harnery 0.24.0 → 0.26.0

package/dist/core/harnesses/attestation.js

@@ -0,0 +1,171 @@
+/**
+ * Durable record of what an installed vendor CLI was actually observed doing
+ * (ADR 0038).
+ *
+ * The conformance bench proves Harnery's own planner and normalizer against a
+ * committed fixture. That is an adapter check, not a vendor check
+ * (ADR 0037). An attestation is the other half: one bounded live turn, its
+ * observations, and the vendor version they were observed on.
+ *
+ * A record stores structural facts only. No prompt text, no completion text,
+ * no host paths. It lives under the host's coordination directory and is never
+ * published.
+ */
+import { mkdirSync, readdirSync, readFileSync, renameSync, unlinkSync, writeFileSync, } from "node:fs";
+import { resolve } from "node:path";
+import { monorepoRoot } from "../agents/coord-client.js";
+import { stableDigest } from "../workflow/durable-record.js";
+export const ATTESTATION_SCHEMA_VERSION = 2;
+/** Dimensions one minimal live turn can honestly establish. Everything else
+ * needs a purpose-built scenario and stays outside the record rather than
+ * being guessed at. */
+export const ATTESTABLE_DIMENSIONS = ["invocation", "finalResult", "sessionId", "cost"];
+export function attestationsDir(opts = {}) {
+    const root = opts.coordRoot ?? monorepoRoot();
+    if (!root)
+        throw new Error("Not in a coord-aware repo (coord root resolved to null).");
+    return resolve(root, ".harnery", "harnesses", "attestations");
+}
+/** Stable identity of a declaration, so editing a claim invalidates the
+ * attestation that was recorded against the old one. */
+export function profileDigest(profile) {
+    return stableDigest({
+        id: profile.id,
+        binary: profile.binary,
+        capabilities: profile.capabilities,
+    });
+}
+function digestOf(record) {
+    return stableDigest(record);
+}
+export function sealAttestation(record) {
+    return { ...record, record_digest: digestOf(record) };
+}
+function attestationPath(harness, opts) {
+    if (!/^[a-z0-9][a-z0-9._-]*$/i.test(harness)) {
+        throw new Error(`unsafe harness id for an attestation path: ${harness}`);
+    }
+    return resolve(attestationsDir(opts), `${harness}.json`);
+}
+/** Replace-in-place write. Unlike a workflow record an attestation is meant to
+ * be re-recorded, so this is a mutable atomic swap rather than an immutable
+ * create. */
+export function writeAttestation(record, opts = {}) {
+    const path = attestationPath(record.harness, opts);
+    mkdirSync(attestationsDir(opts), { recursive: true, mode: 0o700 });
+    const temporary = `${path}.tmp-${process.pid}`;
+    try {
+        writeFileSync(temporary, `${JSON.stringify(record, null, 2)}\n`, {
+            encoding: "utf8",
+            mode: 0o600,
+        });
+        renameSync(temporary, path);
+    }
+    catch (error) {
+        try {
+            unlinkSync(temporary);
+        }
+        catch {
+            // The temp file may never have been created; the original error wins.
+        }
+        throw error;
+    }
+    return path;
+}
+/** Null for absent, unreadable, malformed, wrong-schema, or tampered records.
+ * A record that fails its own digest is discarded rather than trusted, because
+ * the whole point of the file is that it was not hand-written. */
+export function readAttestation(harness, opts = {}) {
+    let raw;
+    try {
+        raw = readFileSync(attestationPath(harness, opts), "utf8");
+    }
+    catch {
+        return null;
+    }
+    let parsed;
+    try {
+        parsed = JSON.parse(raw);
+    }
+    catch {
+        return null;
+    }
+    return validateAttestation(parsed, harness);
+}
+export function validateAttestation(value, harness) {
+    if (!value || typeof value !== "object")
+        return null;
+    const record = value;
+    if (record.schema_version !== ATTESTATION_SCHEMA_VERSION)
+        return null;
+    if (typeof record.harness !== "string" || (harness && record.harness !== harness))
+        return null;
+    if (typeof record.binary_version !== "string" || typeof record.profile_digest !== "string") {
+        return null;
+    }
+    if (typeof record.observed_at !== "string" || typeof record.record_digest !== "string")
+        return null;
+    if (!record.observations || typeof record.observations !== "object")
+        return null;
+    const { record_digest, ...body } = record;
+    if (digestOf(body) !== record_digest)
+        return null;
+    return record;
+}
+/** An attestation speaks only for the vendor version and declaration it was
+ * recorded against. */
+export function isAttestationCurrent(record, binaryVersion, profile, subscriptionOnly) {
+    if (!record || !binaryVersion)
+        return false;
+    if (record.binary_version !== binaryVersion)
+        return false;
+    if (subscriptionOnly !== undefined && record.subscription_only !== subscriptionOnly)
+        return false;
+    return record.profile_digest === profileDigest(profile);
+}
+export function listAttestations(opts = {}) {
+    let names;
+    try {
+        names = readdirSync(attestationsDir(opts));
+    }
+    catch {
+        return [];
+    }
+    const records = [];
+    for (const name of names) {
+        if (!name.endsWith(".json"))
+            continue;
+        const record = readAttestation(name.slice(0, -".json".length), opts);
+        if (record)
+            records.push(record);
+    }
+    return records.sort((a, b) => a.harness.localeCompare(b.harness));
+}
+/** Both harness-derived proof inputs, read once, for a workflow run
+ * (ADR 0038). Callers inject the result so the engine performs no capability
+ * lookups of its own. A harness with no current attestation simply has no
+ * citation; that absence is not a proof unknown. */
+export function harnessProofInputs(profiles, opts) {
+    const harnessEvidence = {};
+    const harnessAttestations = {};
+    for (const profile of profiles) {
+        harnessEvidence[profile.id] = { toolEvidence: profile.capabilities.toolEvidence };
+        let record = null;
+        try {
+            record = readAttestation(profile.id, opts);
+        }
+        catch {
+            // No coord root or unreadable store: run unattested rather than fail.
+            continue;
+        }
+        if (!isAttestationCurrent(record, opts.versionProbe(profile.binary), profile, opts.subscriptionOnly)) {
+            continue;
+        }
+        harnessAttestations[profile.id] = {
+            binary_version: record.binary_version,
+            observed_at: record.observed_at,
+            record_digest: record.record_digest,
+        };
+    }
+    return { harnessEvidence, harnessAttestations };
+}

