@dzhechkov/harness-cli 0.3.238 → 0.3.239

package/.dz-manifest.json

@@ -9,7 +9,7 @@
       },
       {
         "path": "README.md",
-        "sha256": "ef6603652143fc1517c81506dad9d5e7b79e5d5138aad9f6281ad77492d58b71"
+        "sha256": "6cdf43e972df9144b6eb93beed515bb216c1545bdd69b696b773a3dbe2d5b679"
       },
       {
         "path": "coverage/coverage-final.json",
@@ -41,15 +41,15 @@
       },
       {
         "path": "dist/cli.d.ts.map",
-        "sha256": "8f8807920c7ddd901361eb5a602df9c85fa69726e47535a146937038eab6b281"
+        "sha256": "dfb260e21d46dfc8020eef0bf5ba570f9c335964181cf90446c1c5d761626415"
       },
       {
         "path": "dist/cli.js",
-        "sha256": "5f5fbcc7c040a272f6af992ef1f9c39482b941d74dbcadec9f3d9008babcb08a"
+        "sha256": "a5b7d2fc4f582c507dfab54f1cc4754ca6700f7e686a3c72871afa807cceac25"
       },
       {
         "path": "dist/cli.js.map",
-        "sha256": "420d1178a513e54f315095a17e9c5722543ca6cbdbebd9a674d4d50ef61b8aa0"
+        "sha256": "8fb0c46e092ac7240af9d254f272da649d911b3307b39dc60ac1ef43d349834f"
       },
       {
         "path": "dist/index.d.ts",
@@ -73,7 +73,7 @@
       },
       {
         "path": "package.json",
-        "sha256": "8fbaf10d29b5733cf9445e1ef5804a3e124f733c3390a99937d2d8dc3b4cf02a"
+        "sha256": "82a481991b03522d4da62fe49c032c03445928a46a226e8cbb05498eb9321950"
       },
       {
         "path": "src/bin.ts",
@@ -81,7 +81,7 @@
       },
       {
         "path": "src/cli.ts",
-        "sha256": "131f23a50187e1be19875d4ccd76d8c354b1fbb8d193c03e73fda16ff64f51dc"
+        "sha256": "ccb1af3966a258b17eed99ce8b830639166ad8a8a5075ca027d8f30077025d19"
       },
       {
         "path": "src/index.ts",
@@ -101,5 +101,5 @@
       }
     ]
   },
-  "signature": "spl/yr0En3yUAxhYjhIdot8D/Tri4JeoMrweQkA9GF2F9UyzDco3FLpaqCkvwlhGvj4phKb4qB0Nm3e+GAMaDw=="
+  "signature": "mJYQHwfoPEK5x7cn8B0AbYnqRpUO9iy3qHhWrIHQ/4xqgsTOb/DMSDScPDY2GxfvZsk2pUVuPQS8xjWMKXH9BA=="
 }

package/README.md

@@ -434,6 +434,57 @@ dz parity — codex  (capabilities: shell, skills, mcp)
 **When to use:** before promising a workflow to a teammate on Cursor/Codex/Hermes; when choosing
 a target for a project; as the requirements input for porting a feature to more targets.
 
+### Do your skills actually register? — `dz skills-verify`
+
+Shipping a skill pack is not the same as a skill **registering**. A layout test that asserts
+`SKILL.md` exists on disk verifies a **proxy**; the property that matters is whether Claude Code
+loads it. `@dzhechkov/health-advisor` 1.2.0 shipped to npm with a green layout test and **zero**
+skills registering — the gap was only found by hand, after publish. `dz skills-verify` closes it.
