cool-workflow 0.1.83 → 0.1.85

package/docs/multi-agent-operator-ux.7.md

@@ -325,3 +325,9 @@ _No changes in v0.1.82._
 ## Hardening and Onboarding (v0.1.83)
 
 Loaders fail closed on corrupt state; store writes are made safe under more than one writer; a new cw doctor checks your setup; help lists every command; and the docs are put into Basic English.
+
+## Privacy Release (v0.1.84)
+
+No other change to this page in v0.1.84.
+
+0.1.85

package/docs/node-snapshot-diff-replay.7.md

@@ -146,3 +146,9 @@ _No change in behavior in v0.1.82 (the node projection field set was brought tog
 ## Hardening and Onboarding (v0.1.83)
 
 Loaders fail closed on corrupt state; store writes are made safe under more than one writer; a new cw doctor checks your setup; help lists every command; and the docs are put into Basic English.
+
+## Privacy Release (v0.1.84)
+
+No other change to this page in v0.1.84.
+
+0.1.85

package/docs/observability-cost-accounting.7.md

@@ -205,3 +205,9 @@ _No changes in v0.1.82._
 ## Hardening and Onboarding (v0.1.83)
 
 Loaders fail closed on corrupt state; store writes are made safe under more than one writer; a new cw doctor checks your setup; help lists every command; and the docs are put into Basic English.
+
+## Privacy Release (v0.1.84)
+
+No other change to this page in v0.1.84.
+
+0.1.85

package/docs/project-index.md

@@ -1,16 +1,16 @@
 # Cool Workflow Project Index
 
-Generated from the current repository code on 2026-06-16 by `npm run sync:project-index`.
+Generated from the current repository code on 2026-06-18 by `npm run sync:project-index`.
 
 ## Snapshot
 
 - Package: `cool-workflow`
-- Version: `0.1.83`
-- Source modules: `61`
+- Version: `0.1.85`
+- Source modules: `62`
 - Workflow apps: `7`
-- Docs: `49`
-- Smoke tests: `92`
-- Repository: http://local_proxy@127.0.0.1:33273/git/coo1white/cool-workflow
+- Docs: `50`
+- Smoke tests: `101`
+- Repository: https://github.com/coo1white/cool-workflow
 