package/dist/core/harnesses/bench.d.ts

@@ -1,13 +1,23 @@
+import type { HarnessAttestation } from "./attestation.js";
 import type { HarnessRegistry } from "./registry.js";
 import type { CapabilitySupport, HarnessCapabilityDimension, HarnessId } from "./types.js";
 export type BenchVerdict = "supported" | "partial" | "unsupported" | "not_applicable" | "unknown" | "skipped" | "drift";
-export type BenchDimension = "registration" | "binary" | HarnessCapabilityDimension;
+export type BenchDimension = "registration" | "binary" | "contract" | HarnessCapabilityDimension;
+/** How a result was established (ADR 0037). Orthogonal to the verdict: this
+ * answers "how do we know", not "what is true".
+ *
+ * - `adapter`: checked against Harnery's own planner, normalizer, or fixture.
+ *   Proves the adapter contract, says nothing about the installed vendor CLI.
+ * - `attested`: checked by observing the installed vendor CLI on this host.
+ * - `declared`: not checked. The value is the declaration, repeated. */
+export type BenchBasis = "adapter" | "attested" | "declared";
 export interface BenchResult {
     harness: HarnessId;
     dimension: BenchDimension;
     declared: CapabilitySupport | "not_applicable";
     observed: BenchVerdict;
     verdict: BenchVerdict;
+    basis: BenchBasis;
     note?: string;
 }
 export interface HarnessBenchReport {
@@ -16,6 +26,9 @@ export interface HarnessBenchReport {
     harnesses: HarnessId[];
     results: BenchResult[];
     summary: Record<BenchVerdict, number>;
+    /** Result counts per basis, so a caller can tell a clean report from an
+     * unmeasured one without walking every result. */
+    basisSummary: Record<BenchBasis, number>;
     drift: boolean;
     skipped: boolean;
 }
@@ -24,6 +37,18 @@ export interface HarnessBenchOptions {
     dimensions?: readonly HarnessCapabilityDimension[];
     /** Test seam and alternate host probe. A null version means unavailable. */
     versionProbe?: (binary: string) => string | null;
+    /** Where to look for live attestations (ADR 0038). Defaults to the coord
+     * root. Reading them never runs a model turn. */
+    coordRoot?: string;
+    /** Test seam. Defaults to reading the attestation store. */
+    attestationReader?: (harness: HarnessId) => HarnessAttestation | null;
+    /** Only cite attestations recorded under this billing mode. Omitted means
+     * any mode is acceptable for reporting purposes. */
+    subscriptionOnly?: boolean;
 }
 export declare function runHarnessBench(registry: HarnessRegistry, opts?: HarnessBenchOptions): HarnessBenchReport;
+/** Ask an installed vendor CLI what version it is. Null when it is absent or
+ * refuses to answer. Shared with the live attestation probe so both halves of
+ * the capability story key on the same string. */
+export declare function probeBinaryVersion(binary: string): string | null;
 //# sourceMappingURL=bench.d.ts.map

