cool-workflow 0.2.0 → 0.2.2

package/docs/project-index.md

@@ -1,15 +1,15 @@
 # Cool Workflow Project Index
 
-Generated from the current repository code on 2026-07-05 by `npm run sync:project-index`.
+Generated from the current repository code on 2026-07-08 by `npm run sync:project-index`.
 
 ## Snapshot
 
 - Package: `cool-workflow`
-- Version: `0.2.0`
-- Source modules: `129`
+- Version: `0.2.2`
+- Source modules: `145`
 - Workflow apps: `8`
-- Docs: `61`
-- Smoke tests: `173`
+- Docs: `62`
+- Smoke tests: `181`
 - Repository: https://github.com/coo1white/cool-workflow
 
 ## Architecture
@@ -82,6 +82,7 @@ multi-agent host -> topology -> blackboard/coordinator
 - [cli/entry.ts](../src/cli/entry.ts)
 - [cli/io.ts](../src/cli/io.ts)
 - [cli/parseargv.ts](../src/cli/parseargv.ts)
+- [core/capability-data.ts](../src/core/capability-data.ts)
 - [core/capability-table.ts](../src/core/capability-table.ts)
 - [core/format/help.ts](../src/core/format/help.ts)
 - [core/format/state-explosion-text.ts](../src/core/format/state-explosion-text.ts)
@@ -113,6 +114,9 @@ multi-agent host -> topology -> blackboard/coordinator
 - [core/trust/telemetry-ledger.ts](../src/core/trust/telemetry-ledger.ts)
 - [core/types.ts](../src/core/types.ts)
 - [core/types/boundary.ts](../src/core/types/boundary.ts)
+- [core/types/execution-backend.ts](../src/core/types/execution-backend.ts)
+- [core/types/observability.ts](../src/core/types/observability.ts)
+- [core/util/collate.ts](../src/core/util/collate.ts)
 - [mcp/dispatch.ts](../src/mcp/dispatch.ts)
 - [mcp/server.ts](../src/mcp/server.ts)
 - [shell/agent-config.ts](../src/shell/agent-config.ts)
@@ -181,6 +185,18 @@ multi-agent host -> topology -> blackboard/coordinator
 - [shell/workbench-text.ts](../src/shell/workbench-text.ts)
 - [shell/workbench.ts](../src/shell/workbench.ts)
 - [shell/worker-cli.ts](../src/shell/worker-cli.ts)
+- [wiring/capability-table/basics.ts](../src/wiring/capability-table/basics.ts)
+- [wiring/capability-table/exec-backend.ts](../src/wiring/capability-table/exec-backend.ts)
+- [wiring/capability-table/index.ts](../src/wiring/capability-table/index.ts)
+- [wiring/capability-table/multi-agent.ts](../src/wiring/capability-table/multi-agent.ts)
+- [wiring/capability-table/parity.ts](../src/wiring/capability-table/parity.ts)
+- [wiring/capability-table/pipeline.ts](../src/wiring/capability-table/pipeline.ts)
+- [wiring/capability-table/registry-core.ts](../src/wiring/capability-table/registry-core.ts)
+- [wiring/capability-table/reporting.ts](../src/wiring/capability-table/reporting.ts)
+- [wiring/capability-table/scheduling-registry.ts](../src/wiring/capability-table/scheduling-registry.ts)
+- [wiring/capability-table/state.ts](../src/wiring/capability-table/state.ts)
+- [wiring/capability-table/trust-ledger.ts](../src/wiring/capability-table/trust-ledger.ts)
+- [wiring/capability-table/workflow-apps.ts](../src/wiring/capability-table/workflow-apps.ts)
 
 ## Workflow Apps
 
@@ -251,6 +267,7 @@ multi-agent host -> topology -> blackboard/coordinator
 - [State Explosion Management](state-explosion-management.7.md)
 - [STATE-NODE(7)](state-node.7.md)
 - [Team Collaboration](team-collaboration.7.md)
+- [Trust Audit Anchor](trust-audit-anchor.7.md)
 - [Trust Model & Limitations](trust-model.md)
 - [Unix-Inspired Workflow Principles](unix-principles.md)
 - [Vendor Manifest Loadability](vendor-manifest-loadability.7.md)
@@ -263,6 +280,8 @@ multi-agent host -> topology -> blackboard/coordinator
 
