Repository navigation
ISS-14136, ISS-14127, ISS-14112: No-surface criteria, gpt-6.2 - #433
Merged
Merged
Conversation
Move every ClosedLoop and symphony-alpha specific line out of vibe and vibe-core into the new optional vibe-closedloop-provider plugin, which ships text and data only. With the provider bound, sessions keep today's behavior; without it, vibe and fleet work from the repository itself: the person's description, plans in the private git directory, the AGENTS.md test command, the lockfile install, and a requested PR or named recipient at the end. - provider binding, provider text assembly and data access in vibe-core - guards read their rules from the bound provider and stay silent outside an activated chat - main sync gates its provider-only machinery on the provider's recipe - the recipes opt-in file is validated and withholds recipes with one reason - the provider is offered for install once, never re-enabled or reinstalled - generic scan gate, generic walk fixture, eval arms, and generic twins for the provider-bound tests - vibe 1.1.81, vibe-core 1.0.5, vibe-closedloop-provider 1.0.0
- writer ledgers parse capabilities by shape; grants are checked against the session binding at register and launch - the Claude launcher resolves provider text for its roles - writers inherit the lead's binding through a Provider line, enforced by the Codex launcher, the Claude launcher and the Agent guard - the recipes opt-in file is validated and judged in the session repository - adopting a session keeps its binding; moved provider releases re-resolve - every installed provider's plan folder stays private - generic fleet recovery ids and duties, the worktree exclude rule, the guardrails formatter from the verify recipe - dangling reference paths fixed, provider text free of ticket ids and home paths, the inventory snapshot and mapping checked in pytest - the eval provider arm installs provider-resolved text - bundles build byte-identically from any working directory
Bind the tracker in the tracker-duty tests, run spawned scripts outside the real checkout so the cwd repository leg cannot pick it, derive the provider line in fleet tests from the fixture binding, fix the shadowed prototype deploymentHost check and rebuild dist, list provider.mjs as a vibe-core wrapper, and record the changelog word count.
Pass the lead's binding to remote fleet writers, read the recipes opt-in from the main checkout for older sessions and for a named checkout, know UI files by extension without an areas recipe, never write a legacy provider record generic, keep old hook scope records active, restore the exact ticket pattern for Prior removal, keep clerkOrgId for older vibe, report a main sync with no check as needing review, require the sole workflow for an undeclared lane source, keep generic plans in the common git dir, move the repository E2E text into provider blocks, and stop offering an install over an unreadable provider.
…c-vibe-v2 # Conflicts: # CHANGELOG.md # plugins/vibe/.claude-plugin/plugin.json # plugins/vibe/.codex-plugin/plugin.json
…ider - provider.mjs assemble writes a vibe file merged for the session's binding, and assemble --blocks gives every bound block of a loaded SKILL.md in one call; the reading rules use them - each generic region and its provider block share their identical lines once, so assembled text is unchanged and the provider is smaller - only an unbound capability falls back to the generic text: a bound provider that is gone, a missing block or an unreadable record stops with its reason and fix (exit 4) in the CLI, the launcher, every script and the hooks, and a guard refuses the action it guards - session records are written whole and renamed into place - the 1.1.82 word count is measured at this tree
ISS-14039 (#427) took vibe 1.1.82 and vibe-core 1.0.5. Keep both sides of the claude-worker launcher (the broken-provider mapping and the review model), rebuild the bundles, drop the ticket slugs the merge brought into runtime files, and remeasure the word count against the new main.
ISS-14028 and ISS-14024 (#428) took vibe 1.1.84 and vibe-core 1.0.7. The unit-status guard keeps the bound provider's rule and gains the chat's own program board; its tests keep both cases. The fleet-resume header covers both changes, the bundles are rebuilt, the ticket slugs the merge brought into runtime comments are dropped, the launcher mark written into installed launchers stays as an allowlisted compatibility line, and the word count is remeasured against the new main.
The evals policy file is the one source for the default Codex model; cl-sweep, the vibe agent binding and vibe-bench read it. cl-sweep ticket owners honour the per-machine writer override, and a saved session on another model is replaced at its next idle boundary instead of refused. vibe 1.1.87.
Brings in ISS-14077 (#430). vibe stays 1.1.87 on top of 1.1.86.
Operator ruling 2026-11-13: the generic orchestrator no longer refuses a new gpt-5.5 session, so an explicit or override model works the same on all three transports. Removes the two constants only that refusal used. CHANGELOG regenerated with /update-documentation.
