Skip to content

ISS-13446: Run vibe and fleet in any repository - #426

Merged
shafty023 merged 19 commits into
mainfrom
feat/iss-13446-generic-vibe
Nov 17, 2026
Merged

shafty023 merged 19 commits into
mainfrom
feat/iss-13446-generic-vibe

Conversation

@shafty023

@shafty023 shafty023 commented Nov 11, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

  • ISS-13446: vibe and fleet-orchestrator run in any git repository. Everything ClosedLoop and symphony-alpha specific moves into a new optional plugin, vibe-closedloop-provider, that ships text and data only (provider.json plus reference blocks). Plan: the approved v2 plan, with Daniel's answers to its open questions (OQ1 to OQ12) recorded on the ticket.

  • Both prerequisites have landed: symphony-alpha #9779 (merged as e3f9f4d8e7, commits .closedloop-ai/vibe-recipes.json, the opt-in this PR reads) and the fleet orchestrator's migration of its orchestration directory into the fleet home. origin/main is merged in, most recently 8a858bb (ISS-14109: Default vibe writers to Opus, overridable models #429 and ISS-14077: Honor a recorded targeted-tests ruling in the handoff gate #430, vibe 1.1.86 / vibe-core 1.0.9; see Merged with main), so this change takes vibe 1.1.88 and vibe-core 1.0.10.

  • Binding: one probe at vibe or fleet start records provider: {name, state, capabilities, ...} on the startup record and the session; scripts, workers and hooks read it and never probe. States are none, bound and bound-disconnected (a failed precondition is surfaced, never a silent downgrade). Records from before bindings existed wait for a chat that binds instead of being written generic.

  • Repository opt-in (OQ5): recipes bind only when the repository commits .closedloop-ai/vibe-recipes.json naming the provider; it is judged in the session's repository, falling back to its main checkout for sessions branched before the file landed. No repository names are hard-coded anywhere.

  • Generic path (vibe-generic installed): description intake with no tracker, operator identity from gh, local #n work ids for fleet units, plans and decision tables under <git common dir>/vibe/<slug>/, bootstrap from AGENTS.md or the lockfile, an unknown layout takes the full review path, UI files known by extension, repo-owned gates run only when the repo ships them, main sync with nothing to run reports NEEDS_REVIEW, an undeclared lane source must be the AGENTS.md test command or the repo's only workflow.

  • Hooks: every vibe-core guard is gated on activation (a vibe-plugin skill in the chat, a launcher-started worker, or a cwd in a vibe worktree); elsewhere a guard prints, writes and calls nothing. ClosedLoop guard rules (unit status, e2e slot) come from the bound provider's data.

  • Fleet: writers inherit the lead's binding through the brief's Provider: line; remote writers get it with the launch (--lead-provider). Status sweeps, ticket units and placement by ticket are provider duties.

  • Install (OQ1), superseded at this head: there is no provider plugin to install. ClosedLoop machines get the built vibe; going generic means installing vibe-generic (see Release-time build).

  • Gates: vibe_generic_scan.py check fails on provider-specific text in runtime files (exact-line allowlist with reasons); the inventory diff test proves no gating line of the base was dropped; generic-walk.test.mjs runs the walk's deterministic steps in a personal repository and requires an empty blocker ledger; the Codex eval runner has provider and generic arms.

  • Two independent adversarial reviews (A and B) ran on the full diff; every finding was verified against the code. A1 to A7 and B2 to B10 are fixed with tests: remote fleet writers could not start under a bound lead; the opt-in was read only from the session branch; generic UI diffs counted as server-only; legacy records could be written generic; old hook scope records stopped activating; the Prior removal check widened; clerkOrgId was no longer written for an older vibe; main sync with no check reported DONE; an undeclared workflow lane source could be any workflow; generic plans lived in the per-worktree git dir; symphony E2E text sat in generic handoff text; an unreadable provider was offered for reinstall.

  • Reported, not changed:

    • A8: vibe-core 1.0.5 installed beside an older vibe. Both ship in this PR, and the start check flags an outdated pair.
    • B11: in a chat that never activated vibe, activation reads the transcript on each hook event. The proposed fix is a state write, which the activation ruling forbids.
  • Known limit, tracked as ISS-14022 (review finding B1) and not changed here: vibe still assumes the default branch is main unless vibe.baseRef is set (session creation, handoff gate, several merge-base diffs, main sync).