 ## Architecture
 
@@ -92,6 +92,7 @@ multi-agent host -> topology -> blackboard/coordinator
 - [evidence-reasoning.ts](../src/evidence-reasoning.ts)
 - [execution-backend.ts](../src/execution-backend.ts)
 - [gates.ts](../src/gates.ts)
+- [mcp-surface.ts](../src/mcp-surface.ts)
 - [multi-agent-eval.ts](../src/multi-agent-eval.ts)
 - [multi-agent-operator-ux.ts](../src/multi-agent-operator-ux.ts)
 - [multi-agent-trust.ts](../src/multi-agent-trust.ts)
@@ -160,6 +161,7 @@ multi-agent host -> topology -> blackboard/coordinator
 - [Real Execution Backend Integrations](real-execution-backends.7.md)
 - [Release And Migration Discipline](release-and-migration.7.md)
 - [Release Tooling](release-tooling.7.md)
+- [Verifiable Report Bundle](report-verifiable-bundle.7.md)
 - [Routines](routines.md)
 - [Run Registry / Control Plane](run-registry-control-plane.7.md)
 - [Run Retention & Provable Reclamation](run-retention-reclamation.7.md)
@@ -189,9 +191,11 @@ Smoke tests mirror the public contracts. The high-signal suites are:
 - [audit-verify-smoke.js](../test/audit-verify-smoke.js)
 - [backend-registry-smoke.js](../test/backend-registry-smoke.js)
 - [block-unapproved-tag-smoke.js](../test/block-unapproved-tag-smoke.js)
+- [bump-version-idempotent-smoke.js](../test/bump-version-idempotent-smoke.js)
 - [candidate-scoring-smoke.js](../test/candidate-scoring-smoke.js)
 - [canonical-workflow-apps-smoke.js](../test/canonical-workflow-apps-smoke.js)
 - [claude-p-agent-wrapper-smoke.js](../test/claude-p-agent-wrapper-smoke.js)
+- [cli-command-surface-smoke.js](../test/cli-command-surface-smoke.js)
 - [cli-jsonmode-parity-smoke.js](../test/cli-jsonmode-parity-smoke.js)
 - [cli-mcp-parity-smoke.js](../test/cli-mcp-parity-smoke.js)
 - [concurrent-failure-semantics-smoke.js](../test/concurrent-failure-semantics-smoke.js)
@@ -199,6 +203,7 @@ Smoke tests mirror the public contracts. The high-signal suites are:
 - [contract-migration-tooling-smoke.js](../test/contract-migration-tooling-smoke.js)
 - [control-plane-scheduling-smoke.js](../test/control-plane-scheduling-smoke.js)
 - [coordinator-blackboard-smoke.js](../test/coordinator-blackboard-smoke.js)
+- [demo-bundle-smoke.js](../test/demo-bundle-smoke.js)
 - [det-ids-b-smoke.js](../test/det-ids-b-smoke.js)
 - [doctor-smoke.js](../test/doctor-smoke.js)
 - [dogfood-release-smoke.js](../test/dogfood-release-smoke.js)
@@ -213,6 +218,7 @@ Smoke tests mirror the public contracts. The high-signal suites are:
 - [freebsd-audit-fixes-smoke.js](../test/freebsd-audit-fixes-smoke.js)
 - [h7-custom-profile-persist-smoke.js](../test/h7-custom-profile-persist-smoke.js)
 - [mcp-app-surface-smoke.js](../test/mcp-app-surface-smoke.js)
+- [mcp-surface-registry-smoke.js](../test/mcp-surface-registry-smoke.js)
 - [multi-agent-cli-mcp-surface-smoke.js](../test/multi-agent-cli-mcp-surface-smoke.js)
 - [multi-agent-eval-determinism-regression-smoke.js](../test/multi-agent-eval-determinism-regression-smoke.js)
 - [multi-agent-eval-replay-harness-smoke.js](../test/multi-agent-eval-replay-harness-smoke.js)
@@ -228,15 +234,19 @@ Smoke tests mirror the public contracts. The high-signal suites are:
 - [operator-ux-smoke.js](../test/operator-ux-smoke.js)
 - [parallel-onramp-smoke.js](../test/parallel-onramp-smoke.js)
 - [parity-doc-sync-smoke.js](../test/parity-doc-sync-smoke.js)
+- [pii-redaction-smoke.js](../test/pii-redaction-smoke.js)
 - [pipeline-auto-advance-smoke.js](../test/pipeline-auto-advance-smoke.js)
 - [pipeline-runner-smoke.js](../test/pipeline-runner-smoke.js)
 - [project-index-sync-smoke.js](../test/project-index-sync-smoke.js)
+- [quickstart-bundle-smoke.js](../test/quickstart-bundle-smoke.js)
 - [quickstart-smoke.js](../test/quickstart-smoke.js)
 - [real-execution-backends-smoke.js](../test/real-execution-backends-smoke.js)
 - [registry-corrupt-fail-closed-smoke.js](../test/registry-corrupt-fail-closed-smoke.js)
 - [release-flow-smoke.js](../test/release-flow-smoke.js)
 - [release-gate-smoke.js](../test/release-gate-smoke.js)
 - [release-tooling-smoke.js](../test/release-tooling-smoke.js)
+- [report-bundle-smoke.js](../test/report-bundle-smoke.js)
+- [report-verify-bundle-smoke.js](../test/report-verify-bundle-smoke.js)
 - [result-normalize-smoke.js](../test/result-normalize-smoke.js)
 - [robustness-failclosed-smoke.js](../test/robustness-failclosed-smoke.js)
 - [robustness-hardening-smoke.js](../test/robustness-hardening-smoke.js)
@@ -271,6 +281,7 @@ Smoke tests mirror the public contracts. The high-signal suites are:
 - [verifier-gated-commit-smoke.js](../test/verifier-gated-commit-smoke.js)
 - [verify-import-audit-chain-smoke.js](../test/verify-import-audit-chain-smoke.js)
 - [web-desktop-workbench-smoke.js](../test/web-desktop-workbench-smoke.js)
+- [worker-accept-path-architecture-smoke.js](../test/worker-accept-path-architecture-smoke.js)
 - [worker-isolation-smoke.js](../test/worker-isolation-smoke.js)
 - [worker-retry-count-smoke.js](../test/worker-retry-count-smoke.js)
 - [workflow-app-framework-smoke.js](../test/workflow-app-framework-smoke.js)

package/docs/real-execution-backends.7.md

@@ -153,3 +153,9 @@ _No behavioral change in v0.1.82 (delegated child programs moved out to `scripts
 ## Hardening and Onboarding (v0.1.83)
 