 Smoke tests mirror the public contracts. The high-signal suites are:
 
+- [agent-backend-concurrent-user-env-smoke.js](../test/agent-backend-concurrent-user-env-smoke.js)
+- [agent-backend-user-env-smoke.js](../test/agent-backend-user-env-smoke.js)
 - [agent-config-atomic-write-smoke.js](../test/agent-config-atomic-write-smoke.js)
 - [agent-delegation-drive-smoke.js](../test/agent-delegation-drive-smoke.js)
 - [agent-stream-gate-smoke.js](../test/agent-stream-gate-smoke.js)
@@ -270,6 +289,7 @@ Smoke tests mirror the public contracts. The high-signal suites are:
 - [architecture-review-fast-automation-smoke.js](../test/architecture-review-fast-automation-smoke.js)
 - [architecture-review-fast-phase-cache-smoke.js](../test/architecture-review-fast-phase-cache-smoke.js)
 - [architecture-review-fast-smoke.js](../test/architecture-review-fast-smoke.js)
+- [architecture-review-question-aware-smoke.js](../test/architecture-review-question-aware-smoke.js)
 - [artifact-integrity-smoke.js](../test/artifact-integrity-smoke.js)
 - [audit-verify-smoke.js](../test/audit-verify-smoke.js)
 - [backend-registry-smoke.js](../test/backend-registry-smoke.js)
@@ -310,6 +330,7 @@ Smoke tests mirror the public contracts. The high-signal suites are:
 - [deferred-checkpoint-batching-smoke.js](../test/deferred-checkpoint-batching-smoke.js)
 - [demo-bundle-smoke.js](../test/demo-bundle-smoke.js)
 - [det-ids-b-smoke.js](../test/det-ids-b-smoke.js)
+- [dispatch-legacy-burndown-smoke.js](../test/dispatch-legacy-burndown-smoke.js)
 - [doctor-smoke.js](../test/doctor-smoke.js)
 - [dogfood-architecture-review-smoke.js](../test/dogfood-architecture-review-smoke.js)
 - [dogfood-release-smoke.js](../test/dogfood-release-smoke.js)
@@ -339,6 +360,7 @@ Smoke tests mirror the public contracts. The high-signal suites are:
 - [mcp-app-surface-smoke.js](../test/mcp-app-surface-smoke.js)
 - [mcp-surface-registry-smoke.js](../test/mcp-surface-registry-smoke.js)
 - [mcp-tool-call-coverage-smoke.js](../test/mcp-tool-call-coverage-smoke.js)
+- [metrics-summary-limit-smoke.js](../test/metrics-summary-limit-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)
@@ -405,11 +427,13 @@ Smoke tests mirror the public contracts. The high-signal suites are:
 - [run-restore-failclosed-smoke.js](../test/run-restore-failclosed-smoke.js)
 - [run-resume-drive-smoke.js](../test/run-resume-drive-smoke.js)
 - [run-retention-reclamation-smoke.js](../test/run-retention-reclamation-smoke.js)
+- [run-state-lock-concurrency-smoke.js](../test/run-state-lock-concurrency-smoke.js)
 - [sample-determinism-smoke.js](../test/sample-determinism-smoke.js)
 - [sandbox-env-batch-hardening-smoke.js](../test/sandbox-env-batch-hardening-smoke.js)
 - [sandbox-profile-smoke.js](../test/sandbox-profile-smoke.js)
 - [sched-policy-validation-smoke.js](../test/sched-policy-validation-smoke.js)
 - [schedule-routine-daemon-smoke.js](../test/schedule-routine-daemon-smoke.js)
+- [scheduling-routine-lock-concurrency-smoke.js](../test/scheduling-routine-lock-concurrency-smoke.js)
 - [schema-validation-smoke.js](../test/schema-validation-smoke.js)
 - [security-trust-hardening-smoke.js](../test/security-trust-hardening-smoke.js)
 - [self-audit-hardening-smoke.js](../test/self-audit-hardening-smoke.js)
@@ -427,6 +451,7 @@ Smoke tests mirror the public contracts. The high-signal suites are:
 - [telemetry-metrics-coverage-smoke.js](../test/telemetry-metrics-coverage-smoke.js)
 - [telemetry-verify-signatures-smoke.js](../test/telemetry-verify-signatures-smoke.js)
 - [token-budget-enforcement-smoke.js](../test/token-budget-enforcement-smoke.js)
+- [trust-audit-anchor-smoke.js](../test/trust-audit-anchor-smoke.js)
 - [vendor-manifest-load-smoke.js](../test/vendor-manifest-load-smoke.js)
 - [vendor-preflight-smoke.js](../test/vendor-preflight-smoke.js)
 - [verifier-gated-commit-smoke.js](../test/verifier-gated-commit-smoke.js)