package/dist/core/harnesses/bench.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"bench.d.ts","sourceRoot":"","sources":["../../../src/core/harnesses/bench.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AACrD,OAAO,KAAK,EACV,iBAAiB,EAEjB,0BAA0B,EAC1B,SAAS,EACV,MAAM,YAAY,CAAC;AAGpB,MAAM,MAAM,YAAY,GACpB,WAAW,GACX,SAAS,GACT,aAAa,GACb,gBAAgB,GAChB,SAAS,GACT,SAAS,GACT,OAAO,CAAC;AAEZ,MAAM,MAAM,cAAc,GAAG,cAAc,GAAG,QAAQ,GAAG,0BAA0B,CAAC;AAEpF,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAE,SAAS,CAAC;IACnB,SAAS,EAAE,cAAc,CAAC;IAC1B,QAAQ,EAAE,iBAAiB,GAAG,gBAAgB,CAAC;IAC/C,QAAQ,EAAE,YAAY,CAAC;IACvB,OAAO,EAAE,YAAY,CAAC;IACtB,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,kBAAkB;IACjC,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,EAAE,SAAS,CAAC;IAChB,SAAS,EAAE,SAAS,EAAE,CAAC;IACvB,OAAO,EAAE,WAAW,EAAE,CAAC;IACvB,OAAO,EAAE,MAAM,CAAC,YAAY,EAAE,MAAM,CAAC,CAAC;IACtC,KAAK,EAAE,OAAO,CAAC;IACf,OAAO,EAAE,OAAO,CAAC;CAClB;AAED,MAAM,WAAW,mBAAmB;IAClC,SAAS,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IAC9B,UAAU,CAAC,EAAE,SAAS,0BAA0B,EAAE,CAAC;IACnD,4EAA4E;IAC5E,YAAY,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,MAAM,GAAG,IAAI,CAAC;CAClD;AAYD,wBAAgB,eAAe,CAC7B,QAAQ,EAAE,eAAe,EACzB,IAAI,GAAE,mBAAwB,GAC7B,kBAAkB,CAuDpB"}
+{"version":3,"file":"bench.d.ts","sourceRoot":"","sources":["../../../src/core/harnesses/bench.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAuB,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AAEhF,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AACrD,OAAO,KAAK,EACV,iBAAiB,EAEjB,0BAA0B,EAC1B,SAAS,EACV,MAAM,YAAY,CAAC;AAGpB,MAAM,MAAM,YAAY,GACpB,WAAW,GACX,SAAS,GACT,aAAa,GACb,gBAAgB,GAChB,SAAS,GACT,SAAS,GACT,OAAO,CAAC;AAEZ,MAAM,MAAM,cAAc,GAAG,cAAc,GAAG,QAAQ,GAAG,UAAU,GAAG,0BAA0B,CAAC;AAEjG;;;;;;wEAMwE;AACxE,MAAM,MAAM,UAAU,GAAG,SAAS,GAAG,UAAU,GAAG,UAAU,CAAC;AAE7D,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAE,SAAS,CAAC;IACnB,SAAS,EAAE,cAAc,CAAC;IAC1B,QAAQ,EAAE,iBAAiB,GAAG,gBAAgB,CAAC;IAC/C,QAAQ,EAAE,YAAY,CAAC;IACvB,OAAO,EAAE,YAAY,CAAC;IACtB,KAAK,EAAE,UAAU,CAAC;IAClB,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,kBAAkB;IACjC,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,EAAE,SAAS,CAAC;IAChB,SAAS,EAAE,SAAS,EAAE,CAAC;IACvB,OAAO,EAAE,WAAW,EAAE,CAAC;IACvB,OAAO,EAAE,MAAM,CAAC,YAAY,EAAE,MAAM,CAAC,CAAC;IACtC;sDACkD;IAClD,YAAY,EAAE,MAAM,CAAC,UAAU,EAAE,MAAM,CAAC,CAAC;IACzC,KAAK,EAAE,OAAO,CAAC;IACf,OAAO,EAAE,OAAO,CAAC;CAClB;AAED,MAAM,WAAW,mBAAmB;IAClC,SAAS,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IAC9B,UAAU,CAAC,EAAE,SAAS,0BAA0B,EAAE,CAAC;IACnD,4EAA4E;IAC5E,YAAY,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,MAAM,GAAG,IAAI,CAAC;IACjD;qDACiD;IACjD,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,4DAA4D;IAC5D,iBAAiB,CAAC,EAAE,CAAC,OAAO,EAAE,SAAS,KAAK,kBAAkB,GAAG,IAAI,CAAC;IACtE;wDACoD;IACpD,gBAAgB,CAAC,EAAE,OAAO,CAAC;CAC5B;AAkDD,wBAAgB,eAAe,CAC7B,QAAQ,EAAE,eAAe,EACzB,IAAI,GAAE,mBAAwB,GAC7B,kBAAkB,CAwEpB;AA2KD;;kDAEkD;AAClD,wBAAgB,kBAAkB,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAIhE"}