+
+Two layers:
+
+| layer | needs a session? | what it proves |
+|---|---|---|
+| `--static` | no — instant, CI-safe | which dirs *can* register, and flags the three shapes that never can |
+| default | yes (~3 s) | the **authoritative** listing, read from the session's `system/init` event — no model prose |
+
+```bash
+dz skills-verify --static                       # CI gate: layout only, exits 1 on a problem
+dz skills-verify                                # full check in the current project
+dz skills-verify --dir ../my-app --expect my-skill,my-other-skill
+dz skills-verify --json --strict                # machine-readable; --strict makes inconclusive exit 1
+```
+
+A healthy project, and one with the exact defect that shipped:
+
+```
+$ dz skills-verify
+dz skills-verify: PASS — all 72 expected skill(s) are registered
+  layout: 72 registrable skill dir(s) under /path/.claude/skills
+  session: 106 skill(s) registered · client 2.1.220 · 5 plugin(s) loaded
+
+$ dz skills-verify --dir /tmp/broken-install
+dz skills-verify: FAIL — nothing can register: 24 layout problem(s) and no registrable skill directory
+  layout problems (these can never register):
+    [no-skill-md] no SKILL.md at health-advisor/SKILL.md — this directory cannot register
+    [plugin-manifest-trap] health-advisor/.claude-plugin/plugin.json does not auto-register — plugins load from the marketplace, not from .claude/skills
+    [buried-skill-md] health-advisor/skills/clinical-decision-support/SKILL.md is 2+ levels deep …
+```
+
+A `.claude-plugin/plugin.json` under `.claude/skills/` is reported as an **advisory**, not a failure:
+that layout may load under workspace trust, so the verdict asks the session whether the plugin actually
+loaded (`init.plugins`) instead of judging the layout. A container with skills inside and no matching
+loaded plugin fails, and says exactly that.
+
+**Exit codes: `0` pass · `1` fail · `2` inconclusive.** The third is the point — a missing `claude`
+binary, a login prompt, a timeout, or a session that read a *different* project all yield
+`inconclusive`, **never** a pass. (A session that read another project cannot testify that a skill is
+missing, so its listing is refused rather than believed.) A layout so broken that nothing is
+registrable fails too, instead of passing vacuously on an empty expectation.
+
+**Use it when:** publishing a package that installs skills, after `dz setup` in a new project, or in
+CI (`--static` needs no Claude session at all).
+
 ### Portable delivery gate — `dz delivery-check` (Step-10, on every shell target)
 
 The feature-adr Step-10 Delivery Gate reviews a landed feature across four orthogonal planes and
@@ -584,7 +635,7 @@ Get the whole set with `dz init --target claude-code --preset meta`, or pick one
 
 > **A skill and its npx toolkit are not duplicates — they're a graduation.** Several skills (e.g. `feature-adr`, `design-thinking`) exist BOTH as a skill inside a `dz` preset AND as a standalone `npx` package. The preset's SKILL.md is **fully functional on its own** (the whole methodology — modules + references — travels with it, and it auto-activates by description), and it's the only way to compile that capability to the **non-Claude platforms** (Codex/OpenCode/Hermes/OpenClaude) via `dz`. The npx package adds **project-level runtime governance** around the same skill: a slash command, governance rules, a context shard, and (for feature-adr) reward-learning + `/harvest`. So: pick the **skill/preset** for a working capability across platforms; pick the **npx toolkit** when you want it as a governed, command-driven fixture of one project.