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

@@ -191,3 +191,7 @@ _No behavioral change in v0.1.89 (CLI-surface golden-path + help-output fixes on
 0.1.98
 
 0.2.0
+
+0.2.1
+
+0.2.2

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

@@ -331,3 +331,7 @@ _No behavioral change in v0.1.89 (CLI-surface golden-path + help-output fixes on
 0.1.98
 
 0.2.0
+
+0.2.1
+
+0.2.2

package/docs/release-tooling.7.md

@@ -294,3 +294,7 @@ _No behavioral change in v0.1.89 (CLI-surface golden-path + help-output fixes on
 0.1.98
 
 0.2.0
+
+0.2.1
+
+0.2.2

package/docs/remote-source-review.7.md

@@ -1,15 +1,15 @@
 # Remote-Source Review (`--link`)
 
 CW v0.1.91 lets you point a review at **any repository on the internet** instead
-of only a local path: `cw -q "what are the risks?" --link <url>`. CW materializes
+of only a local path: `cw -q "how does this work?" --link <url>`. CW materializes
 the remote into a local checkout and runs the **existing** review pipeline against
 it — identical downstream to reviewing a folder. A URL passed to `-dir`/`--repo`
 is auto-detected, so `--link <url>` and `-dir <url>` are equivalent.
 
 ```bash
-cw -q "What are the risks?" --link https://github.com/owner/repo
-cw -q "What are the risks?" --link git@gitlab.com:owner/repo.git --ref v1.2.0
-cw -q "What are the risks?" --link https://github.com/owner/repo/archive/refs/heads/main.tar.gz
+cw -q "How does auth work end-to-end here?" --link https://github.com/owner/repo
+cw -q "What are the security risks?" --link git@gitlab.com:owner/repo.git --ref v1.2.0
+cw -q "Is it safe to swap this queue for Redis?" --link https://github.com/owner/repo/archive/refs/heads/main.tar.gz
 cw -q "..." --link <url> --check     # validate the URL + tooling WITHOUT fetching
 ```
 

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

@@ -127,7 +127,7 @@ npx cool-workflow demo 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?" \
+cw quickstart architecture-review --repo . --question "How does auth work end-to-end here?" \
   --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

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

@@ -474,3 +474,7 @@ _No behavioral change in v0.1.89 (CLI-surface golden-path + help-output fixes on
 0.1.98
 
 0.2.0
+
+0.2.1
+
+0.2.2

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

@@ -263,3 +263,7 @@ _No behavioral change in v0.1.89 (CLI-surface golden-path + help-output fixes on
 0.1.98
 
 0.2.0
+
+0.2.1
+
+0.2.2

package/docs/security-trust-hardening.7.md

@@ -37,7 +37,9 @@ Event sources are clear:
 - `runtime-derived`: CW got the event from run state.
 