  • Reading provider text, superseded at this head: runtime assembly (assemble, --blocks, the bind-time folder) is retired; the text is assembled at release time (see Release-time build).

  • Broken provider (orchestrator rulings): only "no provider with is bound" (exit 3) means generic text. A bound provider that is gone, a missing block or reference, or a session or startup record that exists but cannot be read stops with provider-broken: <reason>. Fix: ... (exit 4) in provider.mjs, the Claude launcher (PROVIDER_BROKEN), and every vibe script that reads the provider. Hooks never throw: they show the reason and fix and keep their generic default, and a guard whose default would relax the bound rule refuses the action it guards (unit-status-guard on an MCP status write, shell-safety-guard on an executed e2e script, pr-create-guard when the override decides). Session records are written atomically.

  • Also fixed: prototype-session.mjs shadowed deploymentHost with a const of the same name (a throw on every call) and its bundle was stale; test helpers ran scripts from the real checkout, which the new cwd repository leg could pick up.

  • Versions at this head: vibe 1.1.97, vibe-core 1.0.13 (VIBE_CORE_MIN 1.0.13), vibe-closedloop-provider 1.0.0. These are above main's 1.1.94 / 1.0.12 and every open PR's reserved version. Earlier, at 22c726c: vibe 1.1.88, vibe-core 1.0.10; dist bundles rebuilt. Word count against 1.1.86 (origin/main 8a858bb) at this head (instruction-words.mjs --changelog passes): vibe about 19982 words shorter, the provider adds about 42419, all plugin Markdown about 22833 longer. The merges brought ticket slugs into runtime files (four (ISS-14039) from ISS-14039: Accept a Fable review for a Claude cloud import #427, five ISS-14028/ISS-14024 comments from ISS-14028, ISS-14024: Bind fleet chats to their programs, safe migrate #428, and the ISS-14109/ISS-14077 comments from ISS-14109: Default vibe writers to Opus, overridable models #429 and ISS-14077: Honor a recorded targeted-tests ruling in the handoff gate #430); they are removed here (OQ10). The sixth from ISS-14028, ISS-14024: Bind fleet chats to their programs, safe migrate #428, fleet-home.mjs LAUNCHER_MARK, is text written into installed launchers and matched before an overwrite, so it stays as an allowlisted compatibility line.

  • Merging ISS-14028, ISS-14024: Bind fleet chats to their programs, safe migrate #428: unit-status-guard keeps the bound provider's rule and gains ISS-14028, ISS-14024: Bind fleet chats to their programs, safe migrate #428's per-chat program board (both tests kept; ISS-14028, ISS-14024: Bind fleet chats to their programs, safe migrate #428's chat test now passes the provider rule). provider.json (1310 words) is data only scripts read; per-file provider headers and block markers are never printed to an agent.

Release-time build (this head)

The realistic eval arm showed a real ClosedLoop session reading the generic text and missing main's steps (findings/iss13446-realistic.md). This head drops runtime assembly and builds the ClosedLoop vibe at release time.

Operator rulings this head implements (quoted from the program's rulings record):

  • AC5 and Q1, amended. ISS-13446 AC5 and Q1 said the ClosedLoop knowledge is its own plugin that the user disables to go generic. That is amended. Daniel: going generic = "Install the other vibe (Recommended)" (amends ISS-13446 AC5/Q1: no separate ClosedLoop plugin to disable).
  • Naming. Daniel: "I don't want a separate repo. that's out of the question. we use claude-plugins repo and that's it", then "vibe-generic plugin (Recommended)": a second plugin in the closedloop-ai catalog named vibe-generic. Its manifest name is vibe-generic, and Codex lists its skills as vibe-generic:*. vibe text works under either plugin name. The eval keeps installing the generic files as vibe in its own eval marketplace, with no grader change. The ClosedLoop build keeps the name vibe.
  • Realistic arm. "Yes, install the built folder (Recommended)". This amends the 11-13 wording "no pre-assembly" for a committed folder that CI checks for drift.
  • Recipes. In a repository without the opt-in file, "Inline as on main (Recommended)".
  • Earlier rulings, still binding. ClosedLoop steps stay out of generic vibe. A ClosedLoop session matches main with no extra model-chosen hops on the hot paths. Cold sections live in separate files, one hop. Nothing writes into an installed plugin at runtime.