 
-## All Commands (57)
+## All Commands (58)
 
 ```
 dz setup             --target <name> [--preset <name>] [--select id,id,...] [--skills-dir <dir>] [--memory agentdb] [--no-memory] [--no-hooks] [--install-driver] [--force]
@@ -640,6 +691,7 @@ dz routing [--stage <s>] [--json]   # inspect the learned cost-optimal routing s
 dz bto-optimize --split|--plan|--select|--scope-check|--diff [--json]   # deterministic engine behind /bto-optimize: hold-out split + hard-capped budget + no-regress-on-holdout winner selection (defeats judge-gaming); prose-only, diff-confirmed, never auto-writes
 dz discrimination-check --test <f[,f]> [--base <ref>] [--name <filter>] [--runner <cmd>] [--json]   # §42 test-discrimination gate for feature-adr Step-8: run the ADR's property test in an isolated git worktree at pre-feature base — it MUST go red without the fix; a green is a false green (HIGH finding, advisory, never auto-aborts)
 dz delivery-check --slug <slug> [--context-only] [--findings <f.json>] [--strict] [--author <model>] [--json]   # portable Step-10 Delivery Gate: the `manual` form that travels to every shell target — prints the 4-plane review brief (regressions ‖ security ‖ code-quality ‖ product-honesty) + artifact probes; --findings classifies a fed-back review into a fail-closed ready|blocked hand-off (only cross-validated BLOCKER/HIGH count) and writes features/<slug>/10_delivery_review.md; --strict exits 1 on blocked
+dz skills-verify     [--dir <project>] [--expect a,b] [--static] [--strict] [--timeout <s>] [--json]   # does .claude/skills/ actually REGISTER? --static = instant layout scan (CI-safe, no session): flags dirs that can never register; default also starts a real session and reads the authoritative system/init listing. exit 0 pass / 1 fail / 2 inconclusive — an unobservable registration is NEVER a pass
 dz sign --init --out <path>  |  --pack <dir> --key <path>   # --init: generate the Ed25519 keypair (private OUTSIDE the repo, prints the public key for keys/dz.pub); else sign a pack's manifest + CycloneDX SBOM
 dz sbom --pack <dir> [--out <file>]   # emit the CycloneDX 1.5 SBOM for a pack standalone (file-level bill of materials); print to stdout or write to a file
 dz guard check --op <publish|teach|consolidate|reindex> [--text <s>] [--json] [--force <reason>]   # declarative constraint layer before self-mutating ops: HARD violation → block (exit 1), SOFT → warn; zero-config defaults, .dz/guard.json to customise; dz guard --init | dz guard log (append-only audit). dz publish runs it automatically (--no-guard "<reason>" = logged escape hatch)

package/dist/cli.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"cli.d.ts","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AA8QH,2EAA2E;AAC3E,MAAM,WAAW,KAAK;IACpB,QAAQ,CAAC,GAAG,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;IACxC;;;;OAIG;IACH,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB;;;;;OAKG;IACH,QAAQ,CAAC,aAAa,CAAC,EAAE,iBAAiB,CAAC;CAC5C;AAED,yFAAyF;AACzF,MAAM,MAAM,iBAAiB,GAAG,CAC9B,GAAG,EAAE,MAAM,EACX,IAAI,EAAE;IAAE,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;CAAE,KACvD;IAAE,QAAQ,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,EAAE,OAAO,CAAA;CAAE,CAAC;AAyiL9E,wBAAsB,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,GAAE,KAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAkJ5E"}
