@jinn-network/client 0.1.6 → 0.1.7-canary.205587f0

package/dist/harnesses/impls/learner/harness.js

@@ -1,7 +1,9 @@
-import { CLAUDE_CODE_HARNESS } from '../../names.js';
+import { CLAUDE_CODE_HARNESS, CODEX_HARNESS, canonicalHarnessName } from '../../names.js';
 import { resolvePluginRoot } from './plugin-path.js';
 import { harvestOutput } from './harvest.js';
 import { buildClaudeIsReady } from '../../../preflight/claude-auth.js';
+import { probeCodexDoctor } from '../../../api/codex-doctor-endpoint.js';
+import { isLocalCodexBaseUrl } from './adapters/codex-code.js';
 /**
  * `Harness` shell. Bridges the engine's dispatch contract
  * (`await impl.run(ctx)`) into the harness adapter + markdown plugin.
@@ -16,6 +18,9 @@ export class LearnerHarness {
     adapter;
     pluginRoot;
     claudePath;
+    codexPath;
+    codexBaseUrl;
+    codexDoctorTimeoutMs;
     runtimeMode;
     constructor(config) {
         this.adapter = config.adapter;
@@ -23,12 +28,119 @@ export class LearnerHarness {
         this.version = config.version ?? '0.1.0-shim';
         this.pluginRoot = config.pluginRoot ?? resolvePluginRoot();
         this.claudePath = config.claudePath ?? 'claude';
+        this.codexPath = config.codexPath;
+        this.codexBaseUrl = config.codexBaseUrl;
+        this.codexDoctorTimeoutMs = config.codexDoctorTimeoutMs;
         this.runtimeMode = config.runtimeMode ?? 'bare';
     }
-    isReady = buildClaudeIsReady({
+    claudeIsReady = buildClaudeIsReady({
         getClaudePath: () => this.claudePath,
         getContext: () => this.runtimeMode,
     });
+    /**
+     * Readiness probe.
+     *
+     * The `LearnerHarness` shell backs two distinct CLIs: claude-code (the
+     * default) and Codex (`name === CODEX_HARNESS`). The probe MUST match the
+     * CLI actually invoked — delegating the Codex variant to the claude auth
+     * probe makes a missing/unconfigured `codex` install look always-ready and
+     * burns N/N failed claims (#348, the same-shape bug as #330).
+     *
+     *   - claude-code → `buildClaudeIsReady` (shells `claude auth status`).
+     *   - Codex       → `probeCodexDoctor` (shells `codex --version`, then
+     *     checks for `OPENAI_API_KEY` / a `codex login` auth file).
+     */
+    async isReady(ctx) {
+        if (canonicalHarnessName(this.name) === CODEX_HARNESS) {
+            return this.codexIsReady();
+        }
+        return this.claudeIsReady(ctx);
+    }
+    /**
+     * Codex-specific readiness probe. Shells `codex --version` via the shared
+     * `probeCodexDoctor` helper (same logic the SPA precheck endpoint uses).
+     * Reports:
+     *   - `installed: false` → binary not on PATH → ready=false with an install
+     *     nextStep so the operator sees an actionable message instead of N/N
+     *     failed claims (#348).
+     *   - `exitCode !== 0` → binary exists but `codex --version` failed →
+     *     ready=false pointing at the Codex precheck panel.
+     *   - local `codexBaseUrl` → binary runs and the local sidecar owns auth, so
+     *     Codex auth is not required.
+     *   - `authStatus: 'not_configured'` → binary runs but no `OPENAI_API_KEY`
+     *     and no `codex login` session → ready=false with a sign-in nextStep.
+     *   - `authStatus: 'expired'` → an `auth.json` is present but its OAuth
+     *     session has expired (or the file is malformed) → ready=false with a
+     *     re-login nextStep. Distinct from `not_configured` so a logged-out
+     *     operator with a leftover file is not treated as ready (#366).
+     *   - otherwise → ready=true.