What ships. Both entries are in closedloop-ai (Claude and Codex catalogs):

  • vibe: plugins/vibe-closedloop/, a generated folder with manifest name vibe. Every existing vibe@closedloop-ai install updates into it.
  • vibe-generic: plugins/vibe/, with manifest name vibe-generic in both harnesses. Codex refuses an entry whose name differs from its manifest's ("plugin.json name vibe does not match marketplace plugin name vibe-generic", codex-cli 0.167.2). Real generic users see vibe-generic:* skill names. A bare $vibe still resolves (step 0).
  • vibe-closedloop-provider leaves both catalogs. Its folder is build input only.

The build (tools/vibe-closedloop-build, npm run build):

  • Inputs. It calls provider.mjs assembleProviderText with every declared capability. There is one assembler; the Python twin skill_split.assembled is deleted.
  • Hot text. Hot blocks go inline, and the stubs and generic regions are dropped.
  • Cold text. The deploy and environment sections of vibe SKILL.md go to skills/vibe/references/closedloop-deploy.md and closedloop-environment.md, each reached by one provider-written line.
  • Recipes. They stay inline, under one provider-written line naming .closedloop-ai/vibe-recipes.json.
  • Paths and names. The provider's whole-file references (ticket-template.md, closedloop-graph.md, ticket-units.md) are placed at main's paths (skills/vibe/references/, skills/fleet-orchestrator/references/), and provider:<file> names become main's relative forms. provider/ holds only provider.json; its placed map is what providerPath resolves through. Block sources are not shipped. A build test requires every Markdown file reference in the built tree to resolve. The exceptions that also fail on main (target-repository names, other plugins' files) are listed with their reasons. vibe-generic:, vibe-generic@ and /vibe-generic/ become main's vibe forms.
  • Descriptions. Manifest and frontmatter descriptions are main's, kept in plugins/vibe-closedloop-provider/build.json.
  • Left out. Tests, evals and fixtures are not copied.
  • Determinism. Output is deterministic. Byte-identical copies share git blobs with plugins/vibe.
  • Folder location. Under plugins/, so the version-bump job and the manifest tests cover it with no new wiring. The AC1 scan's plugins/vibe/ prefix does not match it, and the scan is unchanged.
  • CI. A tools-ts matrix row rebuilds and fails on any diff or untracked file.

Discovery and binding.

  • Discovery reads only the contract bundled at <vibe>/provider/provider.json; the sibling-plugin scan is gone.
  • Kept: per-capability binding, connection states, the opt-in check, the check exiting 0 with its reason, and the setup block.

Retired.

  • The provider-rule paragraph, reduced to its two generic sentences. The build drops even those.
  • references/provider-text.md.
  • The bind-time text folder and tree.
  • The model-run assemble, block and path commands.
  • providerTextPolicy.
  • The install offer, provider-install and provider-match.
  • The skill-listed flag.
  • The provider skill.
  • The harness-side assembly in run.py.

Where fixes go. This rule is added to CLAUDE.md (Claude) and AGENTS.md (Codex):

  • Generic behavior goes in plugins/vibe, and vibe_generic_scan.py check gates it.
  • ClosedLoop text and data go in plugins/vibe-closedloop-provider.
  • plugins/vibe-closedloop is build output and never hand-edited; CI's TypeScript (vibe-closedloop-build) drift check enforces that.

Plan-folder privacy.

  • knownProviderContracts lists the provider bundled in every vibe install or checkout, enabled or not. local-plans uses it, so a plan written while bound (<artifactRoot>vibe-plans/) stays private evidence after the session is unbound, whichever vibe the lookup order picks.
  • Limit: a machine with only vibe-generic installed has no provider contract, so it knows no ClosedLoop plan folder.
  • Deliberate expectation change: local-plans.test.mjs expected knownLocalPlanPrefixes(null) to be empty in this checkout. It now expects .closedloop-ai/vibe-plans/, because the built ClosedLoop vibe sits beside the source.