 CW does not keep secrets or raw environment values. Environment audit records
-keep names only.
+keep names only — for example, `worker.agent-env` records which provider-namespace
+env var NAMES (`CW_`, `ANTHROPIC_`, `OPENAI_`, and similar) were forwarded from
+the host process to a delegated agent child, queryable via `cw audit summary`/`cw audit`.
 
 ## Enforcement Boundary
 
@@ -49,6 +51,34 @@ The agent host must still make certain of OS-level read isolation, write isolati
 limits on process execution, limits on network, and environment filtering.
 The audit layer lets you look at that boundary; it is not a kernel.
 
+## Requiring Attested Telemetry
+
+`CW_REQUIRE_ATTESTED_TELEMETRY=1` (or `--require-attested-telemetry`, or a
+durable `agent-config.json` field) is an opt-in, off-by-default gate: once
+on, CW refuses to accept ANY worker result whose usage telemetry is not
+cryptographically `attested` — a delegated hop whose signature does not
+verify, and now also a MANUAL accept (`cw worker output`, `cw result`) that
+carries no delegation telemetry at all. The manual shape used to slip past
+the gate silently; it is now blocked the same way, with its own code
+(`telemetry-missing-blocked`, distinct from a present-but-unverified hop's
+`telemetry-unattested-blocked`).
+
+An operator who genuinely needs to accept an unattested manual result under
+the require flag passes `--allow-unattested` (MCP: `allowUnattested`). This
+is never silent: it writes a `telemetry.gate-override` trust-audit event
+(`decision: "allowed"`, `source: "operator"`) into the same hash-chained log
+every other audit decision goes through, so the override itself is part of
+the auditable record, not a hole in it.
+
+A result-cache hit (the drive loop replaying a previously accepted result) is
+never re-blocked by this gate — the underlying result was already gated (or
+overridden) at its first acceptance. When the require flag is on, a cache
+accept still records a `telemetry.cache-accept` event, so the audit trail
+shows which accepts came from a fresh hop and which from the cache.
+
+With the require flag off (the default), none of this changes: a plain
+`cw worker output` / `cw result` behaves exactly as before.
+
 ## CLI
 
 ```bash
@@ -65,6 +95,8 @@ node scripts/cw.js audit decision <run-id> <worker-id> --path <path>
 node scripts/cw.js audit decision <run-id> <worker-id> --command "npm test"
 node scripts/cw.js audit decision <run-id> <worker-id> --network example.com
 node scripts/cw.js audit decision <run-id> <worker-id> --env SECRET_NAME
+node scripts/cw.js worker output <run-id> <worker-id> <result-file> [--allow-unattested]
+node scripts/cw.js result <run-id> <task-id> <result-file> [--allow-unattested]
 ```
 
 Denied audit decisions are put into audit files and joined to feedback/error

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

@@ -320,3 +320,7 @@ _No behavioral change in v0.1.89 (CLI-surface golden-path + help-output fixes on
 0.1.98
 
 0.2.0
+
+0.2.1
+
+0.2.2

package/docs/team-collaboration.7.md

@@ -256,3 +256,7 @@ _No behavioral change in v0.1.89 (CLI-surface golden-path + help-output fixes on
 0.1.98
 