+     */
+    codexIsReady() {
+        const config = {};
+        if (this.codexPath !== undefined)
+            config.codexPath = this.codexPath;
+        if (this.codexDoctorTimeoutMs !== undefined) {
+            config.codexDoctorTimeoutMs = this.codexDoctorTimeoutMs;
+        }
+        const result = probeCodexDoctor(config);
+        if (!result.installed) {
+            return {
+                ready: false,
+                reason: 'codex binary not installed',
+                nextStep: {
+                    description: 'Install the Codex CLI — see the Codex precheck panel in the operator dashboard for the install command.',
+                    url: '/api/codex/doctor',
+                },
+            };
+        }
+        if (result.exitCode !== 0) {
+            const stderr = result.stderr.trim();
+            const stdout = result.stdout.trim();
+            const detail = stderr.length > 0 ? stderr : stdout;
+            return {
+                ready: false,
+                reason: `codex --version exit ${result.exitCode}${detail ? `: ${detail}` : ''}`,
+                nextStep: {
+                    description: 'Run `codex --version` locally to surface the problem, or open the Codex precheck panel in the operator dashboard.',
+                    url: '/api/codex/doctor',
+                },
+            };
+        }
+        if (this.codexBaseUrl) {
+            if (isLocalCodexBaseUrl(this.codexBaseUrl)) {
+                return { ready: true };
+            }
+            return {
+                ready: false,
+                reason: 'codex base URL must be local',
+                nextStep: {
+                    description: 'Use a local Codex provider URL such as http://127.0.0.1:11434/v1 or unset JINN_CODEX_BASE_URL.',
+                    url: '/api/codex/doctor',
+                },
+            };
+        }
+        if (result.authStatus === 'expired') {
+            return {
+                ready: false,
+                reason: 'codex auth expired',
+                nextStep: {
+                    description: 'Codex sign-in has expired — run `codex login` to refresh the session (or set OPENAI_API_KEY), then re-check the Codex precheck panel in the operator dashboard.',
+                    url: '/api/codex/doctor',
+                },
+            };
+        }
+        if (result.authStatus !== 'ok') {
+            return {
+                ready: false,
+                reason: 'codex auth not configured',
+                nextStep: {
+                    description: 'Sign in to Codex — set OPENAI_API_KEY or run `codex login`, then re-check the Codex precheck panel in the operator dashboard.',
+                    url: '/api/codex/doctor',
+                },
+            };
+        }
+        return { ready: true };
+    }
     supports(spec) {
         if (spec.role === 'evaluation')
             return false;

package/dist/harnesses/impls/learner/harness.js.map

@@ -1 +1 @@