+{"version":3,"file":"cli.d.ts","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAoRH,2EAA2E;AAC3E,MAAM,WAAW,KAAK;IACpB,QAAQ,CAAC,GAAG,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;IACxC;;;;OAIG;IACH,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB;;;;;OAKG;IACH,QAAQ,CAAC,aAAa,CAAC,EAAE,iBAAiB,CAAC;CAC5C;AAED,yFAAyF;AACzF,MAAM,MAAM,iBAAiB,GAAG,CAC9B,GAAG,EAAE,MAAM,EACX,IAAI,EAAE;IAAE,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;CAAE,KACvD;IAAE,QAAQ,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,EAAE,OAAO,CAAA;CAAE,CAAC;AAiuL9E,wBAAsB,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,GAAE,KAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAoJ5E"}

package/dist/cli.js

@@ -6,10 +6,10 @@
 import { chmodSync, existsSync, lstatSync, mkdirSync, mkdtempSync, readFileSync, readdirSync, readlinkSync, realpathSync, renameSync, rmdirSync, rmSync, statSync, symlinkSync, writeFileSync } from 'node:fs';
 import { basename, dirname, isAbsolute, join, relative, resolve, sep } from 'node:path';
 import { fileURLToPath } from 'node:url';
-import { execSync } from 'node:child_process';
+import { execSync, spawn } from 'node:child_process';
 import { homedir, tmpdir } from 'node:os';
 import { createRequire } from 'node:module';
-import { createSkill, getSkillInfo, getWorkflow, isTargetName, listSkills, runDoctor, runInit, benchmarkSkill, benchmarkSkills, scanMcp, reconcileCapabilities, RECONCILE_BANNER, buildRegistry, discoverSkillPackDirs, checkUpstream, compareSkills, checkAllUpstream, sweepSkillDrift, syncCanonicalSkill, checkUpgrades, discoverPackages, discoverSourcePackages, fetchAllDownloads, filterByCategory, pretrain, recommend, generatePlugin, publishPackages, runSetup, runMigrate, searchRegistry, runSync, runVerify, runInitAgentsMd, runInitGeminiMd, TARGET_NAMES, buildParityMatrix, TARGET_CAPABILITIES, TARGET_SHORT_LABELS, WORKFLOW_NAMES, importEcc, recordPattern, resolveLearningBackend, storeStats, consolidateSessions, pruneNoisePatterns, lessonDeltaReport, removePatternsByIds, snapshotStore, recallHybrid, teachGuard, mirrorPatternsToVector, mirrorEntriesToVector, patternVectorEntry, readMemoryLearningConfig, promotePatterns, quarantineExpiryCandidates, pruneQuarantinePatterns, clearAgentdbQuarantine, vectorMirrorEnabled, vectorTierStatus, resolveVectorEngine, reindexVectorStore, harmonizeVectorStore, importRvfCheckpoint, statuslineData, writeFeatureAdrState, computeUsage, deriveUsageCalibration, normalizeClaudeUsageModelKey, readUsageLimits, claimCheck, summarize, queryBookKnowledge, loadStorePatternsSync, patternRecordId, loadStoreRecords, recordToPattern, bundleSkills, brainHome, listBrain, promoteProjectToBrain, updateBrainSource, queryBrain, groundPrompt, expandKu, reindexBrainVectors, buildPrimer, exportBrainSlice, importBrainSlice, registerKusToBrain, RECALL_USAGE_LOG_RELATIVE, RECALL_USAGE_LOG_MAX_BYTES, parseRecallUsageLog, buildRecallUsageReport, buildManifest, buildSbom, resolveTrustRoot, decideVerifyPolicy, generateSigningKeypair, evaluateGuard, resolveRules, auditRecord, guardExitCode, DEFAULT_RULES, decideProvenance, isInsideTree, signManifest, verifyManifest, listSignablePackFiles, assertKeyOutsideTree, decidePublishGate, collectPackageFacts, planReleaseGates, selectAffectedPackages, classifyGateExecutions, buildFailureIssue, buildReleaseNotes, releaseTagName, firstOutputLine, formatPublishError, MANIFEST_NAME, SBOM_NAME, buildArchitectureMap, renderMapHuman, findArchitectureDrift, renderDriftReport, scanWorkspacePackages, loadSubsystemManifest, loadProductVision, checkFeatureAgainstArchitecture, renderArchCheck, planProjectSkills, guidanceForStage, renderInjectionReport, analyzeCorpus, renderRakeReport, renderCriticSection, rakeAsLesson, rakeReward, DEFAULT_RAKE_THRESHOLDS, streamSessionEvents, findLatestTranscript, detectProcessRakes, buildRetro, renderRetro, retroLessonText, PROCESS_SIGNATURES, RETRO_DOMAIN, scanForSetup, buildSetupPlan, scaffoldFromSpec, renderScaffoldPreview, readExistingForScaffold, assembleChallengeContext, buildChallengeBrief, planDiscriminationCheck, classifyDiscrimination, pickAdversaryModel, CHALLENGE_QUESTIONS, loadOutcomes, renderOutcomes, statsForKey, selectAutoCost, recordProvisional, finalizeOutcome, COST_LADDER, splitScenarios, budgetPlan, selectWinner, proseScopeOk, renderProseDiff, readScenarioIds, DEFAULT_MAX_JUDGE_RUNS, collectDeliveryFacts, planDeliveryCheck, renderDeliveryBrief, classifyDelivery, isUsablePlaneResult, renderDeliveryReview, } from '@dzhechkov/harness-core';
+import { createSkill, getSkillInfo, getWorkflow, isTargetName, listSkills, runDoctor, runInit, benchmarkSkill, benchmarkSkills, scanMcp, reconcileCapabilities, RECONCILE_BANNER, buildRegistry, discoverSkillPackDirs, checkUpstream, compareSkills, checkAllUpstream, sweepSkillDrift, syncCanonicalSkill, checkUpgrades, discoverPackages, discoverSourcePackages, fetchAllDownloads, filterByCategory, pretrain, recommend, generatePlugin, publishPackages, runSetup, runMigrate, searchRegistry, runSync, runVerify, runInitAgentsMd, runInitGeminiMd, TARGET_NAMES, buildParityMatrix, TARGET_CAPABILITIES, TARGET_SHORT_LABELS, WORKFLOW_NAMES, importEcc, recordPattern, resolveLearningBackend, storeStats, consolidateSessions, pruneNoisePatterns, lessonDeltaReport, removePatternsByIds, snapshotStore, recallHybrid, teachGuard, mirrorPatternsToVector, mirrorEntriesToVector, patternVectorEntry, readMemoryLearningConfig, promotePatterns, quarantineExpiryCandidates, pruneQuarantinePatterns, clearAgentdbQuarantine, vectorMirrorEnabled, vectorTierStatus, resolveVectorEngine, reindexVectorStore, harmonizeVectorStore, importRvfCheckpoint, statuslineData, writeFeatureAdrState, computeUsage, deriveUsageCalibration, normalizeClaudeUsageModelKey, readUsageLimits, claimCheck, summarize, queryBookKnowledge, loadStorePatternsSync, patternRecordId, loadStoreRecords, recordToPattern, bundleSkills, brainHome, listBrain, promoteProjectToBrain, updateBrainSource, queryBrain, groundPrompt, expandKu, reindexBrainVectors, buildPrimer, exportBrainSlice, importBrainSlice, registerKusToBrain, RECALL_USAGE_LOG_RELATIVE, RECALL_USAGE_LOG_MAX_BYTES, parseRecallUsageLog, buildRecallUsageReport, buildManifest, buildSbom, resolveTrustRoot, decideVerifyPolicy, generateSigningKeypair, evaluateGuard, resolveRules, auditRecord, guardExitCode, DEFAULT_RULES, decideProvenance, isInsideTree, signManifest, verifyManifest, listSignablePackFiles, assertKeyOutsideTree, decidePublishGate, collectPackageFacts, planReleaseGates, selectAffectedPackages, classifyGateExecutions, buildFailureIssue, buildReleaseNotes, releaseTagName, firstOutputLine, formatPublishError, MANIFEST_NAME, SBOM_NAME, buildArchitectureMap, renderMapHuman, findArchitectureDrift, renderDriftReport, scanWorkspacePackages, loadSubsystemManifest, loadProductVision, checkFeatureAgainstArchitecture, renderArchCheck, planProjectSkills, guidanceForStage, renderInjectionReport, analyzeCorpus, renderRakeReport, renderCriticSection, rakeAsLesson, rakeReward, DEFAULT_RAKE_THRESHOLDS, streamSessionEvents, findLatestTranscript, detectProcessRakes, buildRetro, renderRetro, retroLessonText, PROCESS_SIGNATURES, RETRO_DOMAIN, scanForSetup, buildSetupPlan, scaffoldFromSpec, renderScaffoldPreview, readExistingForScaffold, assembleChallengeContext, buildChallengeBrief, planDiscriminationCheck, classifyDiscrimination, pickAdversaryModel, CHALLENGE_QUESTIONS, loadOutcomes, renderOutcomes, statsForKey, selectAutoCost, recordProvisional, finalizeOutcome, COST_LADDER, splitScenarios, budgetPlan, selectWinner, proseScopeOk, renderProseDiff, readScenarioIds, DEFAULT_MAX_JUDGE_RUNS, collectDeliveryFacts, planDeliveryCheck, renderDeliveryBrief, classifyDelivery, isUsablePlaneResult, renderDeliveryReview, scanSkillsLayout, parseInitFacts, verifyRegistration, registrationExitCode, renderRegistrationReport, } from '@dzhechkov/harness-core';