Name and lookup.

  • One shared module (plugin-roots.mjs, vibe-core canonical, vibe mirror) holds DEFAULT_MARKETPLACE and the two plugin names. No hook keeps its own copy.
  • Each install reads its own name and marketplace from its install path.
  • Checkouts look up vibe-generic first. Installs look up vibe first, so with both installed the ClosedLoop vibe runs the hooks and binds. The check reports conflict: "both-installed" (a code, no new sentence).
  • A new catalog-names check fails when a catalog entry name differs from its manifest name.

Compatibility shims kept (an installed older client can still call these):

  • provider.mjs tree|assemble|block|path print one provider-broken: ... is retired line and exit 4.
  • plugin-version.mjs provider|provider-install|provider-match answer {"ok":false,"retired":...}.
  • --skill-listed and --expect are accepted and ignored.
  • --closedloop/--graph remain aliases.
  • liveness.mjs re-exports isEntryPoint.

Eval arms.

  • provider-installed (the realistic arm) installs the committed plugins/vibe-closedloop as vibe unchanged. It writes symphony-alpha's opt-in file ({ "schema": 1, "provider": "vibe-closedloop-provider" }, as committed at abe87d83) into the run's copy of the case folder before the real check.
  • The generic arm installs plugins/vibe as vibe, renaming only the snapshot's manifest, as main's baseline does.
  • No grader, fixture or case changed. The four generic cases' graders name no plugin-prefixed skill.
  • The realistic arm runs with recipes withheld: run.py's case folders are not git checkouts, so the opt-in file is present but unused (provider.mjs finds the repository by walking up to .git). The built text has recipes inline either way, so the arm compares like for like. Not built: having run.py git init the run's copy of each provider-arm case folder before the bind. It is only revisited if Step 1 shows a failure caused by withheld recipes, because it changes what the model sees.

Harness fixes.

  • run.py resolves its private homes to real paths.
  • 102 main guards in vibe and vibe-core call one isEntryPoint that compares real paths.
  • Known copies of the same guard left alone: tools/shared/cli.ts runWhenMain (bundled into other plugins), the closedloop-core gh-monitor-pr scripts, tested-head.mjs (runs from a lone copy), and the eval checker iss-13760-reply-format.mjs.

Generic noise, for reading the generic arm.

  • The eight root files a generic session reads (vibe, handoff and fleet SKILL.md, preflight, quality-loop, guardrails, guardrails-reviewer, storybook-decomposer) carry 126 stub lines and 236 marker lines out of 4,437.
  • All of plugins/vibe Markdown carries 299 stubs and 602 marker lines.

Merged with main

At this head (regular merges, no rebase):

Earlier merges:

origin/main 8a858bb is merged in (merge commit 72a2a7c). It brought #429 (ISS-14109: overridable model defaults) and #430 (ISS-14077: targeted-tests rulings in the handoff gate, cloud-import PR titles). Main's behaviour is kept; this PR's provider extraction is applied on top where both touched the same lines. 18 files conflicted:

  • CHANGELOG.md: main's vibe 1.1.86 and vibe-core 1.0.9 entries kept; this PR's entries renumbered to vibe 1.1.88, vibe-core 1.0.10 and vibe-closedloop-provider 1.0.0 (1.1.87 is held by ISS-14112's open work), word counts remeasured against 8a858bb.
  • vibe and vibe-core plugin.json (Claude and Codex), test_vibe_ws8_load_contract.py, test_pr_takeover_skill_contract.py, reply-checker.test.ts: version numbers only, now 1.1.88 and 1.0.10.
  • plugins/vibe/README.md: main's model-defaults text kept inside this PR's generic README; version line 1.1.88.
  • review-models.mjs, claude-worker.ts: main's model resolver and model/reviewModel launch fields kept as written, with this PR's binding checks (sessionBinding, requireSupportedCapabilities, brief inheritance, recordModes(binding)) around them; ticket slugs dropped from the comments. The dist bundle was rebuilt, not hand-merged.
  • handoff-gate.mjs, targeted-tests-ruling.mjs: main's ruling logic kept; this PR's exitOnProviderBroken added; slugs dropped from comments.
  • pr-facts.mjs: judgment. Main added a ruling id parameter to selectionProblems; this PR had added the provider's PR rules parameter to the same function. Both are kept: selectionProblems(runsFile, files, requiredTests, commit, rules = GENERIC_PR_RULES, rulingId = null), and the ruled-test check passes rules to e2eCarriedForCi.
  • pr-facts.test.mjs, vibe-triage.test.mjs: main's new ruling and title cases kept, run under this PR's bound-provider fixtures.
  • plugins/vibe/agents/vibe-verify-worker.md and vibe-environment-worker.md: judgment. This PR had moved the sections main edited into the provider's copies of those agents, so main's targeted-tests text (the verify worker's ruling paragraph, --rulings on the gate and evidence commands) went into vibe-closedloop-provider/.../references/agents/ only. Operator ruling (Daniel, 2026-11-13): this text stays in the provider copies only ("only closedloop sessions will have access to closedloop-graph"). A session with no provider bound keeps the existing fallback (vibe-verify-worker.md lines 86 to 97): the tests related to each changed file, through the repository's documented command.

Checks on 72a2a7c, through the local heavy slot: ruff, pyright, plugin manifests, vibe-bench, the whole Python lane (2806 passed, 3 skipped), the 53 node test files the merge touched (637/637), vibe-workers typecheck and build with the committed bundle matching, claude-worker.test.ts 69/69, the changelog word count. vibe-workers' full vitest run had 37 timeouts at load average about 57 and no assertion failure; the 5 in claude-worker.test.ts (merge-touched) passed on rerun, and the 32 in main-sync* and scope-intake files the merge did not touch are left to CI. The ISS-14077 ruling tests pass with the provider bound; with no provider bound, the 9 that need the provider's tracker ticket or lanes fail as expected, as the ruling above intends.

A bound session behaves like main (previous head f1d1e02; superseded by Release-time build)

Operator rulings (Daniel, 2026-11-13): #426 cannot merge while it fails an eval that passes on main, and the fix goes in the PR, never in the evals (no grader, fixture or case edits). The bar: every case main passes, passes with the provider bound, at main's rate. Design chosen: fix how a bound session reads the provider's text at runtime, with a realistic eval arm as the merge measure.

The model evals and four reviews found that a real bound session did not read what the eval's provider arm read. run.py pre-assembled every vibe file with every capability bound, while a real session read the raw generic text and was told to run provider.mjs assemble per file, which writes under the home folder and stops on a refused write. The setup-check case also failed because the start opened with plugin-version.mjs provider and the old --closedloop/--graph flags had become an unknown option. What changed:

  • Start: when the listed provider skill's description names its start calls (its first sentence, under 140 characters, pinned by a test), the session makes them and runs one plugin-version.mjs check with --closedloop/--graph, accepted again as aliases of --connection, as on main. Otherwise plugin-version.mjs provider runs first; it now writes nothing. The skill list never decides "no provider": a check with no connection states while a provider is installed records unprobed, names the calls, binds nothing, and every provider read and vibe-sessions.mjs new/adopt stop until it runs again with the states. In Claude Code the provider counts as loaded when its skill name is in the list.
  • check always exits 0, prints the setup block in every state (a broken provider's reason and fix included, and with no provider at all), and carries the install offer. Fixed on the way: plugin-version.mjs provider threw whenever an install offer stood (a constant read before its declaration).
  • Per-capability binding: a capability binds on its own needs. With the graph down, tickets, plans, memory and recipes stay bound, the graph is listed in missing, and the setup worker restores it, as main did (8a858bb SKILL.md:416-428). A stub naming several capabilities is served when any of them is bound, so its ClosedLoop text stays; graph-only stubs keep their generic text until the graph is back. Before this, one unanswered connection made the whole session generic (launcher grants, live-ticket slugs, PR evidence rules, the unit-status and e2e-slot guard rules).
  • Bound text is written once, at bind time: the escalated check writes every stubbed vibe file assembled for its binding, a blocks file per stubbed SKILL.md and binding.json (repo root, provider, versions, vibe root, capabilities) to ~/.codex/vibe/provider-text/<key>/, keyed by the repository's main checkout (or the working folder outside a checkout), replaced whole, with index.json naming each folder. A reader opens files there, by provider.mjs tree or by reading index.json, only when binding.json names its repository and its vibe install; otherwise it runs the check again, and assemble is the last resort. Codex's default and workspace-write sandboxes can read that folder (checked with codex sandbox on codex-cli 0.167.2); only the escalated check writes it. In a bound session an exit 3 from assemble, block or path is a stop, not a generic fallback.
  • Precedent comes only from the live app in the operator's signed-in browser, never public help centers or documentation, in every session. This restores main (8a858bb precedent-check.mjs:31-32, :58, :158; vibe-requirements-worker.md:81).
  • A writer brief with no Provider: line inherits the lead's binding instead of being refused (main refused nothing here); only a line naming another binding is refused.
  • With the provider bound, vibe-preflight.sh finds the checkout exactly as main did (8a858bb vibe-preflight.sh:205-238): --repo (another repository is refused, a claude-plugins checkout is used for that run only), the remembered checkout, then the home-folder search, with choose-repo for several clones. The remotes come from provider data. A session with no provider also accepts the checkout it runs in. No preflight difference from main remains.
  • vibe-bench sessions bind the provider through the real check (--skill-listed); its old flags work again.
  • Eval runner: new arm provider-installed copies the raw vibe text, installs the provider and runs the real check in each case folder before the model starts (HOME set to the run's own folder, the release read from a loopback copy of the eval marketplace). The provider arm still pre-assembles every capability, so it never exercises a partial binding; the new arm is the merge measure. The old arms are kept.

Breaking changes

  • vibe@closedloop-ai now installs the built ClosedLoop folder (plugins/vibe-closedloop); ClosedLoop sessions read main's text. Generic vibe is a new plugin, vibe-generic@closedloop-ai, whose skills are vibe-generic:*. There is no separate provider plugin to install or disable.
  • Retired commands answer with a retired refusal instead of running: provider.mjs tree|assemble|block|path (exit 4) and plugin-version.mjs provider|provider-install|provider-match; --skill-listed and --expect are ignored.
  • board.mjs refuses writes to an orchestration directory that is not a program in a fleet home (fleet-home.mjs migrate moves one). Session records gain an optional provider field; environmentOrgId is written beside the earlier clerkOrgId, which is still read and written.

Test plan

This head (e316b80). The merges of #432, #436, #434 and #435 and the layout fix are verified on laptop 2's heavy slot unless noted:

  • Passed:
    • lint and pyright;
    • the build tool (references test included) and the drift check;
    • plugin manifests and pytest plugins/ (at 5d4a82ac);
    • node tests the merges touch (fleet, capture-guard, cloud-worker, provider, local-plans, instruction-words: 143/0);
    • vibe-workers dist freshness;
    • main-sync-ci and main-sync-report-admission;
    • the bench, apart from the known phase3 quarantine case.
  • Python contracts the merges touch, plus test_run.py: 153 passed (local heavy slot).
  • vibe-workers timeouts, not failures: on laptop 2, main-sync-input-boundary (4), main-sync (8) and scope-intake (1) failed only with Test timed out and spawn ETIMEDOUT, even run alone. Load was 18 to 24, with another fleet holding the heavy slot. tools/vibe-workers/src is unchanged since 22c726c, whose CI vibe-workers job passed; main-sync-input-boundary passed 63/63 locally. Left to CI.
  • Not rerun: the full Node Tests job. At laptop 2's load, its e2e-lock.test.mjs children hung for over 2 hours, the same hang recorded for earlier heads. CI runs it.