-{"version":3,"file":"harness.js","sourceRoot":"","sources":["../../../../src/harnesses/impls/learner/harness.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,mBAAmB,EAAE,MAAM,gBAAgB,CAAC;AAMrD,OAAO,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AACrD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAC7C,OAAO,EAAE,kBAAkB,EAAE,MAAM,mCAAmC,CAAC;AAEvE;;;;;;;GAOG;AACH,MAAM,OAAO,cAAc;IAChB,IAAI,CAAS;IACb,OAAO,CAAS;IACR,OAAO,CAAiB;IACxB,UAAU,CAAS;IACnB,UAAU,CAAS;IACnB,WAAW,CAA0C;IAEtE,YAAY,MAA4B;QACtC,IAAI,CAAC,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;QAC9B,IAAI,CAAC,IAAI,GAAG,MAAM,CAAC,IAAI,IAAI,mBAAmB,CAAC;QAC/C,IAAI,CAAC,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,YAAY,CAAC;QAC9C,IAAI,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU,IAAI,iBAAiB,EAAE,CAAC;QAC3D,IAAI,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU,IAAI,QAAQ,CAAC;QAChD,IAAI,CAAC,WAAW,GAAG,MAAM,CAAC,WAAW,IAAI,MAAM,CAAC;IAClD,CAAC;IAED,OAAO,GAAG,kBAAkB,CAAC;QAC3B,aAAa,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,UAAU;QACpC,UAAU,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,WAAW;KACnC,CAAC,CAAC;IAEH,QAAQ,CAAC,IAAiE;QACxE,IAAI,IAAI,CAAC,IAAI,KAAK,YAAY;YAAE,OAAO,KAAK,CAAC;QAC7C,uEAAuE;QACvE,2EAA2E;QAC3E,2EAA2E;QAC3E,sDAAsD;QACtD,EAAE;QACF,wEAAwE;QACxE,yEAAyE;QACzE,gEAAgE;QAChE,oEAAoE;QACpE,gEAAgE;QAChE,8EAA8E;QAC9E,0CAA0C;QAC1C,EAAE;QACF,4DAA4D;QAC5D,2EAA2E;QAC3E,sEAAsE;QACtE,sEAAsE;QACtE,kCAAkC;QAClC,IAAI,IAAI,CAAC,UAAU,KAAK,eAAe,IAAI,IAAI,CAAC,UAAU,KAAK,mBAAmB,EAAE,CAAC;YACnF,OAAO,KAAK,CAAC;QACf,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IAED,KAAK,CAAC,GAAG,CAAC,GAAmB;QAC3B,MAAM,MAAM,GAAG,GAAG,CAAC,IAAI,CAAC,MAAM,IAAI,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;QAC3D,MAAM,MAAM,GAAsB;YAChC,MAAM,EAAE,GAAG,CAAC,IAAI,CAAC,EAAE;YACnB,SAAS,EAAE,GAAG,CAAC,SAAS;YACxB,OAAO,EAAE,GAAG,CAAC,OAAO;YACpB,UAAU,EAAE,GAAG,CAAC,IAAI,CAAC,UAAU;YAC/B,KAAK,EAAE,GAAG,CAAC,SAAS,EAAE,KAAK;YAC3B,WAAW,EAAE,GAAG,CAAC,SAAS,EAAE,KAAK;YACjC,QAAQ,EAAE,GAAG,CAAC,IAAqC;YACnD,YAAY,EAAE,GAAG,CAAC,YAAY;YAC9B,UAAU,EAAE,GAAG,CAAC,UAAU;YAC1B,WAAW,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,iBAAiB,IAAI,EAAE,CAAC,CAAC;YAC/C,aAAa,EAAE,MAAM,CAAC,OAAO;YAC7B,WAAW,EAAE,MAAM,CAAC,KAAK;YACzB,YAAY,EAAE,GAAG,CAAC,YAAY,EAAE;YAChC,KAAK,EAAE,GAAG,CAAC,KAAK;YAChB,IAAI,EAAE,GAAG,CAAC,IAAI;SACf,CAAC;QAEF,MAAM,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,IAAI,CAAC,UAAU,CAAC,CAAC;QAEpD,MAAM,QAAQ,GAAG,MAAM,aAAa,CAAC,GAAG,CAAC,UAAU,EAAE,SAAS,EAAE,GAAG,CAAC,IAAI,CAAC,CAAC;QAC1E,OAAO;YACL,GAAG,QAAQ;YACX,QAAQ,EAAE,EAAE,GAAG,QAAQ,CAAC,QAAQ,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE;SACpD,CAAC;IACJ,CAAC;CACF"}
+{"version":3,"file":"harness.js","sourceRoot":"","sources":["../../../../src/harnesses/impls/learner/harness.ts"],"names":[],"mappings":"AAMA,OAAO,EAAE,mBAAmB,EAAE,aAAa,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAC;AAM1F,OAAO,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AACrD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAC7C,OAAO,EAAE,kBAAkB,EAAE,MAAM,mCAAmC,CAAC;AACvE,OAAO,EAAE,gBAAgB,EAAE,MAAM,uCAAuC,CAAC;AACzE,OAAO,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