package/dist/core/harnesses/bench.js

@@ -1,5 +1,6 @@
 import { spawnSync } from "node:child_process";
 import { HARNESS_SPECS } from "../hooks/harness/events.js";
+import { isAttestationCurrent, readAttestation } from "./attestation.js";
 import { HARNESS_CAPABILITY_DIMENSIONS } from "./types.js";
 const EMPTY_SUMMARY = {
     supported: 0,
@@ -10,12 +11,44 @@ const EMPTY_SUMMARY = {
     skipped: 0,
     drift: 0,
 };
+const EMPTY_BASIS_SUMMARY = {
+    adapter: 0,
+    attested: 0,
+    declared: 0,
+};
+/** A stored attestation only counts when it still matches the installed
+ * version and the current declaration. A stale one is ignored rather than
+ * trusted, so an upgraded vendor CLI silently drops back to adapter basis. */
+function loadAttestation(id, version, adapter, opts) {
+    let record = null;
+    try {
+        record = opts.attestationReader
+            ? opts.attestationReader(id)
+            : readAttestation(id, { coordRoot: opts.coordRoot });
+    }
+    catch {
+        // No coord root, unreadable store: the bench still runs, just unattested.
+        return null;
+    }
+    return isAttestationCurrent(record, version, adapter.profile, opts.subscriptionOnly)
+        ? record
+        : null;
+}
+/** First version-shaped token in a string, or null when there is none.
+ * Tolerates the vendor prefixes and suffixes real CLIs print, so
+ * `codex-cli 0.144.5` and `2.1.197 (Claude Code)` both reduce to a token. */
+function versionToken(value) {
+    if (!value)
+        return null;
+    const match = value.match(/\d+(?:\.\d+)+(?:[-.][0-9a-z]+)*/i);
+    return match ? match[0].toLowerCase() : null;
+}
 export function runHarnessBench(registry, opts = {}) {
     const ids = opts.harnesses?.length ? [...opts.harnesses] : registry.ids();
     const dimensions = opts.dimensions?.length
         ? [...new Set(opts.dimensions)]
         : [...HARNESS_CAPABILITY_DIMENSIONS];
-    const versionProbe = opts.versionProbe ?? probeVersion;
+    const versionProbe = opts.versionProbe ?? probeBinaryVersion;
     const results = [];
     for (const id of ids) {
         const adapter = registry.require(id);
@@ -25,6 +58,7 @@ export function runHarnessBench(registry, opts = {}) {
             declared: "supported",
             observed: "supported",
             verdict: "supported",
+            basis: "adapter",
             note: `${adapter.profile.integrationMode}; ${adapter.profile.authModel}`,
         });
         const version = versionProbe(adapter.profile.binary);
@@ -34,35 +68,105 @@ export function runHarnessBench(registry, opts = {}) {
             declared: "supported",
             observed: version ? "supported" : "skipped",
             verdict: version ? "supported" : "skipped",
+            // A skip establishes nothing about the vendor, so it is not an attestation.
+            basis: version ? "attested" : "declared",
             note: version ?? `${adapter.profile.binary} not found on PATH`,
         });
+        results.push(contractResult(id, adapter, version));
         const observations = observeAdapter(adapter);
+        const attestation = loadAttestation(id, version, adapter, opts);
         for (const dimension of dimensions) {
             const claim = adapter.profile.capabilities[dimension];
-            const observed = observations[dimension];
+            const live = attestation?.observations[dimension];
+            // A live observation of the installed CLI outranks a fixture check of
+            // Harnery's own normalizer. Everything else keeps its adapter basis.
+            const { observed, basis } = live
+                ? { observed: live, basis: "attested" }
+                : observations[dimension];
             results.push({
                 harness: id,
                 dimension,
                 declared: claim.support,
                 observed,
                 verdict: reconcile(claim.support, observed),
-                note: claim.note,
+                basis,
+                note: live ? `attested on ${attestation?.binary_version}` : claim.note,
             });
         }
     }
     const summary = { ...EMPTY_SUMMARY };
-    for (const result of results)
+    const basisSummary = { ...EMPTY_BASIS_SUMMARY };
+    for (const result of results) {
         summary[result.verdict]++;
+        basisSummary[result.basis]++;
+    }
     return {
         generatedAt: new Date().toISOString(),
         mode: "offline",
         harnesses: ids,
         results,
         summary,
+        basisSummary,
         drift: summary.drift > 0,
         skipped: summary.skipped > 0,
     };
 }
+/** Compare the vendor contract a declaration was validated against with the
+ * one actually installed (ADR 0037). This is the only dimension attested
+ * without a model turn, because a version string costs nothing to read.
+ *
+ * A mismatch is `drift`, not failure: a newer vendor CLI is not automatically
+ * broken, but the declaration is no longer backed by an observation. An
+ * unrecorded or unparseable declared version stays `unknown` rather than being
+ * inferred from the installed one. */
+function contractResult(id, adapter, version) {
+    const recorded = adapter.profile.verified;
+    const base = { harness: id, dimension: "contract", declared: "supported" };
+    if (!version) {
+        return {
+            ...base,
+            observed: "skipped",
+            verdict: "skipped",
+            basis: "declared",
+            note: `${adapter.profile.binary} not found on PATH; installed contract unobservable`,
+        };
+    }
+    const declaredToken = versionToken(recorded?.version);
+    if (!declaredToken) {
+        return {
+            ...base,
+            observed: "unknown",
+            verdict: "unknown",
+            basis: "declared",
+            note: recorded
+                ? `verified.version "${recorded.version}" is not a version; installed ${version}`
+                : `no verified vendor contract recorded; installed ${version}`,
+        };
+    }
+    const installedToken = versionToken(version) ?? version.trim().toLowerCase();
+    if (installedToken === declaredToken) {
+        return {
+            ...base,
+            observed: "supported",
+            verdict: "supported",
+            basis: "attested",
+            note: `declaration validated against ${recorded?.version} on ${recorded?.date}`,
+        };
+    }
+    return {
+        ...base,
+        observed: "drift",
+        verdict: "drift",
+        basis: "attested",
+        note: `declaration validated against ${recorded?.version} (${recorded?.date}); installed ${version}`,
+    };
+}
+/** Checked against Harnery's planner, normalizer, or fixture. */
+function fromAdapter(observed) {
+    return { observed, basis: "adapter" };
+}
+/** No check exists for this dimension yet. */
+const NOT_CHECKED = { observed: "unknown", basis: "declared" };
 function observeAdapter(adapter) {
     const profile = adapter.profile;
     const effort = profile.effortValues[0];
@@ -98,43 +202,41 @@ function observeAdapter(adapter) {
         : undefined;
     const hookSubcommands = new Set(hookSpec?.events.map((event) => event.subcommand) ?? []);
     return {
-        invocation: !planningFailed && argv[0] === profile.binary && argv.includes(request.prompt)
+        invocation: fromAdapter(!planningFailed && argv[0] === profile.binary && argv.includes(request.prompt)
             ? "supported"
-            : "unsupported",
-        modelSelection: !planningFailed && argv.includes(request.model ?? "") ? "supported" : "unsupported",
-        effortSelection: effort === undefined
+            : "unsupported"),
+        modelSelection: fromAdapter(!planningFailed && argv.includes(request.model ?? "") ? "supported" : "unsupported"),
+        effortSelection: fromAdapter(effort === undefined
             ? "unsupported"
             : !planningFailed && argv.some((arg) => arg.includes(effort))
                 ? "supported"
-                : "unsupported",
-        maxTurns: !planningFailed && argv.includes(String(request.maxTurns)) ? "supported" : "unsupported",
-        finalResult: finalMatches ? "supported" : "unsupported",
-        sessionId: sessionObserved === "supported" && normalized?.sessionId !== fixture.sessionId
+                : "unsupported"),
+        maxTurns: fromAdapter(!planningFailed && argv.includes(String(request.maxTurns)) ? "supported" : "unsupported"),
+        finalResult: fromAdapter(finalMatches ? "supported" : "unsupported"),
+        sessionId: fromAdapter(sessionObserved === "supported" && normalized?.sessionId !== fixture.sessionId
             ? "unsupported"
-            : sessionObserved,
-        cost: costObserved === "supported" && normalized?.costUsd !== fixture.costUsd
+            : sessionObserved),
+        cost: fromAdapter(costObserved === "supported" && normalized?.costUsd !== fixture.costUsd
             ? "unsupported"
-            : costObserved,
-        toolEvidence: normalized && "toolEvidence" in normalized ? "supported" : "unsupported",
-        policyMapping: "unknown",
-        interruption: "unknown",
-        streaming: "unknown",
-        steering: "unknown",
-        resume: "unknown",
-        images: "unknown",
-        contextTelemetry: "unknown",
+            : costObserved),
+        toolEvidence: fromAdapter(normalized && "toolEvidence" in normalized ? "supported" : "unsupported"),
+        policyMapping: NOT_CHECKED,
+        interruption: NOT_CHECKED,
+        streaming: NOT_CHECKED,
+        steering: NOT_CHECKED,
+        resume: NOT_CHECKED,
+        images: NOT_CHECKED,
+        contextTelemetry: NOT_CHECKED,
         preCompactionSignal: hookSpec
-            ? hookSubcommands.has("pre-compact")
-                ? "supported"
-                : "unsupported"
-            : "unknown",
+            ? fromAdapter(hookSubcommands.has("pre-compact") ? "supported" : "unsupported")
+            : NOT_CHECKED,
         postCompactionSignal: hookSpec
-            ? hookSubcommands.has("post-compact") ||
+            ? fromAdapter(hookSubcommands.has("post-compact") ||
                 (adapter.profile.id === "claude-code" && hookSubcommands.has("session-start"))
                 ? "supported"
-                : "unsupported"
-            : "unknown",
-        compaction: "unknown",
+                : "unsupported")
+            : NOT_CHECKED,
+        compaction: NOT_CHECKED,
     };
 }
 function reconcile(declared, observed) {
@@ -143,7 +245,10 @@ function reconcile(declared, observed) {
     }
     return declared === observed ? observed : "drift";
 }
-function probeVersion(binary) {
+/** Ask an installed vendor CLI what version it is. Null when it is absent or
+ * refuses to answer. Shared with the live attestation probe so both halves of
+ * the capability story key on the same string. */
+export function probeBinaryVersion(binary) {
     const result = spawnSync(binary, ["--version"], { encoding: "utf8", timeout: 5_000 });
     if (result.error || result.status !== 0)
         return null;