 import { getPreset, PRESET_NAMES } from '@dzhechkov/harness-presets';
 import { scanGitHub, analyzeRepo, generateReport, deepAnalyze, scanAllSources, ScoutMemory } from '@dzhechkov/scout';
 const USAGE = `dz - DZ cross-platform harness CLI
@@ -35,6 +35,7 @@ Usage:
   dz release [--filter <name>] [--tag] [--publish] [--json] [--dry-run] [--no-issue]   (VERIFIED release: 4 HARD gates in FRONT of dz publish — full package test suites, audit >=high, node --check of every dist/bin file, bin smoke-boot via "node <bin> --help" — any red gate STOPS the release (exit 1) + best-effort gh issue; all green ⇒ re-sign reminder, then prints the ready dz publish command (or chains with --publish); never duplicates publish's own gates)
   dz parity [--target <name>] [--json]   (the honest feature×target map, COMPUTED from the capability model — which harness feature is full / manual / absent on each of the ${TARGET_NAMES.length} targets, and via which form)
   dz delivery-check --slug <slug> [--context-only] [--findings <f.json>] [--strict] [--author <model>] [--json]   (portable Step-10 Delivery Gate: prints the 4-plane review brief + artifact probes; --findings classifies a fed-back review into a fail-closed ready|blocked hand-off and writes features/<slug>/10_delivery_review.md; --strict exits 1 on blocked)
+  dz skills-verify [--dir <project>] [--expect a,b] [--static] [--strict] [--json]   (does .claude/skills/ actually REGISTER? --static = instant layout scan for CI; default reads the authoritative system/init listing from a real session. exit 0 pass / 1 fail / 2 inconclusive — never a false pass)
   dz setup --target <name> [--preset <name>] [--select id,id,...] [--skills-dir <dir>] [--project <dir>] [--memory agentdb] [--no-memory] [--no-hooks] [--install-driver] [--force] [--enrich]
   dz teach "<pattern>" [--reward <0-1>] [--domain <name>] [--type rule|success-pattern|lesson-learned] [--project <dir>] [--no-mirror]   (--project pins the learned store to <dir>/.dz, not the cwd — pin to a canonical brain)
   dz teach --from-json <file> [--project <dir>] [--no-mirror]   (bulk-import a 'dz recall --all --json' export — share a learned store across machines)
@@ -5302,6 +5303,176 @@ function nameFor(t, outcome) {
  *   --author <model>         cosmetic: the reviewer to dispatch, printed alongside the brief
  *   --json                   machine contract { planesChecked, planesSkipped, findings, handoff, artifact }
  */
+/**
+ * Env vars an INHERITED Claude session leaks into a child. Left in place, the probe can silently
+ * read the parent's project instead of the target — the exact confound that made a hand-rolled
+ * 2026-07-23 probe untrustworthy until controls were added (feature skills-verify, ADR-001).
+ */
+const PROBE_SCRUB_ENV = [
+    'CLAUDE_CODE_CHILD_SESSION',
+    'CLAUDE_CODE_SESSION_ID',
+    'CLAUDECODE',
+    'CLAUDE_CODE_ENTRYPOINT',
+    'CLAUDE_CODE_EXECPATH',
+    'CLAUDE_PLUGIN_DATA',
+    'AI_AGENT',
+];
+/**
+ * Start a real Claude session in `projectDir` and capture its stream until the `system/init` event
+ * lands, then kill it — the model never answers, so the probe costs ~no tokens. Never throws:
+ * every failure becomes an `error` string, which the pure classifier turns into `inconclusive`.