AAE/D;;;;;;;GAOG;AACH,MAAM,OAAO,cAAc;IAChB,IAAI,CAAS;IACb,OAAO,CAAS;IACR,OAAO,CAAiB;IACxB,UAAU,CAAS;IACnB,UAAU,CAAS;IACnB,SAAS,CAAqB;IAC9B,YAAY,CAAqB;IACjC,oBAAoB,CAAqB;IACzC,WAAW,CAA0C;IAEtE,YAAY,MAA4B;QACtC,IAAI,CAAC,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;QAC9B,IAAI,CAAC,IAAI,GAAG,MAAM,CAAC,IAAI,IAAI,mBAAmB,CAAC;QAC/C,IAAI,CAAC,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,YAAY,CAAC;QAC9C,IAAI,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU,IAAI,iBAAiB,EAAE,CAAC;QAC3D,IAAI,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU,IAAI,QAAQ,CAAC;QAChD,IAAI,CAAC,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;QAClC,IAAI,CAAC,YAAY,GAAG,MAAM,CAAC,YAAY,CAAC;QACxC,IAAI,CAAC,oBAAoB,GAAG,MAAM,CAAC,oBAAoB,CAAC;QACxD,IAAI,CAAC,WAAW,GAAG,MAAM,CAAC,WAAW,IAAI,MAAM,CAAC;IAClD,CAAC;IAEgB,aAAa,GAAG,kBAAkB,CAAC;QAClD,aAAa,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,UAAU;QACpC,UAAU,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,WAAW;KACnC,CAAC,CAAC;IAEH;;;;;;;;;;;;OAYG;IACH,KAAK,CAAC,OAAO,CACX,GAAiE;QAEjE,IAAI,oBAAoB,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,aAAa,EAAE,CAAC;YACtD,OAAO,IAAI,CAAC,YAAY,EAAE,CAAC;QAC7B,CAAC;QACD,OAAO,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC;IACjC,CAAC;IAED;;;;;;;;;;;;;;;;;;OAkBG;IACK,YAAY;QAClB,MAAM,MAAM,GAA0D,EAAE,CAAC;QACzE,IAAI,IAAI,CAAC,SAAS,KAAK,SAAS;YAAE,MAAM,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,CAAC;QACpE,IAAI,IAAI,CAAC,oBAAoB,KAAK,SAAS,EAAE,CAAC;YAC5C,MAAM,CAAC,oBAAoB,GAAG,IAAI,CAAC,oBAAoB,CAAC;QAC1D,CAAC;QACD,MAAM,MAAM,GAAG,gBAAgB,CAAC,MAAM,CAAC,CAAC;QACxC,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC;YACtB,OAAO;gBACL,KAAK,EAAE,KAAK;gBACZ,MAAM,EAAE,4BAA4B;gBACpC,QAAQ,EAAE;oBACR,WAAW,EACT,yGAAyG;oBAC3G,GAAG,EAAE,mBAAmB;iBACzB;aACF,CAAC;QACJ,CAAC;QACD,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;YAC1B,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;YACpC,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;YACpC,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;YACnD,OAAO;gBACL,KAAK,EAAE,KAAK;gBACZ,MAAM,EAAE,wBAAwB,MAAM,CAAC,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAC,KAAK,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE;gBAC/E,QAAQ,EAAE;oBACR,WAAW,EACT,mHAAmH;oBACrH,GAAG,EAAE,mBAAmB;iBACzB;aACF,CAAC;QACJ,CAAC;QACD,IAAI,IAAI,CAAC,YAAY,EAAE,CAAC;YACtB,IAAI,mBAAmB,CAAC,IAAI,CAAC,YAAY,CAAC,EAAE,CAAC;gBAC3C,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;YACzB,CAAC;YACD,OAAO;gBACL,KAAK,EAAE,KAAK;gBACZ,MAAM,EAAE,8BAA8B;gBACtC,QAAQ,EAAE;oBACR,WAAW,EACT,gGAAgG;oBAClG,GAAG,EAAE,mBAAmB;iBACzB;aACF,CAAC;QACJ,CAAC;QACD,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;YACpC,OAAO;gBACL,KAAK,EAAE,KAAK;gBACZ,MAAM,EAAE,oBAAoB;gBAC5B,QAAQ,EAAE;oBACR,WAAW,EACT,iKAAiK;oBACnK,GAAG,EAAE,mBAAmB;iBACzB;aACF,CAAC;QACJ,CAAC;QACD,IAAI,MAAM,CAAC,UAAU,KAAK,IAAI,EAAE,CAAC;YAC/B,OAAO;gBACL,KAAK,EAAE,KAAK;gBACZ,MAAM,EAAE,2BAA2B;gBACnC,QAAQ,EAAE;oBACR,WAAW,EACT,+HAA+H;oBACjI,GAAG,EAAE,mBAAmB;iBACzB;aACF,CAAC;QACJ,CAAC;QACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