package/dist/core/harnesses/index.d.ts

@@ -1,6 +1,10 @@
 export type { Spawner, SpawnRequest, SpawnResult } from "../workflow/types.js";
-export type { BenchDimension, BenchResult, BenchVerdict, HarnessBenchOptions, HarnessBenchReport, } from "./bench.js";
-export { runHarnessBench } from "./bench.js";
+export type { AttestationOutcome, HarnessAttestationReport, HarnessAttestationResult, RunHarnessAttestationOptions, } from "./attest.js";
+export { ATTESTATION_PROMPT, runHarnessAttestation } from "./attest.js";
+export type { AttestableDimension, AttestationStoreOptions, HarnessAttestation, } from "./attestation.js";
+export { ATTESTABLE_DIMENSIONS, ATTESTATION_SCHEMA_VERSION, attestationsDir, harnessProofInputs, isAttestationCurrent, listAttestations, profileDigest, readAttestation, validateAttestation, writeAttestation, } from "./attestation.js";
+export type { BenchBasis, BenchDimension, BenchResult, BenchVerdict, HarnessBenchOptions, HarnessBenchReport, } from "./bench.js";
+export { probeBinaryVersion, runHarnessBench } from "./bench.js";
 export type { BuiltinHarnessId } from "./profiles.js";
 export { BUILTIN_HARNESS_IDS, BUILTIN_HARNESS_PROFILES, builtinHarnessProfile, isBuiltinHarness, validateHarnessEffort, } from "./profiles.js";
 export { createBuiltinHarnessRegistry, HarnessRegistry } from "./registry.js";