At 22c726c: 73ad772's CI had two red tests, both fixed:

  • the CHANGELOG word-count phrase pin (the entry is regenerated through /update-documentation);
  • main-sync-report-closure.test.ts plan-folder privacy (the behavior is restored and the expectation unchanged).
  • Every CI job, run in full locally the CI way through the heavy slot (ci.yml has no path filters, so all of them run on every PR):
    • Lint, Type Check, Plugin Manifest Versions: green.
    • Tests (pytest plugins/): 2822 passed, 3 skipped.
    • Node Tests (skill scripts): 2186 passed, 0 failed, 1 skipped.
    • TypeScript (vibe-workers): 863/863, with fresh dist bundles.
    • TypeScript (design-inventory, code-review-cost, guided-manual-qa, erd-diagram, vibe-closedloop-build): green, with no dist drift.
    • Vibe Eval T0: typecheck, validate:plugins and test_run.py 67/67 pass; one bench test fails, the known phase3 case below.
  • New for the privacy fix: two provider tests and one local-plans assertion; the mutation proof fails all of them, plus the closure test.

At 73ad772: targeted tests through the local heavy slot.

  • Every behavior change has a test that fails on f1d1e02. Mutation proofs: build 15/15, retirement 14/14, rule restore 3/3, naming 31/31, eval arms 7/7, main guards 4/4. Two retirement proofs edit SOURCE text only: a word in the ticket-worker provider block fails test_vibe_forbidden_path_routing_contract, and a word in plugins/vibe SKILL.md fails test_vibe_connection_restore_contract.
  • Final runs: build tool 13/13 and drift clean; plugin-manifests 16/16 (new catalog-names check); test_vibe_generic_scan 35/35, unchanged; test_run.py 67/67; ruff and pyright clean. The lanes ran every vibe and vibe-core node test file (194) and Python test file (71), one at a time.
  • Step 0, no evals, in scratch HOME, CLAUDE_CONFIG_DIR and CODEX_HOME (codex-cli 0.167.2, Claude Code 2.1.323):
    • vibe alone binds all 5 capabilities from its bundled provider in both harnesses.
    • vibe-generic alone binds none and lists vibe-generic:*.
    • Both installed: both are enabled, the check reports both-installed, and ClosedLoop binds.
    • A bare $vibe resolves to vibe-generic:vibe (one codex exec).
    • Not checked: which skill Claude's model picks with both enabled.
  • Environmental, recorded and not rerun: vibe-bench phase3 fails the same way on origin/main 8a858bb (expired 2026-11-13 eval quarantine deadlines).
  • Not run: the realistic and generic eval arms; team-lead runs them on this head.