 0.2.0
+
+0.2.1
+
+0.2.2

package/docs/trust-audit-anchor.7.md

@@ -0,0 +1,69 @@
+# Trust Audit Anchor
+
+CW v0.2.1 adds the Trust Audit Anchor: a way to see when the END of a run's
+trust-audit log was cut off. The hash chain in `audit/events.jsonl` lets
+`cw audit verify` see an edited event, a removed middle event, a bad line, and
+a mixed-era forgery. But one tamper shape gets past a pure chain walk: take
+the last N lines off the file, and what is left is a shorter but fully
+consistent chain — verify stays green. The anchor closes that hole, and it
+does so without changing any old output byte.
+
+## Design Discipline
+
+- Mechanism, not policy: the kernel gives you two small parts — a read of the
+  chain head, and a check against a head you saved earlier. WHEN you save a
+  head (after a run, before you publish, at export time) is your policy.
+- Fail-closed: a saved head that is not on the chain, or an event count that
+  comes up short, makes verify exit non-zero with the distinct check code
+  `trust-audit-truncated`. A bad `--expect-head` / `--expect-count` value is
+  an error, never a check silently made weaker.
+- POLA: with no anchor flags, `cw audit verify` output is byte-for-byte what
+  it was before this feature (no `anchor` key, same checks, same exit rules).
+- Reuse: the anchor rides on the existing eventHash chain — no new file, no
+  new state, no new hash form. `cw audit head` is a read-only projection.
+- Parity: `audit.head` is on both front doors (`cw audit head` and the MCP
+  tool `cw_audit_head`); the anchor args are on both `cw audit verify` and
+  `cw_audit_verify` (`expectHead` / `expectCount`).
+
+## CLI
+
+```text
+node dist/cli.js audit head <run-id>
+# -> { "schemaVersion": 1, "runId": "...", "eventCount": 87,
+#      "headHash": "sha256:..." }
+
+node dist/cli.js audit verify <run-id> --expect-head <hash> --expect-count <n>
+# green: verified true, "anchor": { ..., "satisfied": true }, exit 0
+# cut tail: verified false, checks carry trust-audit-truncated, exit 1
+```
+
+The head is the hash the NEXT appended event will link from: the last event's
+`eventHash`, or the run's genesis hash when the log is empty. Save the pair
+`{headHash, eventCount}` somewhere the log's writer cannot reach — a CI
+variable, a note in your PR, the output of `cw run export` (the export
+manifest hashes the event log bytes, so an export IS an anchor in file form).
+
+## How the check works
+
+`verifyTrustAudit` walks the chain as before, and keeps the trail of head
+hashes it saw (genesis, then the hash after each event). With an anchor:
+
+- `expectCount`: the walked log must have at least that many events. Fewer =
+  `trust-audit-truncated` (check name `anchor-count`).
+- `expectHead`: the saved head must be ON the trail. A log that was cut and
+  then padded back with new events reaches the old count, but the new events
+  link from an earlier point — the old head is no longer on the trail, so
+  this still fails (check name `anchor-head`).
+
+## Compatibility
+
+Trust Audit Anchor is introduced in CW v0.2.1. Fields are additive and
+optional; older run state loads unchanged. A plain `cw audit verify` keeps
+its exact old output. The `anchor` key appears in the JSON only when the
+caller passed an anchor flag.
+
+## See Also
+
+security-trust-hardening(7), cli-mcp-parity(7), report-verifiable-bundle(7)
+
+0.2.2

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

@@ -96,11 +96,21 @@ The console is read-only. It offers no actions. If a later release adds an actio
 capability core entry that already exists — never a side code path — so it cannot drift from the
 CLI/MCP and is covered by the parity gate.
 
+Authentication is opt-in, not on by default: set `CW_WORKBENCH_TOKEN` and every
+request must carry it as `Authorization: Bearer <token>` or `?token=<token>`
+(timing-safe compare); left unset, the loopback/read-only/GET-only controls
+above are the only defense and any local process can read. Pass
+`--require-token` to `cw workbench serve` to fail closed instead: the server
+refuses to start at all unless `CW_WORKBENCH_TOKEN` is already set. This is a
+strict opt-in — leaving it off keeps today's default behavior unchanged.
+(`--require-token` has no effect on `cw_workbench_serve`'s MCP path, which
+never actually binds a listener.)
+
 ## Surfaces
 
 ```
 cw workbench view <run-id> [--json]         # five-panel WorkbenchRunView for one run
-cw workbench serve [--port N] [--scope repo|home] [--once|--json]
+cw workbench serve [--port N] [--scope repo|home] [--once|--json] [--require-token]
 ```
 
 `cw workbench serve` with `--once`/`--json` prints the serve descriptor (bind
@@ -264,3 +274,7 @@ _No behavioral change in v0.1.89 (CLI-surface golden-path + help-output fixes on
 0.1.98
 