 Loaders fail closed on corrupt state; store writes are made safe under more than one writer; a new cw doctor checks your setup; help lists every command; and the docs are put into Basic English.
+
+## Privacy Release (v0.1.84)
+
+No other change to this page in v0.1.84.
+
+0.1.85

package/docs/release-and-migration.7.md

@@ -293,3 +293,9 @@ _No changes in v0.1.82._
 ## Hardening and Onboarding (v0.1.83)
 
 Loaders fail closed on corrupt state; store writes are made safe under more than one writer; a new cw doctor checks your setup; help lists every command; and the docs are put into Basic English.
+
+## Privacy Release (v0.1.84)
+
+No other change to this page in v0.1.84.
+
+0.1.85

package/docs/release-tooling.7.md

@@ -143,12 +143,21 @@ The per-platform difference is config, not code — set the reviewer agent here:
 
 | Platform | Reviewer config |
 |---|---|
-| Claude    | `CW_AGENT_COMMAND="claude -p {{input}}"` |
+| Claude    | `CW_AGENT_COMMAND="claude -p --permission-mode acceptEdits {{input}}"` |
 | Codex     | `CW_AGENT_COMMAND="codex exec {{input}}"` |
 | Gemini    | `CW_AGENT_COMMAND="gemini -p {{input}}"` |
 | OpenCode  | `CW_AGENT_COMMAND="opencode run -m <provider/model> {{input}}"` |
 | DeepSeek  | via OpenCode (`-m deepseek/deepseek-chat`) or `CW_AGENT_ENDPOINT=<deepseek-compatible HTTP agent>` |
 
+The reviewer's last act is to **write** the verdict file, so a headless agent
+needs file-write permission. For `claude -p`, recent CLIs default to a mode that
+silently denies `Write` (no prompt in headless mode), so the reviewer reaches a
+verdict but cannot persist it and the flow fails closed at `[3/3] verify` with
+`no verdict written`. The `--permission-mode acceptEdits` flag in the preset above
+fixes this; Read/Bash (which the review itself needs) keep working. The reviewer
+CLI must also be logged in (`claude auth login` / `claude auth status` →
+`loggedIn: true`) — a fresh shell or CI runner is not.
+
 `{{input}}` is put in place of the reviewer prompt file path. Gemini and OpenCode
 also get generated MCP manifests (`.gemini-plugin/`, `.opencode-plugin/`) so the
 `cw_*` tools are there as MCP tools in those hosts. The verdict path
@@ -200,3 +209,9 @@ _No changes to the release-tooling contract in v0.1.82._
 ## Hardening and Onboarding (v0.1.83)
 
 Loaders fail closed on corrupt state; store writes are made safe under more than one writer; a new cw doctor checks your setup; help lists every command; and the docs are put into Basic English.
+
+## Privacy Release (v0.1.84)
+
+`release-flow` now writes reviewer input with repo-local paths, so local user home names do not enter the saved review prompt.
+
+0.1.85