Earlier heads

  • Full lanes, because the diff is too broad to target: provider.mjs (new, so not in the test-selection index) is imported by nearly every vibe and vibe-core test. code_tests_for_changes on the last round's 61 files selected 102 unit test files (hooks, fleet, handoff and vibe scripts, claude-worker); the full lanes cover them all.
  • Local, the way ci.yml runs them (harness env stripped, fixture HOME): ruff, pyright and pytest (2806) pass. Node Tests: 201 files, 2074 tests, 0 failed. Plugin Manifest Versions, Vibe Eval T0 (vibe-bench typecheck, validate:plugins, tests, Codex test_run.py) and vibe-workers typecheck, vitest and dist freshness all pass.
  • e2e-lock.test.mjs hung once under a loaded full run (load average about 40 from other sessions). Its signal cases then passed 3 of 3 in isolation, and its one change (the top-level catch calls failScript) is not on the signal paths. It passed in the final full run.
  • vibe-workers main-sync.test.ts and main-sync-input-boundary.test.ts timed out (16 to 44 s against 15 s) at load average 37 to 45. They passed 122/122 on one sequential rerun.
  • The 26 cl-sweep tests that failed locally failed only because Claude Code's env vars (CLAUDECODE, CLAUDE_CODE_SESSION_ID) leaked into the lane. Run the CI way, 69/69 pass.
  • New tests: assemble output equals assembleProviderText (none, each capability, all); --blocks order and labels; exit 2, 3 and 4; an unreadable record is broken; a script exits 4 (local-plans); a guard refuses while generic checks still run, and grep/cat of e2e paths pass (pre-tool-use); a hook warns and still runs (program-change); PROVIDER_BROKEN mapping and CLI exit 4 (claude-worker).
  • Reviews: two adversarial reviews of the full diff, plus a Fable 5.2 weight review and a Fable 5.2 adversarial review of the final round. Every finding was fixed or is listed above.
  • Cloud lanes not run (targeted-tests ruling; CI on this PR is the Linux gate). Plugin Version Bump checked locally.
  • After merging main (ISS-14039: Accept a Fable review for a Claude cloud import #427): ruff, pyright, vibe python pytest, the node tests of the files ISS-14039: Accept a Fable review for a Claude cloud import #427 touched plus the provider and version tests (135/135), claude-worker vitest (67/67) with dist rebuilt, manifests and vibe-bench all pass.
  • After merging main 7c34119 (ISS-14028, ISS-14024: Bind fleet chats to their programs, safe migrate #428): ruff, pyright and pytest pass; Node Tests the CI way 201 files, 2123 tests, 0 failed; manifests and vibe-bench (54 passed) pass through the local heavy slot. vibe-workers through the heavy slot: 843 passed and 18 timed out (16 Test timed out at 5 s, 2 spawnSync node ETIMEDOUT), with no assertion failure: claude-worker.test.ts 8, main-sync.test.ts 7, main-sync-ci, main-sync-input-references and main-sync-report-admission 1 each, at load average 50 to 65 on 10 cores from other sessions. After the ISS-14039: Accept a Fable review for a Claude cloud import #427 merge the same claude-worker file passed 67/67 and main-sync 122/122 in a sequential run; they are left to CI on clean runners rather than rerun here.
  • Bound-session fix (f1d1e02, tree 455739f1), every lane on laptop 2 through its heavy slot, on exactly the committed tree: ruff, pyright, plugin manifests, release-smoke node tests, test_run.py, vibe-bench (typecheck, validate, tests), vibe-workers (typecheck, vitest 860/863, committed bundle fresh), the Node Tests lane the CI way (201 files, 2143 pass, 0 fail, 1 skipped), claude-worker.test.ts plus main-sync.test.ts alone 128/128 (their 3 timeouts in the full vitest run were at load), the whole Python lane (2817 passed, 3 skipped), the changelog word count and instruction-words.test.mjs. Every new behavior has a test that failed when the change it covers was reverted.
  • Not rerun: e2e-lock.test.mjs hung in the full node run at laptop 2's load (its e2e-lock.mjs run children never finished). This fix does not touch e2e-lock or anything it imports, and the same hang under load is recorded above at an earlier head; CI runs it. The node lane above is every other file.
  • CHANGELOG and READMEs follow /update-documentation (the vibe README's eval count 121 to 125; the vibe-core and provider READMEs were accurate).

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.
shafty023 added a commit that referenced this pull request Nov 12, 2026
Merge origin/main (ISS-14039, #427) and take vibe 1.1.84 and vibe-core 1.0.7, past the numbers #427 used and #426 has reserved. VIBE_CORE_MIN, the README version line, the release-pin tests and the CHANGELOG entries follow, and the vibe-workers bundles are rebuilt.
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.
…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.
…-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.
shafty023 added a commit that referenced this pull request Nov 17, 2026
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.
shafty023 added a commit that referenced this pull request Nov 17, 2026
Merge feat/iss-13446-generic-vibe (e316b80) per the stack-on-#426 ruling.
The built plugin keeps evals/codex/model-policy.json, the one default source cl-sweep reads; plugins/vibe-closedloop rebuilt.
worker-model AC-1 copies every module its code imports.
The built ClosedLoop vibe named no connection in its Plugin version command,
so sessions invented --connection names and lost the setup check eval. A
provider block now carries main's text, naming --closedloop and --graph;
generic vibe keeps its --connection form.
@shafty023
shafty023 merged commit 7d5bacf into main Nov 17, 2026
12 of 13 checks passed
@shafty023
shafty023 deleted the feat/iss-13446-generic-vibe branch November 17, 2026 17:24
shafty023 added a commit that referenced this pull request Nov 17, 2026
Brings in main at 7d5bacf (#426 squashed). Only .gitattributes is new; every other #426 file already matches this branch.
shafty023 added a commit that referenced this pull request Nov 18, 2026
Merge origin/main 7d5bacf, the squash of #426, which this branch already contained up to e316b80.
Each file is this branch's content plus main's changes since e316b80; plugins/vibe-closedloop rebuilt.
Versions unchanged: vibe 1.1.98, provider 1.0.1.
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