 0.2.0
+
+0.2.1
+
+0.2.2

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": "0.2.0",
+    "version": "0.2.2",
     "license": "BSD-2-Clause",
     "homepage": "https://github.com/coo1white/cool-workflow",
     "author": {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cool-workflow",
-  "version": "0.2.0",
+  "version": "0.2.2",
   "bin": {
     "cool-workflow": "scripts/cw.js",
     "cw": "scripts/cw.js"
@@ -43,6 +43,7 @@
     "build": "tsc -p tsconfig.json",
     "check": "tsc -p tsconfig.json --noEmit",
     "dist:check": "node scripts/dist-drift-check.js",
+    "purity:check": "node scripts/purity-gate.js",
     "golden-path": "node scripts/golden-path.js",
     "dogfood:release": "node scripts/dogfood-release.js",
     "canonical-apps": "node scripts/canonical-apps.js",
@@ -68,6 +69,7 @@
     "test:fast": "npm run build --if-present && node dist/cli.js version > /dev/null && node test/run-all.js --concurrency auto",
     "test:ci": "npm run build && node dist/cli.js version > /dev/null && node test/run-all.js --sample 55",
     "test:coverage": "node dist/cli.js version > /dev/null && node scripts/coverage-gate.js --concurrency auto",
+    "test:unit": "node dist/cli.js version > /dev/null && node test/run-unit.js",
     "eval:replay": "tsc -p tsconfig.json && node test/multi-agent-eval-replay-harness-smoke.js",
     "ci": "npm run build && npm run check && npm run test && npm run release:check",
     "validate:schema": "node scripts/validate-run-state-schema.js"

package/scripts/agents/agent-adapter-core.js

@@ -18,7 +18,7 @@ cw:result block that EXACTLY follows this schema:
   "findings": [
     {
       "id": "unique-kebab-id",
-      "title": "short risk title",
+      "title": "short finding title",
       "severity": "P0",
       "classification": "real",
       "evidence": ["path/to/file.ts:42"]
@@ -471,6 +471,25 @@ function persistStderr(resultPath, text) {
   }
 }
 
+// Build the failure-path diagnostic string for a non-zero exit. Real OS-level
+// stderr wins when present (the common case: a crash, a killed process, a
+// vendor CLI that DOES write to stderr) — byte-identical to before this
+// helper existed. When stderr is EMPTY, fall back to whatever partial/
+// structured text the wrapper already parsed from stdout before the child
+// died (a stream-json `result` event with is_error:true is how claude
+// reports terminal errors like an auth failure — that text was being parsed
+// and then silently thrown away). If there is no partial text either, say so
+// plainly instead of just repeating the bare exit code. Never throws; always
+// returns a non-empty string.
+function buildFailureDetail({ label, code, childStderr, partialText }) {
+  const stderrText = String(childStderr || "").trim();
+  if (stderrText) return stderrText;
+  const partial = String(partialText || "").trim();
+  const codeText = `${label} exited ${code === null ? "(timeout/killed)" : code}`;
+  if (partial) return `${codeText} — ${label} stdout (no stderr) said: ${partial}`;
+  return codeText;
+}
+
 module.exports = {
   RESULT_CONTRACT,
   buildPrompt,
@@ -485,5 +504,6 @@ module.exports = {
   flushJsonLines,
   writeResult,
   emitReport,
-  persistStderr // save a failed agent's stderr to <workerDir>/logs/agent-stderr.log (shared by all wrappers)
+  persistStderr, // save a failed agent's stderr to <workerDir>/logs/agent-stderr.log (shared by all wrappers)
+  buildFailureDetail // fold in parsed partial stdout text when real stderr is empty (shared by all wrappers)
 };

package/scripts/agents/claude-p-agent.js

@@ -34,7 +34,7 @@ const { spawn, spawnSync } = require("node:child_process");
 // wrappers instead of carrying a private copy. A drifted inline copy (ASCII
 // hyphens silently became em-dashes here) meant claude was sent a different
 // instruction text than the other providers for the same contract.
-const { buildPrompt, createRenderer, persistStderr, toolLabel, summarizeToolResult } = require("./agent-adapter-core");
+const { buildPrompt, createRenderer, persistStderr, toolLabel, summarizeToolResult, buildFailureDetail } = require("./agent-adapter-core");
 