package/docs/report-verifiable-bundle.7.md

@@ -0,0 +1,123 @@
+# Verifiable Report Bundle
+
+CW v0.1.85 adds the Verifiable Report Bundle: a way to hand someone a run's report
+that they can check themselves, OFFLINE, with one command — no source repo, no
+pre-existing `.cw` tree, and no key passed on the side.
+
+Before this, a run could be exported to a portable `<id>.cwrun.json` archive
+([Run Registry / Control Plane](run-registry-control-plane.7.md)) and its bytes
+re-proven with `cw run inspect-archive`. But two things stopped a stranger from
+fully verifying it:
+
+- The ed25519 PUBLIC key that re-checks the signed telemetry came only from
+  `--pubkey` or the `CW_AGENT_ATTEST_PUBKEY` environment variable — it did not
+  travel with the archive.
+- `inspect-archive` proves the archive's own file digests, not the telemetry hash
+  chain or the signatures. Re-proving those meant `cw run import` into a real
+  `.cw` tree and then `cw run verify-import` — too much to ask of a recipient.
+
+The Verifiable Report Bundle closes both gaps without a new archive format.
+
+## Mechanism vs Policy
+
+The MECHANISM is three small additions to the existing export/verify path:
+
+1. `cw run export <run> --with-trust-key <pem-or-path>` embeds the operator's
+   ed25519 PUBLIC key into the archive under a new optional `trust` block
+   (`{ publicKeyPem, algorithm: "ed25519" }`). Only a public key is ever embedded;
+   CW never exports a private key. The flag defaults to `CW_AGENT_ATTEST_PUBKEY`,
+   so one configured key both attests at record-time and travels with the export.
+   A raw inline PEM begins with `-----`, which the CLI parses as a flag — so the
+   CLI form takes a key FILE PATH; an inline PEM is accepted programmatically and
+   via the environment variable.
+
+2. `cw report verify-bundle <bundle>` verifies the bundle SELF-CONTAINED and
+   OFFLINE. It reuses the existing verifiers end to end: `inspectArchive` for the
+   archive bytes, then a restore into a throwaway temp dir (auto-removed) so
+   `verifyImportedRun` re-proves the telemetry hash chain and the trust-audit
+   chain, then `verifyTelemetrySignatures` re-runs ed25519 over each attested hop
+   using the key the bundle carries. It writes nothing to any registry.
+
+3. `cw report bundle <run>` is the PRODUCER counterpart: it exports a sealed
+   bundle (step 1) and then immediately self-verifies it (step 2), returning the
+   archive path and the verification verdict together. It fails closed — a solo
+   operator never hands off a report whose bundle does not verify (for example, no
+   trust key configured under `--strict-signatures`). `--extract-report` also
+   writes the human-readable `report.md` next to the bundle so the shippable pair
+   is produced in one command. It is pure composition; it spawns nothing.
+
+`cw quickstart <app> --bundle [--with-trust-key K]` folds step 3 into the
+one-command entry: after the drive reaches `status: complete`, it seals the run
+with `reportBundle` (anchored to the run's own repo) and returns the verdict as
+`result.bundle`. It is gated on completion — a blocked/partial run is **never**
+sealed (the operator gets a `hint`, not a half-shipped artifact) — and the CLI
+exits non-zero when `result.bundle.ok` is false. `quickstart` stays `cli-only`;
+MCP hosts compose `cw_run_drive_step` + `cw_report_bundle` for the same outcome.
+
+Because the README headline runs quickstart cross-directory (your shell cwd is not
+the `--repo`), the run is resolved from its own repo, but the bundle's OUTPUT —
+the `.cwrun.json` and any `--extract-report` — lands in **your cwd**, not the
+analyzed repo (so `cw quickstart … --bundle --extract-report out.md && send out.md`
+works and the repo's working tree is never polluted). `result.bundle.archivePath`
+and `result.bundle.reportExtractedTo` report the absolute paths.
+
+The POLICY is fail-closed and self-describing:
+
+- Key precedence is **bundle > `--pubkey` > `CW_AGENT_ATTEST_PUBKEY`**, so a bundle
+  with an embedded key verifies the same on any machine; the override/env only
+  apply when the bundle omits a key.
+- `ok` is true only when the archive bytes, the telemetry chain, and the
+  trust-audit chain all verify AND no attested signature failed re-verification.
+- A bundle with attested telemetry but no available key DEGRADES by default
+  (`signatureKeyProvided: false`, the intact chain still decides `ok`).
+  `--strict-signatures` refuses such a bundle instead.
+- `--extract-report <path>` writes the bundle's `report.md` out for a human to
+  read alongside the machine verdict. If extraction is requested but the bundle
+  has no `report.md` (or the write fails), that is a failure, not a silent no-op:
+  a `extract-report` / `report-md-unavailable` check is recorded and `ok` is
+  false — so a producer never ships a green verdict with no report attached.
+
+## Fail closed
+
+`cw report verify-bundle` exits non-zero whenever `ok` is false, so
+`cw report verify-bundle <file> && ship` cannot pass on a forgery. This holds even
+for a telemetry chain forged so that every archive file digest still matches: the
+file-digest layer (`inspect-archive`) would wave it through, but the embedded hash
+chain breaks at the next record and the signature re-verification rejects inflated
+usage. A missing, unreadable, or schema-unsupported bundle is also `ok: false`.
+
+## Usage
+
+```
+# See it in 30 seconds, hermetic — no agent, no API key, no repo:
+npx cool-workflow demo bundle
+#   -> builds a sealed bundle, forges it two ways, and shows verify-bundle
+#      catching both offline with only the embedded public key.
+
+# Run the review AND get a shippable, client-verifiable bundle from ONE command:
+cw quickstart architecture-review --repo . --question "What are the risks?" \
+  --agent-command "claude -p" --bundle --with-trust-key ./trust-pub.pem
+#   -> after the drive COMPLETES, the run is sealed into a self-verified bundle and
+#      the verdict is folded into the quickstart JSON (result.bundle). Exits non-zero
+#      if that bundle would not verify. A run that did not complete is never sealed.
+
+# Produce-and-prove from an existing run: export sealed + self-verify + emit the report.
+cw report bundle <run-id> --with-trust-key ./trust-pub.pem \
+  --output report.cwrun.json --extract-report report.md
+#   -> exits non-zero if the produced bundle would not verify (don't ship it).
+
+# Or the two steps separately:
+cw run export <run-id> --with-trust-key ./trust-pub.pem --output report.cwrun.json
+
+# Anyone, anywhere, offline — no repo, no key handed over, no install beyond npx:
+npx cool-workflow report verify-bundle report.cwrun.json
+npx cool-workflow report verify-bundle report.cwrun.json --extract-report report.md
+npx cool-workflow report verify-bundle report.cwrun.json --strict-signatures --json
+```
+
+The same capability is on the MCP surface as `cw_report_verify_bundle`, so an agent
+host re-verifies a bundle through the identical core entry.
+
+See also [Security / Trust Hardening](security-trust-hardening.7.md) and the trust
+model's honest ceiling: a signature proves non-repudiable attribution to the key
+holder, not ground-truth measurement.