+ */
+function probeInitStream(projectDir, timeoutMs) {
+    return new Promise((resolveProbe) => {
+        const env = { ...process.env };
+        for (const key of PROBE_SCRUB_ENV)
+            delete env[key];
+        env.CLAUDE_PROJECT_DIR = projectDir;
+        let out = '';
+        let err = '';
+        let settled = false;
+        let child;
+        const finish = (error) => {
+            if (settled)
+                return;
+            settled = true;
+            clearTimeout(timer);
+            try {
+                child?.kill('SIGTERM');
+            }
+            catch {
+                /* the child may already be gone */
+            }
+            resolveProbe({ stream: out, error });
+        };
+        const timer = setTimeout(() => finish(`no init event within ${Math.round(timeoutMs / 1000)}s (is \`claude\` logged in?)`), timeoutMs);
+        try {
+            // NOT `--bare`: that mode skips plugin credentials and fails with "Not logged in".
+            child = spawn('claude', ['-p', 'ok', '--output-format', 'stream-json', '--verbose'], {
+                cwd: projectDir,
+                env,
+                stdio: ['ignore', 'pipe', 'pipe'],
+            });
+        }
+        catch (error) {
+            finish(`cannot run \`claude\`: ${error instanceof Error ? error.message : String(error)}`);
+            return;
+        }
+        child.stdout?.on('data', (chunk) => {
+            out += chunk.toString();
+            // Reuse the pure parser so a partially-flushed line simply waits for more data.
+            if (parseInitFacts(out))
+                finish(null);
+        });
+        child.stderr?.on('data', (chunk) => {
+            err += chunk.toString();
+        });
+        child.on('error', (error) => finish(`cannot run \`claude\`: ${error.message}`));
+        child.on('close', (code) => {
+            if (parseInitFacts(out))
+                finish(null);
+            else
+                finish(err.trim() || `\`claude\` exited ${code} without an init event`);
+        });
+    });
+}
+/**
+ * `dz skills-verify` — does this project's `.claude/skills/` actually REGISTER? (feature
+ * skills-verify, ADR-001.) L1 static scan is instant and CI-safe; the live layer reads the
+ * authoritative `system/init` listing. Fail-closed: unobservable ⇒ inconclusive, never pass.
+ */
+async function cmdSkillsVerify(options, flags, cwd, write) {
+    const json = flags.has('json');
+    if (flags.has('help')) {
+        write('dz skills-verify [--dir <project>] [--expect a,b] [--static] [--strict] [--timeout <s>] [--json]');
+        write('  Verifies that a project\'s .claude/skills/ actually register in Claude Code.');
+        write('  --static  layout scan only (no Claude session, CI-safe): flags dirs that can never register');
+        write('  default   also starts a real session and reads the authoritative system/init listing');
+        write('  exit: 0 pass · 1 fail · 2 inconclusive (--strict makes inconclusive exit 1)');
+        return 0;
+    }
+    const allowedFlags = new Set(['json', 'help', 'static', 'strict']);
+    const allowedOptions = new Set(['dir', 'expect', 'timeout']);
+    const usage = '  allowed: --dir <project>, --expect a,b, --timeout <s>, --static, --strict, --json';
+    if (options.has('_positional_0')) {
+        const message = `unexpected argument "${options.get('_positional_0')}"`;
+        write(json ? JSON.stringify({ error: message, exitCode: 1 }) : `dz skills-verify: ${message}\n${usage}`);
+        return 1;
+    }
+    for (const flag of flags) {
+        if (!allowedFlags.has(flag)) {
+            write(json ? JSON.stringify({ error: `unknown option --${flag}`, exitCode: 1 }) : `dz skills-verify: unknown option --${flag}\n${usage}`);
+            return 1;
+        }
+    }
+    for (const key of options.keys()) {
+        if (key.startsWith('_positional_'))