package/dist/core/harnesses/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/core/harnesses/index.ts"],"names":[],"mappings":"AAAA,YAAY,EAAE,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AAC/E,YAAY,EACV,cAAc,EACd,WAAW,EACX,YAAY,EACZ,mBAAmB,EACnB,kBAAkB,GACnB,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAC7C,YAAY,EAAE,gBAAgB,EAAE,MAAM,eAAe,CAAC;AACtD,OAAO,EACL,mBAAmB,EACnB,wBAAwB,EACxB,qBAAqB,EACrB,gBAAgB,EAChB,qBAAqB,GACtB,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,4BAA4B,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAC9E,YAAY,EACV,eAAe,EACf,iBAAiB,EACjB,cAAc,EACd,mBAAmB,EACnB,mBAAmB,EACnB,0BAA0B,EAC1B,SAAS,EACT,iBAAiB,EACjB,cAAc,EACd,gBAAgB,GACjB,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,6BAA6B,EAAE,MAAM,YAAY,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/core/harnesses/index.ts"],"names":[],"mappings":"AAAA,YAAY,EAAE,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AAC/E,YAAY,EACV,kBAAkB,EAClB,wBAAwB,EACxB,wBAAwB,EACxB,4BAA4B,GAC7B,MAAM,aAAa,CAAC;AACrB,OAAO,EAAE,kBAAkB,EAAE,qBAAqB,EAAE,MAAM,aAAa,CAAC;AACxE,YAAY,EACV,mBAAmB,EACnB,uBAAuB,EACvB,kBAAkB,GACnB,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,qBAAqB,EACrB,0BAA0B,EAC1B,eAAe,EACf,kBAAkB,EAClB,oBAAoB,EACpB,gBAAgB,EAChB,aAAa,EACb,eAAe,EACf,mBAAmB,EACnB,gBAAgB,GACjB,MAAM,kBAAkB,CAAC;AAC1B,YAAY,EACV,UAAU,EACV,cAAc,EACd,WAAW,EACX,YAAY,EACZ,mBAAmB,EACnB,kBAAkB,GACnB,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,kBAAkB,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AACjE,YAAY,EAAE,gBAAgB,EAAE,MAAM,eAAe,CAAC;AACtD,OAAO,EACL,mBAAmB,EACnB,wBAAwB,EACxB,qBAAqB,EACrB,gBAAgB,EAChB,qBAAqB,GACtB,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,4BAA4B,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAC9E,YAAY,EACV,eAAe,EACf,iBAAiB,EACjB,cAAc,EACd,mBAAmB,EACnB,mBAAmB,EACnB,0BAA0B,EAC1B,SAAS,EACT,iBAAiB,EACjB,cAAc,EACd,gBAAgB,GACjB,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,6BAA6B,EAAE,MAAM,YAAY,CAAC"}

package/dist/core/harnesses/index.js

@@ -1,4 +1,6 @@
-export { runHarnessBench } from "./bench.js";
+export { ATTESTATION_PROMPT, runHarnessAttestation } from "./attest.js";
+export { ATTESTABLE_DIMENSIONS, ATTESTATION_SCHEMA_VERSION, attestationsDir, harnessProofInputs, isAttestationCurrent, listAttestations, profileDigest, readAttestation, validateAttestation, writeAttestation, } from "./attestation.js";
+export { probeBinaryVersion, runHarnessBench } from "./bench.js";
 export { BUILTIN_HARNESS_IDS, BUILTIN_HARNESS_PROFILES, builtinHarnessProfile, isBuiltinHarness, validateHarnessEffort, } from "./profiles.js";
 export { createBuiltinHarnessRegistry, HarnessRegistry } from "./registry.js";
 export { HARNESS_CAPABILITY_DIMENSIONS } from "./types.js";

package/dist/core/harnesses/profiles.d.ts

@@ -14,8 +14,8 @@ export declare const BUILTIN_HARNESS_PROFILES: {
         readonly modelFamily: "claude";
         readonly effortValues: readonly ["low", "medium", "high", "xhigh", "max"];
         readonly verified: {
-            readonly date: "2026-07-20";
-            readonly version: "current CLI contract";
+            readonly date: "2026-07-25";
+            readonly version: "2.1.197 (Claude Code)";
         };
         readonly capabilities: HarnessCapabilities;
     };
@@ -31,8 +31,8 @@ export declare const BUILTIN_HARNESS_PROFILES: {
         readonly modelFamily: "gpt";
         readonly effortValues: readonly ["none", "minimal", "low", "medium", "high", "xhigh"];
         readonly verified: {
-            readonly date: "2026-07-21";
-            readonly version: "codex-cli 0.145.0-alpha.18";
+            readonly date: "2026-07-25";
+            readonly version: "codex-cli 0.144.5";
         };
         readonly capabilities: HarnessCapabilities;
     };
@@ -48,8 +48,8 @@ export declare const BUILTIN_HARNESS_PROFILES: {
         readonly modelFamily: "multi";
         readonly effortValues: readonly [];
         readonly verified: {
-            readonly date: "2026-07-21";
-            readonly version: "2026.07.16-899851b";
+            readonly date: "2026-07-25";
+            readonly version: "2026.07.23-e383d2b";
         };
         readonly capabilities: HarnessCapabilities;
     };

package/dist/core/harnesses/profiles.js

@@ -39,7 +39,7 @@ export const BUILTIN_HARNESS_PROFILES = {
         authModel: "own-auth",
         modelFamily: "claude",
         effortValues: ["low", "medium", "high", "xhigh", "max"],
-        verified: { date: "2026-07-20", version: "current CLI contract" },
+        verified: { date: "2026-07-25", version: "2.1.197 (Claude Code)" },
         capabilities: capabilities({
             effortSelection: supported("Mapped to `--effort <level>`."),
             maxTurns: supported("Mapped to `--max-turns <n>`."),
@@ -61,7 +61,7 @@ export const BUILTIN_HARNESS_PROFILES = {
         authModel: "own-auth",
         modelFamily: "gpt",
         effortValues: ["none", "minimal", "low", "medium", "high", "xhigh"],
-        verified: { date: "2026-07-21", version: "codex-cli 0.145.0-alpha.18" },
+        verified: { date: "2026-07-25", version: "codex-cli 0.144.5" },
         capabilities: capabilities({
             effortSelection: supported('Mapped to `-c model_reasoning_effort="<level>"`.'),
             maxTurns: unsupported("codex exec exposes no turn-ceiling flag."),
@@ -83,7 +83,7 @@ export const BUILTIN_HARNESS_PROFILES = {
         authModel: "own-auth",
         modelFamily: "multi",
         effortValues: [],
-        verified: { date: "2026-07-21", version: "2026.07.16-899851b" },
+        verified: { date: "2026-07-25", version: "2026.07.23-e383d2b" },
         capabilities: capabilities({
             effortSelection: unsupported("Cursor embeds effort in some parameterized model ids; Harnery does not rewrite model ids."),
             maxTurns: unsupported("cursor-agent exposes no turn-ceiling flag."),