package/docs/run-registry-control-plane.7.md

@@ -404,3 +404,9 @@ _No behavioral change in v0.1.82 (run resolution now threads an explicit base di
 ## Hardening and Onboarding (v0.1.83)
 
 Loaders fail closed on corrupt state; store writes are made safe under more than one writer; a new cw doctor checks your setup; help lists every command; and the docs are put into Basic English.
+
+## Privacy Release (v0.1.84)
+
+No other change to this page in v0.1.84.
+
+0.1.85

package/docs/run-retention-reclamation.7.md

@@ -204,3 +204,9 @@ _No changes in v0.1.82._
 ## Hardening and Onboarding (v0.1.83)
 
 Loaders fail closed on corrupt state; store writes are made safe under more than one writer; a new cw doctor checks your setup; help lists every command; and the docs are put into Basic English.
+
+## Privacy Release (v0.1.84)
+
+No other change to this page in v0.1.84.
+
+0.1.85

package/docs/state-explosion-management.7.md

@@ -282,3 +282,9 @@ _No changes in v0.1.82._
 ## Hardening and Onboarding (v0.1.83)
 
 Loaders fail closed on corrupt state; store writes are made safe under more than one writer; a new cw doctor checks your setup; help lists every command; and the docs are put into Basic English.
+
+## Privacy Release (v0.1.84)
+
+No other change to this page in v0.1.84.
+
+0.1.85

package/docs/team-collaboration.7.md

@@ -218,3 +218,9 @@ _No changes in v0.1.82._
 ## Hardening and Onboarding (v0.1.83)
 
 Loaders fail closed on corrupt state; store writes are made safe under more than one writer; a new cw doctor checks your setup; help lists every command; and the docs are put into Basic English.
+
+## Privacy Release (v0.1.84)
+
+No other change to this page in v0.1.84.
+
+0.1.85

package/docs/web-desktop-workbench.7.md

@@ -226,3 +226,9 @@ _No changes in v0.1.82._
 ## Hardening and Onboarding (v0.1.83)
 
 Loaders fail closed on corrupt state; store writes are made safe under more than one writer; a new cw doctor checks your setup; help lists every command; and the docs are put into Basic English.
+
+## Privacy Release (v0.1.84)
+
+No other change to this page in v0.1.84.
+
+0.1.85

package/manifest/plugin.manifest.json

@@ -2,7 +2,7 @@
   "_comment": "SINGLE SOURCE OF TRUTH for every vendor manifest. Edit THIS file, then run `npm run gen:manifests`. Do NOT hand-edit the generated vendor manifests (.claude-plugin/, .codex-plugin/, .agents/, .mcp.json) — `npm run gen:manifests -- --check` (run by release:check) will fail if they drift from this source.",
   "identity": {
     "name": "cool-workflow",
-    "version": "0.1.83",
+    "version": "0.1.85",
     "license": "BSD-2-Clause",
     "homepage": "https://github.com/coo1white/cool-workflow",
     "author": {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cool-workflow",
-  "version": "0.1.83",
+  "version": "0.1.85",
   "bin": {
     "cool-workflow": "scripts/cw.js",
     "cw": "scripts/cw.js"

package/scripts/bump-version.js

@@ -59,7 +59,15 @@ function main() {
   if (!SEMVER.test(next)) fail(`"${next}" is not a x.y.z version`);
 
   const current = JSON.parse(fs.readFileSync(path.join(pluginRoot, "package.json"), "utf8")).version;
-  if (next === current) fail(`already at ${next}`);
+  if (next === current) {
+    // Idempotent: re-running the bump for the same version is a no-op (exit 0)
+    // rather than a hard fail. Required by `release-flow --cut --version X` when
+    // the bump was already landed in a prior PR (e.g. a release-prep PR that
+    // committed the version surfaces, then a follow-up fix on top — cut should
+    // still be able to commit the verdict and tag without re-bumping).
+    process.stdout.write(`bump:version: already at ${next}, no surfaces to update\n`);
+    process.exit(0);
+  }
 