+            continue;
+        if (!allowedOptions.has(key)) {
+            write(json ? JSON.stringify({ error: `unknown option --${key}`, exitCode: 1 }) : `dz skills-verify: unknown option --${key}\n${usage}`);
+            return 1;
+        }
+    }
+    const projectDir = resolve(cwd, options.get('dir') ?? '.');
+    const scan = scanSkillsLayout(projectDir);
+    const expected = options.has('expect')
+        ? (options.get('expect') ?? '')
+            .split(',')
+            .map((s) => s.trim())
+            .filter(Boolean)
+        : scan.registrable;
+    // ── L1 only: deterministic, no session, safe for CI ──
+    if (flags.has('static')) {
+        const exitCode = scan.findings.length > 0 ? 1 : 0;
+        if (json) {
+            write(JSON.stringify({ mode: 'static', ...scan, expected, exitCode }, null, 2));
+        }
+        else {
+            write(`dz skills-verify (static): ${scan.registrable.length} registrable skill dir(s) under ${scan.skillsRoot}`);
+            if (!scan.exists)
+                write('  no .claude/skills/ directory here');
+            for (const f of scan.findings)
+                write(`  [${f.kind}] ${f.detail}`);
+            write(scan.findings.length ? `  ${scan.findings.length} layout problem(s) — these can never register` : '  no layout problems found');
+            write('  (static is a PROXY — run without --static to read the real registration listing)');
+        }
+        return exitCode;
+    }
+    // ── L2: the authoritative listing ──
+    const timeoutSec = Number(options.get('timeout') ?? '180');
+    const timeoutMs = Number.isFinite(timeoutSec) && timeoutSec > 0 ? timeoutSec * 1000 : 180_000;
+    if (!json)
+        write(`dz skills-verify: starting a session in ${projectDir} to read the real registration listing…`);
+    const { stream, error } = await probeInitStream(projectDir, timeoutMs);
+    // `init.skills` carries names, not provenance: a USER-level skill of the same name would satisfy
+    // the expectation while the project's own copy stays broken. Collect the collisions so the
+    // classifier can refuse to attribute registration to this project (Codex QE #2).
+    const userSkillsDir = join(homedir(), '.claude', 'skills');
+    const ambiguous = expected.filter((name) => existsSync(join(userSkillsDir, name, 'SKILL.md')));
+    // realpath so a symlinked project still matches its canonical cwd (Codex QE #6).
+    const canonical = (p) => {
+        try {
+            return realpathSync(p);
+        }
+        catch {
+            return resolve(p);
+        }
+    };
+    const result = verifyRegistration({
+        projectDir,
+        scan,
+        probe: error === null ? { ok: true, stream } : { ok: false, error },
+        // The provenance check RAN (that is what `checked: true` asserts) — see `ambiguous` above.
+        provenance: { checked: true, ambiguous },
+        ...(options.has('expect') ? { expected } : {}),
+    }, { resolvePath: canonical });
+    const exitCode = registrationExitCode(result.verdict, flags.has('strict'));
+    if (json)
+        write(JSON.stringify({ mode: 'live', ...result, skillsRoot: scan.skillsRoot, exitCode }, null, 2));
+    else
+        write(renderRegistrationReport(result, scan));
+    return exitCode;
+}
 function cmdDeliveryCheck(options, flags, cwd, write) {
     let repoRoot = cwd;
     try {
@@ -5870,6 +6041,8 @@ export async function runCli(argv, io = {}) {
                 return cmdDiscriminationCheck(options, flags, cwd, write);
             case 'delivery-check':
                 return cmdDeliveryCheck(options, flags, cwd, write);
+            case 'skills-verify':
+                return cmdSkillsVerify(options, flags, cwd, write);
             case 'routing':
                 return cmdRouting(options, flags, cwd, write);
             case 'bto-optimize':