package/dist/core/workflow/engine.js

@@ -245,6 +245,7 @@ async function executeWorkflow(scriptPath, absScript, opts, resumeState, workCon
                 agents: [],
                 evidence: [],
                 harnessEvidence: opts.harnessEvidence,
+                harnessAttestations: opts.harnessAttestations,
                 policy: policy
                     ? {
                         config: policy,
@@ -925,6 +926,7 @@ async function executeWorkflow(scriptPath, absScript, opts, resumeState, workCon
                 agents: Array.from(agentProofs.values()),
                 evidence: evidenceRecords,
                 harnessEvidence: opts.harnessEvidence,
+                harnessAttestations: opts.harnessAttestations,
                 policy: policy
                     ? {
                         config: policy,
@@ -1002,6 +1004,7 @@ async function executeWorkflow(scriptPath, absScript, opts, resumeState, workCon
                 agents: Array.from(agentProofs.values()),
                 evidence: evidenceRecords,
                 harnessEvidence: opts.harnessEvidence,
+                harnessAttestations: opts.harnessAttestations,
                 policy: policy
                     ? {
                         config: policy,

package/dist/core/workflow/proof.d.ts

@@ -1,6 +1,6 @@
 import type { RepoSnapshot } from "../context/index.js";
 import type { NormalizedPolicy, PolicyDecision, PolicyIsolation, PolicyNetworkAccess } from "../policy/index.js";
-import type { AcceptanceCriterion, AcceptanceResult, AcceptanceSummary, HarnessEvidenceCapability, ResultDigest, WorkflowAgentProof, WorkflowAttemptContext, WorkflowEvidenceInput, WorkflowEvidenceRecord, WorkflowMeta, WorkflowProof, WorkflowWorkContext } from "./types.js";
+import type { AcceptanceCriterion, AcceptanceResult, AcceptanceSummary, HarnessAttestationCitation, HarnessEvidenceCapability, ResultDigest, WorkflowAgentProof, WorkflowAttemptContext, WorkflowEvidenceInput, WorkflowEvidenceRecord, WorkflowMeta, WorkflowProof, WorkflowWorkContext } from "./types.js";
 import type { WorkspaceAttestation, WorkspaceBinding, WorkspaceUnsupportedExecutionEvidence } from "./workspaces/index.js";
 export interface NormalizedWorkflowMeta {
     name: string;
@@ -24,6 +24,9 @@ export interface BuildWorkflowProofInput {
     agents: WorkflowAgentProof[];
     evidence: WorkflowEvidenceRecord[];
     harnessEvidence?: Readonly<Record<string, HarnessEvidenceCapability | undefined>>;
+    /** Live attestations backing each harness's claims (ADR 0038). Injected by
+     * the caller so proof stays free of filesystem lookups. */
+    harnessAttestations?: Readonly<Record<string, HarnessAttestationCitation | undefined>>;
     policy?: {
         config: Readonly<NormalizedPolicy>;
         decisions: readonly PolicyDecision[];