   const touched = [];
   const note = (rel) => touched.push(rel);

package/scripts/canonical-apps.js

@@ -83,7 +83,7 @@ const canonicalApps = [
       "--source",
       "plugins/cool-workflow/docs/workflow-app-framework.7.md",
       "--scope",
-      "Cool Workflow v0.1.83",
+      "Cool Workflow v0.1.85",
       "--freshness",
       "as of release preparation"
     ]
@@ -117,14 +117,14 @@ function main() {
     assert.ok(summary, `${app.id} must appear in app list`);
     assert.equal(summary.sourceKind, "app-directory");
     assert.equal(summary.legacy, false);
-    assert.equal(summary.version, "0.1.83");
+    assert.equal(summary.version, "0.1.85");
 
     const validation = runJson(["app", "validate", manifestPath]);
     assert.equal(validation.valid, true, `${app.id} manifest must validate`);
 
     const shown = runJson(["app", "show", app.id]);
     assert.equal(shown.app.id, app.id);
-    assert.equal(shown.app.version, "0.1.83");
+    assert.equal(shown.app.version, "0.1.85");
     assert.ok(shown.app.metadata.canonical, `${app.id} must be marked canonical`);
     assert.ok(shown.app.sandboxProfiles.length > 0, `${app.id} must declare sandbox profiles`);
     assertTaskIdsUnique(shown);
@@ -135,7 +135,7 @@ function main() {
     const plan = runJson(["plan", app.id, ...app.args(workspace)]);
     const state = JSON.parse(fs.readFileSync(plan.statePath, "utf8"));
     assert.equal(state.workflow.app.id, app.id);
-    assert.equal(state.workflow.app.version, "0.1.83");
+    assert.equal(state.workflow.app.version, "0.1.85");
     assert.equal(state.workflow.app.metadata.canonical, true);
     assert.ok(state.tasks.some((task) => task.requiresEvidence), `${app.id} plan must include evidence gates`);
     assert.ok(state.tasks.every((task) => task.sandboxProfileId), `${app.id} plan must include sandbox hints`);

package/scripts/dogfood-release.js

@@ -5,7 +5,7 @@ const { spawnSync } = require("node:child_process");
 const fs = require("node:fs");
 const path = require("node:path");
 
-const TARGET_VERSION = "0.1.83";
+const TARGET_VERSION = "0.1.85";
 const PREVIOUS_VERSION = "0.1.31";
 const pluginRoot = path.resolve(__dirname, "..");
 const repoRoot = path.resolve(pluginRoot, "..", "..");

package/scripts/golden-path.js

@@ -33,7 +33,7 @@ function main() {
     const appValidation = runJson(["app", "validate", "end-to-end-golden-path"], pluginRoot);
     assert.equal(appValidation.valid, true);
     assert.equal(appValidation.summary.id, "end-to-end-golden-path");
-    assert.equal(appValidation.summary.version, "0.1.83");
+    assert.equal(appValidation.summary.version, "0.1.85");
 
     const plan = runJson(
       [
@@ -42,7 +42,7 @@ function main() {
         "--repo",
         tmp,
         "--question",
-        "Prove the deterministic v0.1.83 end-to-end golden path."
+        "Prove the deterministic v0.1.85 end-to-end golden path."
       ],
       pluginRoot
     );
@@ -52,7 +52,7 @@ function main() {
 
     let state = readJson(plan.statePath);
     assert.equal(state.workflow.app.id, "end-to-end-golden-path");
-    assert.equal(state.workflow.app.version, "0.1.83");
+    assert.equal(state.workflow.app.version, "0.1.85");
     assert.equal(state.loopStage, "interpret");
 
     const dispatch = runJson(["dispatch", plan.runId, "--limit", "1", "--sandbox", "readonly"], tmp);
@@ -195,7 +195,7 @@ function main() {
     assert.equal(reportPath, plan.reportPath);
     assert.ok(fs.existsSync(reportPath));
     const report = fs.readFileSync(reportPath, "utf8");
-    assert.match(report, /Workflow App: end-to-end-golden-path@0\.1\.83/);
+    assert.match(report, /Workflow App: end-to-end-golden-path@0\.1\.85/);
     assert.match(report, /## Candidates/);
     assert.match(report, /## Trust Audit/);
     assert.match(report, /## Acceptance Rationale/);