 const inputPath = process.argv[2];
 const resultPath = process.argv[3];
@@ -179,9 +179,14 @@ child.on("close", (code) => {
   render.finishLive(); // stop the spinner + restore the cursor BEFORE any further output
   render.writeTranscript(transcriptPath); // full narration + tool I/O always saved
   if (code !== 0) {
-    const detail = childStderr.trim() || `claude exited ${code === null ? "(timeout/killed)" : code}`;
+    const stderrText = childStderr.trim();
+    // childStderr is the OS-level stream; claude's real failure reason (auth,
+    // rate limit, relay error) is often reported as a stream-json `result`
+    // event on STDOUT instead, already parsed into `resultText` above — fold
+    // it in here so a silent-stderr failure still leaves a readable reason.
+    const detail = buildFailureDetail({ label: "claude", code, childStderr: stderrText, partialText: resultText });
     persistStderr(resultPath, detail);
-    if (childStderr.trim()) process.stderr.write(`${childStderr.trim()}\n`);
+    if (stderrText) process.stderr.write(`${stderrText}\n`);
     process.stderr.write(`claude exited ${code === null ? "(timeout/killed)" : code}\n`);
     process.exit(code === null ? 1 : code);
   }

package/scripts/agents/gemini-agent.js

@@ -17,6 +17,7 @@
 const path = require("node:path");
 const { spawn } = require("node:child_process");
 const {
+  buildFailureDetail,
   buildPrompt,
   createRenderer,
   emitReport,
@@ -94,7 +95,10 @@ child.on("close", (code) => {
   render.finishLive();
   render.writeTranscript(transcriptPath);
   if (code !== 0) {
-    const detail = childStderr.trim() || `gemini exited ${code === null ? "(timeout/killed)" : code}`;
+    // gemini's real failure reason is often only in the NDJSON text/result
+    // fragments already parsed into `state`, not in raw OS-level stderr.
+    const partial = state.finalResult || state.textFragments.join("\n\n");
+    const detail = buildFailureDetail({ label: "gemini", code, childStderr: childStderr.trim(), partialText: partial });
     persistStderr(resultPath, detail);
     process.stderr.write(`${detail}\n`);
     process.exit(code === null ? 1 : code);

package/scripts/agents/opencode-agent.js

@@ -22,6 +22,7 @@
 const path = require("node:path");
 const { spawn } = require("node:child_process");
 const {
+  buildFailureDetail,
   buildPrompt,
   createRenderer,
   emitReport,
@@ -147,7 +148,10 @@ child.on("close", (code) => {
   render.finishLive();
   render.writeTranscript(transcriptPath);
   if (code !== 0) {
-    const detail = childStderr.trim() || `opencode exited ${code === null ? "(timeout/killed)" : code}`;
+    // opencode's real failure reason is often only in the JSONL text/result
+    // fragments already parsed into `state`, not in raw OS-level stderr.
+    const partial = state.finalResult || state.lastMessageText || state.textFragments.join("\n\n");
+    const detail = buildFailureDetail({ label: "opencode", code, childStderr: childStderr.trim(), partialText: partial });
     persistStderr(resultPath, detail);
     process.stderr.write(`${detail}\n`);
     process.exit(code === null ? 1 : code);

package/scripts/bump-version.js

@@ -46,6 +46,23 @@ function replaceFirstVersionField(absPath, next) {
   return true;
 }
 