…e check A bound session now behaves like main. The start is one version check when the provider skill's description names its calls, and --closedloop/--graph are accepted again as aliases of --connection; otherwise the read-only provider probe runs first, and a check with no states records unprobed instead of binding generic. Each capability binds on its own connection, so a missing graph keeps the ClosedLoop text and restores the graph. The check writes the bound text once per repository and binding, and readers open files from that folder instead of assembling each one. check always exits 0 and prints the setup block in every state. Also: precedent stays live-only, a brief without a Provider: line inherits the lead's binding, the preflight finds the checkout as main did when bound, vibe-bench binds through the real check, and the eval runner gains a provider-installed arm.
Criteria audit and unit review accept a reviewer's No surface mark with a reason and a named test row in place of an app run; an unmarked criterion keeps its rule. The surface decision and the mark live in one module. The duplicate-helper rule no longer flags a helper inside a committed bundle the repo builds from that source, or one bundle against another bundle of the same source; a hand-written copy is still flagged. A frozen, shrink-only guard lists every bare-word classifier in the gate scripts. vibe 1.1.89.
…-generic - tools/vibe-closedloop-build assembles plugins/vibe plus the provider into the committed plugins/vibe-closedloop (the `vibe` entry), with cold deploy and environment sections in their own files, recipes inline, and a CI drift check. plugins/vibe is published as `vibe-generic`. - Retire runtime provider text: the rule paragraph (two generic sentences stay, dropped by the build), provider-text.md, the bind-time folder, model-run assemble/block/path/tree, the install offer, provider-install, provider-match, the skill-listed flag and the provider skill. Discovery reads only the provider bundled in vibe. - Each install finds its own plugin name and marketplace; with both installed the ClosedLoop vibe runs the hooks and binds, and the check reports both-installed. - Eval arms install the built folder (with the repository opt-in file in the case folder) and plugins/vibe as `vibe`; no harness assembly. - Resolve run.py's private homes and every main guard to real paths.
…the word-count entry - provider.mjs knownProviderContracts lists the provider bundled in every vibe install or checkout, enabled or not; local-plans uses it so a plan written while bound stays private when the session is not, whichever vibe the lookup order picks. Binding and lookup orders are unchanged. - The vibe CHANGELOG entry names the full Markdown scope of its word count again, with the generated folder's exclusion stated after it.
… 1.1.97 - The build places the provider's whole-file references (ticket-template, closedloop-graph, ticket-units) at main's paths in plugins/vibe-closedloop and rewrites provider: names to main's relative forms; provider/ holds only provider.json, whose placed map providerPath resolves through. A build test requires every Markdown file reference in the built tree to resolve. - vibe 1.1.97, vibe-core 1.0.13 (VIBE_CORE_MIN raised to it); #432's new cloud code follows the branch's generic rules.
Merge feat/iss-13446-generic-vibe (e316b80) per the stack-on-#426 ruling. Generic behavior stays in plugins/vibe; the ticket-units and handoff unit-review pointers move to the provider. Ticket ids leave generic runtime text per OQ10, provenance kept. Build excludes every test fixture module by name; plugins/vibe-closedloop rebuilt.
The provider gained this PR's ticket-units and handoff pointer edits, so it takes its own PATCH bump with a CHANGELOG entry.
Folds #431 (ISS-14112) into this branch: one vibe 1.1.98 release and one changelog entry. ISS-14136: a reviewer can mark a criterion as having no surface, with a reason, and the audit then asks for a named test row instead of an app run. ISS-14127: the duplicate-helper rule skips the launcher bundle built from the flagged source. ISS-14112: Codex workers default to gpt-6.2-sol, with gpt-6.2-astra for planning and review lanes on HIGH tickets, falling back to gpt-5.6-sol when the running Codex does not list the model.
Criteria lists read or report every checkbox form. A no-surface mark must name a test in a file a test runner collects, and an indented mark is accepted. The three cl-sweep transports share one model-list launch helper and one session model field helper, and the cl-sweep instructions now describe the one init fallback and its modelNote. ISS-14112 tests carry their ticket in their titles.
Criteria lists read only - and * boxes under a criteria heading, as before, and report every other checkbox form, so criterion ids do not move. The shared test-file owner decides which files a no-surface mark may name. The App Server and generic orchestrator fallbacks have their own tests, and the cl-sweep docs name both commands that take the fallback.
Inside an acceptance criteria section, a + or ordered box under a sub-heading or list label is reported under the criteria heading instead of dropped. The changelog names both commands that take the model fallback.
One report-only rule covers both the Acceptance criteria heading and the template list: inside either section, a box the audit does not read is reported, never read, and ids stay as main reads them.
The template criteria list reports every box in the span it reads, so an indented list never drops one silently, and a shallower sibling label closes it so an unrelated checklist is not reported. Read ids are unchanged from main.
Reading stays exactly as on main. Inside a criteria section every checkbox is read or reported; the template list closes only at a heading or a line at or left of its own line, so a sibling section indented between them is reported rather than dropped (orchestrator decision: AC-4 outranks the earlier close rule).