package/scripts/parity-check.js

@@ -9,8 +9,8 @@
 // FAIL CLOSED ON DRIFT (BSD discipline, same shape as gen-manifests --check):
 //   1. STATIC parity — the declared capability registry must exactly match the
 //      live MCP tool list (tools/list) and the CLI dispatch tokens parsed from
-//      dist/cli.js. A tool or command on one surface but not the other, or not
-//      declared in src/capability-registry.ts, is release-blocking.
+//      the built CLI dispatch surface. A tool or command on one surface but not
+//      the other, or not declared in src/capability-registry.ts, is release-blocking.
 //   2. PAYLOAD parity — for every capability declared `payloadIdentical`, the
 //      `cw <cmd> --json` payload must equal the `cw_<tool>` MCP result on a real
 //      bootstrap run (whitespace + generation-moment ISO timestamps aside).
@@ -31,50 +31,16 @@ const cli = path.join(pluginRoot, "dist", "cli.js");
 const mcpServer = path.join(pluginRoot, "dist", "mcp-server.js");
 const registry = require(path.join(pluginRoot, "dist", "capability-registry.js"));
 
-// Read capabilities that are safe to probe on a freshly planned run with only a
-// runId (or no args). Each: [capability, cliArgvAfterRunId, mcpTool].
-// runId-less reads use [] for cli args and {} mcp args.
-const RUN_PROBES = [
-  ["status", ["--json"], "cw_status"],
-  ["operator.status", ["--json"], "cw_operator_status"],
-  ["operator.report", ["--json"], "cw_operator_report"],
-  ["graph", ["--json"], "cw_operator_graph"],
-  ["report", ["--json"], "cw_report"],
-  ["next", [], "cw_next"],
-  ["state.check", [], "cw_state_check"],
-  ["contract.show", [], "cw_contract_show"],
-  ["node.list", [], "cw_node_list"],
-  ["node.graph", ["--json"], "cw_node_graph"],
-  ["worker.summary", ["--json"], "cw_worker_summary"],
-  ["candidate.summary", ["--json"], "cw_candidate_summary"],
-  ["feedback.summary", ["--json"], "cw_feedback_summary"],
-  ["commit.summary", ["--json"], "cw_commit_summary"],
-  ["audit.summary", [], "cw_audit_summary"],
-  ["multi-agent.summary", ["--json"], "cw_multi_agent_summary"],
-  ["workbench.view", ["--json"], "cw_workbench_view"],
-  ["metrics.show", ["--json"], "cw_metrics_show"],
-  ["review.status", ["--json"], "cw_review_status"],
-  ["comment.list", ["--json"], "cw_comment_list"],
-  ["run.drive", ["--json"], "cw_run_drive"],
-  ["gc.plan", ["--json"], "cw_gc_plan"],
-  ["gc.verify", ["--json"], "cw_gc_verify"]
-];
-const GLOBAL_PROBES = [
-  ["list", "cw_list"],
-  ["app.list", "cw_app_list"],
-  ["topology.list", "cw_topology_list"],
-  ["sandbox.list", "cw_sandbox_list"],
-  ["backend.list", "cw_backend_list"],
-  ["backend.agent.config.show", "cw_backend_agent_config_show"],
-  ["metrics.summary", "cw_metrics_summary"]
-];
-
 function capById(id) {
   const cap = registry.CAPABILITY_REGISTRY.find((entry) => entry.capability === id);
   assert.ok(cap, `probe references unknown capability ${id}`);
   return cap;
 }
 
+function jsonFlag(cap) {
+  return cap.cli.jsonMode === "flag" ? ["--json"] : [];
+}
+
 // ---- 1. static surface parity ---------------------------------------------
 function liveMcpTools() {
   const result = execFileSync(node, [mcpServer], {
@@ -91,10 +57,14 @@ function liveMcpTools() {
 }
 
 function cliDispatchTokens() {
-  const source = fs.readFileSync(cli, "utf8");
+  const source = cliDispatchSources().map((file) => fs.readFileSync(file, "utf8")).join("\n");
   return [...new Set([...source.matchAll(/case\s+"([^"]+)":/g)].map((match) => match[1]))];
 }
 