package/dist/core/workflow/proof.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"proof.d.ts","sourceRoot":"","sources":["../../../src/core/workflow/proof.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACxD,OAAO,KAAK,EACV,gBAAgB,EAChB,cAAc,EACd,eAAe,EACf,mBAAmB,EACpB,MAAM,oBAAoB,CAAC;AAK5B,OAAO,KAAK,EACV,mBAAmB,EACnB,gBAAgB,EAChB,iBAAiB,EACjB,yBAAyB,EAEzB,YAAY,EACZ,kBAAkB,EAClB,sBAAsB,EACtB,qBAAqB,EACrB,sBAAsB,EACtB,YAAY,EAEZ,aAAa,EAIb,mBAAmB,EACpB,MAAM,YAAY,CAAC;AAGpB,OAAO,KAAK,EACV,oBAAoB,EACpB,gBAAgB,EAEhB,qCAAqC,EACtC,MAAM,uBAAuB,CAAC;AAoB/B,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,mBAAmB,EAAE,CAAC;CACnC;AAED,MAAM,WAAW,uBAAuB;IACtC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,WAAW,CAAC,EAAE,QAAQ,CAAC,mBAAmB,CAAC,CAAC;IAC5C,cAAc,CAAC,EAAE,QAAQ,CAAC,sBAAsB,CAAC,CAAC;IAClD,IAAI,EAAE,sBAAsB,CAAC;IAC7B,MAAM,EAAE,WAAW,GAAG,QAAQ,CAAC;IAC/B,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,YAAY,CAAC;IACrB,KAAK,EAAE,YAAY,CAAC;IACpB,MAAM,EAAE,kBAAkB,EAAE,CAAC;IAC7B,QAAQ,EAAE,sBAAsB,EAAE,CAAC;IACnC,eAAe,CAAC,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,yBAAyB,GAAG,SAAS,CAAC,CAAC,CAAC;IAClF,MAAM,CAAC,EAAE;QACP,MAAM,EAAE,QAAQ,CAAC,gBAAgB,CAAC,CAAC;QACnC,SAAS,EAAE,SAAS,cAAc,EAAE,CAAC;QACrC,SAAS,EAAE,eAAe,CAAC;QAC3B,aAAa,EAAE,mBAAmB,CAAC;KACpC,CAAC;IACF,gBAAgB,CAAC,EAAE,gBAAgB,CAAC;IACpC,oBAAoB,CAAC,EAAE,oBAAoB,CAAC;IAC5C,iBAAiB,CAAC,EAAE,qCAAqC,CAAC;IAC1D,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,wBAAgB,qBAAqB,CACnC,IAAI,EAAE,YAAY,GAAG,SAAS,EAC9B,YAAY,EAAE,MAAM,GACnB,sBAAsB,CAqCxB;AAED,wBAAgB,oBAAoB,CAAC,KAAK,EAAE;IAC1C,KAAK,EAAE,qBAAqB,CAAC;IAC7B,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,EAAE,WAAW,CAAC,MAAM,CAAC,CAAC;IACnC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG,sBAAsB,CA6BzB;AAED,wBAAgB,gBAAgB,CAC9B,QAAQ,EAAE,mBAAmB,EAAE,EAC/B,QAAQ,EAAE,sBAAsB,EAAE,GACjC;IAAE,QAAQ,EAAE,gBAAgB,EAAE,CAAC;IAAC,OAAO,EAAE,iBAAiB,CAAA;CAAE,CAuB9D;AAED,wBAAgB,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,YAAY,CAQjF;AAED,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,uBAAuB,GAAG,aAAa,CAiDhF;AAgCD,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,aAAa,GAAG,IAAI,CAO3E;AAED,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,aAAa,CAgDjF;AA2BD,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,aAAa,GAAG,MAAM,CAoChE"}
+{"version":3,"file":"proof.d.ts","sourceRoot":"","sources":["../../../src/core/workflow/proof.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACxD,OAAO,KAAK,EACV,gBAAgB,EAChB,cAAc,EACd,eAAe,EACf,mBAAmB,EACpB,MAAM,oBAAoB,CAAC;AAK5B,OAAO,KAAK,EACV,mBAAmB,EACnB,gBAAgB,EAChB,iBAAiB,EACjB,0BAA0B,EAC1B,yBAAyB,EAEzB,YAAY,EACZ,kBAAkB,EAClB,sBAAsB,EACtB,qBAAqB,EACrB,sBAAsB,EACtB,YAAY,EAEZ,aAAa,EAIb,mBAAmB,EACpB,MAAM,YAAY,CAAC;AAGpB,OAAO,KAAK,EACV,oBAAoB,EACpB,gBAAgB,EAEhB,qCAAqC,EACtC,MAAM,uBAAuB,CAAC;AAoB/B,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,mBAAmB,EAAE,CAAC;CACnC;AAED,MAAM,WAAW,uBAAuB;IACtC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,WAAW,CAAC,EAAE,QAAQ,CAAC,mBAAmB,CAAC,CAAC;IAC5C,cAAc,CAAC,EAAE,QAAQ,CAAC,sBAAsB,CAAC,CAAC;IAClD,IAAI,EAAE,sBAAsB,CAAC;IAC7B,MAAM,EAAE,WAAW,GAAG,QAAQ,CAAC;IAC/B,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,YAAY,CAAC;IACrB,KAAK,EAAE,YAAY,CAAC;IACpB,MAAM,EAAE,kBAAkB,EAAE,CAAC;IAC7B,QAAQ,EAAE,sBAAsB,EAAE,CAAC;IACnC,eAAe,CAAC,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,yBAAyB,GAAG,SAAS,CAAC,CAAC,CAAC;IAClF;+DAC2D;IAC3D,mBAAmB,CAAC,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,0BAA0B,GAAG,SAAS,CAAC,CAAC,CAAC;IACvF,MAAM,CAAC,EAAE;QACP,MAAM,EAAE,QAAQ,CAAC,gBAAgB,CAAC,CAAC;QACnC,SAAS,EAAE,SAAS,cAAc,EAAE,CAAC;QACrC,SAAS,EAAE,eAAe,CAAC;QAC3B,aAAa,EAAE,mBAAmB,CAAC;KACpC,CAAC;IACF,gBAAgB,CAAC,EAAE,gBAAgB,CAAC;IACpC,oBAAoB,CAAC,EAAE,oBAAoB,CAAC;IAC5C,iBAAiB,CAAC,EAAE,qCAAqC,CAAC;IAC1D,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,wBAAgB,qBAAqB,CACnC,IAAI,EAAE,YAAY,GAAG,SAAS,EAC9B,YAAY,EAAE,MAAM,GACnB,sBAAsB,CAqCxB;AAED,wBAAgB,oBAAoB,CAAC,KAAK,EAAE;IAC1C,KAAK,EAAE,qBAAqB,CAAC;IAC7B,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,EAAE,WAAW,CAAC,MAAM,CAAC,CAAC;IACnC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG,sBAAsB,CA6BzB;AAED,wBAAgB,gBAAgB,CAC9B,QAAQ,EAAE,mBAAmB,EAAE,EAC/B,QAAQ,EAAE,sBAAsB,EAAE,GACjC;IAAE,QAAQ,EAAE,gBAAgB,EAAE,CAAC;IAAC,OAAO,EAAE,iBAAiB,CAAA;CAAE,CAuB9D;AAED,wBAAgB,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,YAAY,CAQjF;AAED,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,uBAAuB,GAAG,aAAa,CAiDhF;AAgCD,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,aAAa,GAAG,IAAI,CAO3E;AAED,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,aAAa,CAgDjF;AA2BD,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,aAAa,GAAG,MAAM,CAoChE"}