A heading deeper than the one the template criteria list sits under no longer closes its report section, matching the heading form; a checkbox left of the list line still closes it.
The cl-sweep test fixture runs every test with an empty temporary HOME and a codex shim ahead of any real codex, failing the file when a test reaches it. The model-list fake also answers core discovery, which 9d3fe99's fake refused, so tests fell through to a live Codex locally and failed in CI. Production code is unchanged.
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
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.
Summary
This pull request covers three tickets: ISS-14136, ISS-14127 and ISS-14112. ISS-14112 was folded in from #431 under the operator's ruling "Fold #431 into #433" (2026-11-17): one version, one changelog set, one CI run and one merge. #431 will be closed with a pointer to this pull request.
ISS-14136: no-surface criteria
No surface: <criterion id>: <reason>in the audit file, plus a MET row that names a test (<file>::<test name>). For a marked criterion,criteria-audit.mjsasks for that test instead of an app run, andunit-review.mjsneeds no capture. Both still require the code review row. This follows the operator's ruling of 2026-11-13 (option B).criterion-surface.mjs. Both the criteria audit and the unit review ask it.criteria-list.mjs) now reports criteria checkboxes it does not read instead of dropping them silently. It still reads the same ids as main. The audit and the unit review refuse those boxes with a message that says how to write them.references/spec-card.md, "Audit rows", and the provider'sticket-units.md) say how to record a mark.ISS-14127: built bundles and the duplicate-helper rule
checkDuplicateHelpersinreview-gap-rules.mjs) treats a generated bundle and the source it inlined as one definition. Two bundles that inlined the same source are treated the same way. The launcher bundle underskills/vibe/scripts/dist/is therefore no longer flagged against the source it was built from.dist/.ISS-14112: Codex models (with the model-switch-6.2 scope)
gpt-6.2-sol. It has one source,codexModelinevals/codex/model-policy.json. This follows the operator rulings of 2026-11-12 and 2026-11-17 ("let's make the switch from gpt-5.6-sol to gpt-6.2-sol and gpt-6.2-astra for high complexity planning/reviews").model-catalog.mjs). When that list lacksgpt-6.2-solorgpt-6.2-astra, or cannot be read, the worker runs ongpt-5.6-soland the launch output says so once. No model is guessed from the CLI version. This follows the operator's ruling of 2026-11-17: "gpt-6.2-sol was added today which means it's tied to a specific version of codex-cli".gpt-6.2-astraonly when cl-analyze graded the ticket HIGH, and ongpt-6.2-solotherwise. Vibe session workers have no grade, so they keep the role defaults. This follows the operator's ruling of 2026-11-17, "Only HIGH-complexity work".gpt-6.2-sol, per the operator's ruling of 2026-11-17, "Switch evals too".vibebuild now shipsevals/codex/model-policy.json, becausemodel-policy.mjsreads it when it loads.Versions
plugins/vibe-closedloop) 1.1.101, rebuilt fromplugins/vibeand the provider.Instruction length
plugins/vibe, measured byinstruction-words.mjs(proof step d02) at this commit. It comes from the text approved for ISS-14112 and from the model-switch ruling.Approved sentences
model-policy.mjswere approved by the operator as written on 2026-11-17 ("Approve both as written"):<model>is not in this Codex's model list, so this worker runs on<fallback>."<error>), so this worker runs on<fallback>instead of<model>."<name>" that read as acceptance criteria in a form or place this audit does not read; write each as a "- [ ]" box directly in the "Acceptance criteria" section or the "- Acceptance criteria:" list";<id>:<file>is not a test file a test runner collects", and "a no-surface mark reads "No surface:<id>:<reason>"".Handoff gate disclosure
Per the orchestrator's answer Q-GATE, under the operator's ruling "Run the worktree's gate", the handoff gate ran from this branch's own built plugin.
The gate ran with a blocking-stdout preload (NODE_OPTIONS) and no code change. handoff-inventory.mjs writes its JSON and then calls process.exit (lines 680 and 681, the same on main), so a child run under spawnSync loses any output past the 64 KiB pipe buffer. This PR's pull-request inventory is 66,790 bytes, so the unpatched gate parsed truncated JSON. Run directly, the same inventory exits 0 with every blocking check true. The fix (write, then set process.exitCode) goes into the ISS-14116 prevention PR.