+function cliDispatchSources() {
+  return [cli, path.join(pluginRoot, "dist", "cli", "command-surface.js")].filter((file) => fs.existsSync(file));
+}
+
 // ---- 2. payload identity ---------------------------------------------------
 // Generation-moment ISO timestamps are presentation metadata, not capability
 // data: the same field carries the wall-clock instant of the render. Neutralize
@@ -144,31 +114,28 @@ async function payloadParity() {
   const runId = plan.runId;
   const mismatches = [];
   const checked = [];
+  const probePlan = registry.payloadProbePlan();
+  assert.deepEqual(probePlan.unclassified, [], "payload probe classification has unclassified capabilities");
+  assert.deepEqual(probePlan.duplicateClassifications, [], "payload probe classification has duplicates");
+  assert.deepEqual(probePlan.invalidClassifications, [], "payload probe classification names non-payload capabilities");
   const mcp = openMcp();
   try {
     await mcp.rpc("initialize", {});
-    for (const [capability, mcpTool] of GLOBAL_PROBES) {
-      const cap = capById(capability);
-      assert.equal(cap.mcp.tool, mcpTool, `probe/registry MCP tool mismatch for ${capability}`);
+    for (const target of registry.payloadProbeTargets()) {
+      const cap = capById(target.capability);
       // jsonMode is the single source for the CLI's --json policy; this probe only
       // appends --json for "flag" verbs and JSON.parse-es the result. The human
       // rendering and "default"-verb no-flag JSON are pinned to cap.cli.jsonMode by
       // the companion test/cli-jsonmode-parity-smoke.js, so cli.ts can't silently
       // re-encode that policy by hand and drift from this registry data.
-      const cliArgv = [...cap.cli.path, ...(cap.cli.jsonMode === "flag" ? ["--json"] : [])];
-      const cliOut = JSON.parse(execFileSync(node, [cli, ...cliArgv], { cwd: workspace, encoding: "utf8" }));
-      const mcpOut = await mcp.tool(mcpTool, { cwd: workspace });
-      checked.push(capability);
-      if (canonical(cliOut) !== canonical(mcpOut)) mismatches.push(capability);
-    }
-    for (const [capability, extraArgv, mcpTool] of RUN_PROBES) {
-      const cap = capById(capability);
-      assert.equal(cap.mcp.tool, mcpTool, `probe/registry MCP tool mismatch for ${capability}`);
-      const cliArgv = [...cap.cli.path, runId, ...extraArgv];
+      const cliArgv = target.kind === "run"
+        ? [...cap.cli.path, runId, ...jsonFlag(cap)]
+        : [...cap.cli.path, ...jsonFlag(cap)];
       const cliOut = JSON.parse(execFileSync(node, [cli, ...cliArgv], { cwd: workspace, encoding: "utf8" }));
-      const mcpOut = await mcp.tool(mcpTool, { cwd: workspace, runId });
-      checked.push(capability);
-      if (canonical(cliOut) !== canonical(mcpOut)) mismatches.push(capability);
+      const mcpArgs = target.kind === "run" ? { cwd: workspace, runId } : { cwd: workspace };
+      const mcpOut = await mcp.tool(cap.mcp.tool, mcpArgs);
+      checked.push(target.capability);
+      if (canonical(cliOut) !== canonical(mcpOut)) mismatches.push(target.capability);
     }
   } finally {
     mcp.server.kill();
@@ -204,6 +171,9 @@ async function main() {
     if (report.missingCliTokens.length) lines.push(`  - registry declares CLI tokens absent from dist/cli.js: ${report.missingCliTokens.join(", ")}`);
     if (report.undeclaredCliTokens.length) lines.push(`  - dist/cli.js dispatches tokens not declared in the registry: ${report.undeclaredCliTokens.join(", ")}`);
     if (report.reasonlessExceptions.length) lines.push(`  - surface-specific / payload-divergent capabilities missing a recorded reason: ${report.reasonlessExceptions.join(", ")}`);
+    if (report.payloadProbeUnclassified.length) lines.push(`  - payload-identical capabilities neither probed nor deferred: ${report.payloadProbeUnclassified.join(", ")}`);
+    if (report.payloadProbeDuplicateClassifications.length) lines.push(`  - payload probe duplicate classifications: ${report.payloadProbeDuplicateClassifications.join(", ")}`);
+    if (report.payloadProbeInvalidClassifications.length) lines.push(`  - payload probe classifications for invalid capabilities: ${report.payloadProbeInvalidClassifications.join(", ")}`);
     if (report.registryLint.length) lines.push(`  - registry lint: ${report.registryLint.join("; ")}`);
     if (payload.mismatches.length) lines.push(`  - cw --json != cw_<tool> payload for: ${payload.mismatches.join(", ")}`);
     lines.push("Reconcile src/capability-registry.ts, cli.ts, and mcp-server.ts so both surfaces render one data source.\n");