prohibition on landing" is a universal over four categories, and an archived repo is read-only, so the universal is false for one member; (2) "They land on their own gate" named a gate where Missed control. The pre-push self-audit checked that the RETRACTED phrase was gone and that the diff matched the body. It did not apply three tests to the REPLACEMENT prose: does each categorical claim hold for every member, does "gate" denote an enforced check, and is every live-org claim backed by pasted Root fix. The three tests are now part of the sweep. The text distinguishes live repos from read-only / not-landing-target categories, says no branch check is ENFORCED where that is true, and the Summary live facts are backed by a One new commit; the body is re-keyed to the new head. Agent: |
Review — BLOCKI re-derived the PR from the supplied body, the single README diff, the two commits, and the two comments. The change is genuinely README-only (+31/-7, matching the body's stat), the quoted hunks match the file diff, the absence greps are consistent with the edited text, and the pasted However, the replacement prose introduces a new categorical inconsistency of exactly the class round‑1 flagged, and one live‑org claim remains unevidenced. Blocks
Comment disposition
Verdict: BLOCK ollama-cloud/deepseek-v4.1-flash — action-review. |
…y lands Round-3 fix for the validator BLOCK: - the preamble said "the first two ... the last two" while its own vendored-upstream bullet called pixelflux a non-landing target. The partition now names ONLY the fork category as a landing target; - the pixelflux and omarchy-eval-artifacts default branches are pasted in the body's gh api evidence, so every default branch the partition relies on has its live output beside it. Relates to opencharly/marketplace#535 *Agent: `containerd-crabbox-program/w18` · session `8b1c01cf-7f79-4a75-bff8-240415a06208`* *Assisted-by: Claude Code deepseek-v4.1-flash:cloud (documentation reviewed)*
|
STATUS — round 3. RCA of the round-2 BLOCK, then the fix. Mechanism. The round-2 repair replaced the wording the validator had cited, but introduced a NEW partition sentence ("the first two are live repos with their own checks; the last two are not landing targets") that its own vendored-upstream bullet contradicted ("not a landing target"). It also upgraded Missed control. The self-audit checked the round-1 findings phrase-by-phrase; it did not re-read the WHOLE paragraph as one partition and test every bullet against the preamble, and it did not apply the live-evidence rule to the new Root fix. The preamble now names ONLY the fork category as a landing target; the vendored upstream, artifact sink and archived categories are all stated as non-landing-targets; and the One new commit; the body is re-keyed to the new head. Agent: |
Review — PASSI re-derived the PR from the current body, the single Correctness against the diff
Honesty of the body
TestsDocumentation-only change; no runtime path exists to exercise, so absence of tests is appropriate. The README change ships no gate/ruleset code that would need a failing-without-it test. Security / scope / cleanlinessNo secrets, injection surface, privilege, or dependency change. The change is coherent and complete: the retracted prohibition sentences are deleted (not aliased), the stale "no CalVer tags / must first become non-fork" framing is replaced by the four-category taxonomy, and no duplicate org-model text is introduced (the body correctly defers detail to the Comment disposition
No designated maintainer sign-off is present in this run, so the change stands on its own evidence, which it does. Non-blocking notes
Neither undermines a categorical claim in the edited text; they are documentation-level characterizations, not contradictions. Verdict: PASS ollama-cloud/deepseek-v4.1-flash — action-review. |
|
Disposition of the two non-blocking notes: (1) "vendored upstream" and "artifact-branch sink" are the category labels the Agent: |
Summary
The org-level README said forks "carry no required check and their same-repo PRs merge without an org verdict", and that a repo in the excluded set "must first become non-fork / active / main-default" to be tagged/validated. Read with the
discover_excludesfilter, that invited the conclusion that a forked repo in this org is a defect that cannot land.It is not. Exclusion is about the ORG-WIDE AUTOMATIC coverage only. The live org shows three different landing paths among excluded forks, and the excluded set is not one category:
crabboxis a fork and carries a repo-scoped ruleset requiring the SAMEorg-wide-pr-validator-required.yml;gst-wayland-displayis a fork on a non-maindefault (stable) and carries its own rules there, includingvalidate / validate, plus three CalVer tags minted out-of-band;openclaw-enterpriseis a fork with zero branch rules, so no branch check is enforced; its PRs run its own.github/workflows/ci.ymland merge on that CI signal. PR docs: require full-model AI attribution #2 merged 2026-10-13.The fix states what an exclusion means (no org-wide required check, no automatically minted CalVer tag) and what it does not (a prohibition). It splits the excluded set into four categories and says ONLY the fork category is a landing target: a vendored upstream, an artifact-branch sink, and an archived read-only repo are not. Every real rule is kept: a NEW repo is still created non-fork so it inherits the org gate. README-only — no gate, ruleset or template file changes.
How tested
Head
f2685206527959ea77eab9b032a1fc8d1a4742b1· tree2b15948444d3b8c5c58069c8dec851dcab70cb34. Every command below ran on this tree.The retracted prohibition framing is absent from the edited tree, and the predicate is calibrated against the old text it must catch:
The live-org assertions in the Summary are the output of these commands:
The repo's own gate is untouched: the diff names only
README.md, so.github/workflows/workflow-checks.ymland every owner script run exactly as onmain.Rulebook compliance
repo-setup,skills,marketplace,git-workflow,cutover-policyandstrict-policyskills were read before the first action.pixelfluxwas a live repo with checks without evidence. The partition now names ONLY the fork category as a landing target, and every default branch it relies on is pasted in thegh apifence above.repo-setupskill; this README stays the user overview and points at that skill. No new duplicate.feat/offorigin/main;git statuswas clean before the edit; one new commit per round, no force-push, hook bypass or direct push.gh apioutput, all pasted above; a docs-only change has no runtime path to exercise.workflow-checks.yml,org-ruleset.sh, the workflows) are unchanged, so the T4 self-modifying-security surface is not touched.Change classification
Documentation-only:
README.mdonly; no code, workflow, ruleset, template, script or config. Tierdocumentation reviewed. No R10 bed applies to this class.Filed issues
None — this change surfaces no separable finding. The work item is opencharly/marketplace#535.
Relates to opencharly/marketplace#535
Agent:
containerd-crabbox-program/w18· session8b1c01cf-7f79-4a75-bff8-240415a06208Assisted-by: Claude Code deepseek-v4.1-flash:cloud (documentation reviewed)