Contract changes
From
app-server-worker-session.mjs:REQUIRED_WORKER_MODELis removed, replaced byDEFAULT_WORKER_MODEL.LEGACY_COMPATIBLE_WORKER_MODELis removed.initializeSessionis changed: it accepts any compact model identifier and recordsrequestedModelExplicit.From
model-policy.mjs:TICKET_OWNER_MODELis changed: it is read fromevals/codex/model-policy.json.LEGACY_UNVERSIONED_MODELandisLegacyTicketOwnerModelare removed.CodexWorkerRolemoves toagent-models.mjs.From the other scripts:
LEGACY_WORKER_MODELis removed fromnative-app-server-orchestrator.mjs.EvidenceAtmoves fromcriteria-audit.mjstocriterion-surface.mjs.namesInis removed fromunit-review.mjs; the surface words now live incriterion-surface.mjs.Each of these changes appears in both
plugins/vibeand the builtplugins/vibe-closedloop.Assertion changes
These expectations follow the operator rulings above.
The
gpt-6-soltogpt-6.2-solswitch, and the rename ofREQUIRED_WORKER_MODELtoDEFAULT_WORKER_MODEL:The default case now uses a model outside the policy (
gpt-5.6-sol), so the test still tells the default apart from the policy model:The legacy model list is removed, so the skill text must no longer name it:
The requested-model option accepts any compact model identifier and refuses one that is not compact:
The published build ships
evals/codex/model-policy.jsonand no other eval file:Overlap
CHANGELOG.md, the two vibe plugin manifests andplugins/vibe/README.md. The pull request that merges second takes the next version and merges the changelog.CHANGELOG.md.Test plan
Targeted tests ruling: the operator's ruling R-TARGETED-TESTS-433 (2026-11-18, recorded in the program rulings file) lets this pull request use targeted tests on the final commit in place of full local lanes. The full CI lanes are left to GitHub CI, which runs them on this pull request after the push.
The targeted inventory runs on the final commit through
proof-fold.sh:pytest plugins/,ruffandpyright;The final head is 529c76a; its full Node Tests lane result is recorded in the PR evidence. The proof run below was on 9d3fe99, before the CI fix: 127 steps ran. 126 matched their expected result, 0 had no result and 0 were deferred. The one mismatch is k04 (below).
pytest plugins/(p01) and the Codex eval mirror (p02): pass.ruff(p03) andpyright(p04): pass.tools/vibe-workersis byte-identical to main 0057c43, so the timeout is not caused by this branch.CI fix: GitHub CI on 9d3fe99 failed "Node Tests (skill scripts)" on 31 cl-sweep tests. The branch's model-list fake refused
codex plugin list, so core discovery fell back to the bundled ChatGPT-app Codex, which CI does not have. 529c76a sandboxes every cl-sweep test in the shared fixture (a temporary HOME, the harness variables cleared, and a codex shim first on PATH that fails the file when a test reaches it) and makes the model-list fake answer core discovery. Production code is unchanged. Four guard tests cover it, with mutations mu46 to mu49.Criteria reader decision (orchestrator, 2026-11-18): ISS-14136 AC-4 ('never dropped silently') outranks the orchestrator's earlier ADV-R5-2 call. The template list's report section now closes only where the boundary is clear: at a heading that is the same depth as, or shallower than, the heading the list sits under (a deeper heading keeps it open, as the heading form does), at another line at or left of the
- Acceptance criteria:line, or at a checkbox left of that line. The section stays open while main's reader still reads. Reading is unchanged from main 0057c43: 0 id differences on 47 reader rows and on 1,250,000 seeded fuzz tickets. The accepted cost is an over-report. A sibling section indented deeper than the list line but shallower than its boxes, such as- Environment:, has its boxes refused under "Acceptance criteria" instead of passed silently; these are the ADV-R5-2 and GR6-2 layouts, and they replace the former close test. Every other trigger row from review rounds 3 to 7 is read or reported. A report-side oracle over a seeded corpus (in the test file) finds no box that sits inside a criteria section, would be read by main, and is neither read nor reported by this reader. The same oracle flags the drops of the two earlier rules. Non-blocking review findings recorded as-is (orchestrator answer Q-R7): GR7-1, ADV-R7-3, GR7-4. k04: main-sync.test.ts timed out under machine load; tools/vibe-workers is byte-identical to main, so the timeout is not caused by this branch.ADV-R8-1 (two
- Acceptance criteria:lists, the second under a deeper heading; a later box is neither read nor reported) is fixed in the ISS-14116 prevention PR, which merges after this one.Seven changed lines of review-class-guards.test.mjs (118, 125, 126, 128, 131, 135, 154) name the guard's MOBILE_WORD fixture, not a narrow or mobile state; the handoff gate's responsive rule (the ISS-14128 misfire) asks a failing-mutation proof for each, and each has a real one recorded in the PR evidence rather than an override.
Recorded non-blocking review notes:
Recorded non-blocking review notes on the CI fix, fixed in the ISS-14116 prevention PR, which merges after this one:
Known environment failures, none caused by this branch:
tools/vibe-workersis byte-identical to main.Missing tests
None. The final verification found no criterion or decision row without an executed test.