+function replaceLockfileVersions(absPath, next) {
+  // Move BOTH lockfile version fields with targeted string swaps, so the byte
+  // formatting npm wrote is kept as it is: the top-level "version" (the first
+  // one in the file) and the root-package entry `"": { "name": ..., "version": ... }`.
+  // Dependency entries are not touched — the second swap is keyed on the
+  // `"": {` root-package opening, which is present only once.
+  const text = fs.readFileSync(absPath, "utf8");
+  let updated = text.replace(/"version":\s*"[^"]*"/, `"version": "${next}"`);
+  updated = updated.replace(
+    /("":\s*\{\s*"name":\s*"[^"]*",\s*"version":\s*)"[^"]*"/,
+    `$1"${next}"`
+  );
+  if (updated === text) return false;
+  fs.writeFileSync(absPath, updated);
+  return true;
+}
+
 function setNestedVersion(absPath, next) {
   // For files where the first `"version"` is NOT the right one, parse + set.
   const json = JSON.parse(fs.readFileSync(absPath, "utf8"));
@@ -75,9 +92,13 @@ function main() {
   // 1. package.json (the single source of truth version:sync now reads from)
   if (replaceFirstVersionField(path.join(pluginRoot, "package.json"), next)) note("package.json");
 
-  // 2. package-lock.json (gitignored install artifact; only if present)
+  // 2. package-lock.json (tracked; only if present). The lockfile keeps the
+  //    version in TWO places: the top-level "version" and the root-package
+  //    entry packages[""].version. The old code moved only the first one, so
+  //    the root-package entry kept an old version till the next `npm install`
+  //    (the v0.1.97 drift seen after the v0.2.0/v0.2.1 cuts). Move both here.
   const lock = path.join(pluginRoot, "package-lock.json");
-  if (fs.existsSync(lock) && replaceFirstVersionField(lock, next)) note("package-lock.json");
+  if (fs.existsSync(lock) && replaceLockfileVersions(lock, next)) note("package-lock.json");
 
   // 2b. Official MCP Registry server metadata (top-level server version + npm package version).
   const serverJson = path.join(pluginRoot, "server.json");
@@ -218,6 +239,7 @@ function contentSurfaceFiles(next) {
     { path: "plugins/cool-workflow/docs/run-retention-reclamation.7.md",     needle: next, desc: "run retention doc" },
     { path: "plugins/cool-workflow/docs/durable-state-and-locking.7.md",     needle: next, desc: "durable state doc" },
     { path: "plugins/cool-workflow/docs/release-and-migration.7.md",         needle: next, desc: "release & migration doc" },
+    { path: "plugins/cool-workflow/docs/trust-audit-anchor.7.md",           needle: next, desc: "trust audit anchor doc" },
   ];
 }
 

package/scripts/canonical-apps.js

@@ -83,7 +83,7 @@ const canonicalApps = [
       "--source",
       "plugins/cool-workflow/docs/workflow-app-framework.7.md",
       "--scope",
-      "Cool Workflow v0.2.0",
+      "Cool Workflow v0.2.2",
       "--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, "0.2.0");
+    assert.equal(summary.version, "0.2.2");
 
     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, "0.2.0");
+    assert.equal(shown.app.version, "0.2.2");
     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, "0.2.0");
+    assert.equal(state.workflow.app.version, "0.2.2");
     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

@@ -6,7 +6,7 @@ const fs = require("node:fs");
 const path = require("node:path");
 const { CoolWorkflowRunner } = require("../dist/shell/orchestrator.js");
 
-const TARGET_VERSION = "0.2.0";
+const TARGET_VERSION = "0.2.2";
 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, "0.2.0");
+    assert.equal(appValidation.summary.version, "0.2.2");
 
     const plan = runJson(
       [
@@ -42,7 +42,7 @@ function main() {
         "--repo",
         tmp,
         "--question",
-        "Prove the deterministic v0.2.0 end-to-end golden path."
+        "Prove the deterministic v0.2.2 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, "0.2.0");
+    assert.equal(state.workflow.app.version, "0.2.2");
     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\.2\.0/);
+    assert.match(report, /Workflow App: end-to-end-golden-path@0\.2\.2/);
     assert.match(report, /## Candidates/);
     assert.match(report, /## Trust Audit/);
     assert.match(report, /## Acceptance Rationale/);