;IACzB,CAAC;IAED,QAAQ,CAAC,IAAiE;QACxE,IAAI,IAAI,CAAC,IAAI,KAAK,YAAY;YAAE,OAAO,KAAK,CAAC;QAC7C,uEAAuE;QACvE,2EAA2E;QAC3E,2EAA2E;QAC3E,sDAAsD;QACtD,EAAE;QACF,wEAAwE;QACxE,yEAAyE;QACzE,gEAAgE;QAChE,oEAAoE;QACpE,gEAAgE;QAChE,8EAA8E;QAC9E,0CAA0C;QAC1C,EAAE;QACF,4DAA4D;QAC5D,2EAA2E;QAC3E,sEAAsE;QACtE,sEAAsE;QACtE,kCAAkC;QAClC,IAAI,IAAI,CAAC,UAAU,KAAK,eAAe,IAAI,IAAI,CAAC,UAAU,KAAK,mBAAmB,EAAE,CAAC;YACnF,OAAO,KAAK,CAAC;QACf,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IAED,KAAK,CAAC,GAAG,CAAC,GAAmB;QAC3B,MAAM,MAAM,GAAG,GAAG,CAAC,IAAI,CAAC,MAAM,IAAI,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;QAC3D,MAAM,MAAM,GAAsB;YAChC,MAAM,EAAE,GAAG,CAAC,IAAI,CAAC,EAAE;YACnB,SAAS,EAAE,GAAG,CAAC,SAAS;YACxB,OAAO,EAAE,GAAG,CAAC,OAAO;YACpB,UAAU,EAAE,GAAG,CAAC,IAAI,CAAC,UAAU;YAC/B,KAAK,EAAE,GAAG,CAAC,SAAS,EAAE,KAAK;YAC3B,WAAW,EAAE,GAAG,CAAC,SAAS,EAAE,KAAK;YACjC,QAAQ,EAAE,GAAG,CAAC,IAAqC;YACnD,YAAY,EAAE,GAAG,CAAC,YAAY;YAC9B,UAAU,EAAE,GAAG,CAAC,UAAU;YAC1B,WAAW,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,iBAAiB,IAAI,EAAE,CAAC,CAAC;YAC/C,aAAa,EAAE,MAAM,CAAC,OAAO;YAC7B,WAAW,EAAE,MAAM,CAAC,KAAK;YACzB,YAAY,EAAE,GAAG,CAAC,YAAY,EAAE;YAChC,KAAK,EAAE,GAAG,CAAC,KAAK;YAChB,IAAI,EAAE,GAAG,CAAC,IAAI;SACf,CAAC;QAEF,MAAM,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,IAAI,CAAC,UAAU,CAAC,CAAC;QAEpD,MAAM,QAAQ,GAAG,MAAM,aAAa,CAAC,GAAG,CAAC,UAAU,EAAE,SAAS,EAAE,GAAG,CAAC,IAAI,CAAC,CAAC;QAC1E,OAAO;YACL,GAAG,QAAQ;YACX,QAAQ,EAAE,EAAE,GAAG,QAAQ,CAAC,QAAQ,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE;SACpD,CAAC;IACJ,CAAC;CACF"}

package/dist/harnesses/impls/learner/plugin-path.d.ts

@@ -1,14 +1 @@
-/**
- * Resolve the learner plugin root from the impl directory's
- * runtime location.
- *
- * Layout assumption: this file lives at
- *   <package>/<src-or-dist>/harnesses/impls/learner/plugin-path.{ts,js}
- * and the plugin lives at
- *   <package>/plugins/learner/
- *
- * Walks up four directories from this file (impls → harness → src/dist →
- * package root) then descends into plugins/learner/. Verifies the
- * expected layout exists and throws with a clear message if not.
- */
 export declare function resolvePluginRoot(): string;

package/dist/harnesses/impls/learner/plugin-path.js

@@ -2,29 +2,49 @@ import { existsSync } from 'node:fs';
 import { dirname, join, resolve } from 'node:path';
 import { fileURLToPath } from 'node:url';
 /**
- * Resolve the learner plugin root from the impl directory's
- * runtime location.
+ * Resolve the learner plugin root from the impl directory's runtime location.
  *
- * Layout assumption: this file lives at
- *   <package>/<src-or-dist>/harnesses/impls/learner/plugin-path.{ts,js}
- * and the plugin lives at
- *   <package>/plugins/learner/
+ * Canonical layout (compiled): this file lives at
+ *   <package>/dist/harnesses/impls/learner/plugin-path.js
+ * and the plugin tree is co-located inside dist/:
+ *   <package>/dist/plugins/learner/
  *
- * Walks up four directories from this file (impls → harness → src/dist →
- * package root) then descends into plugins/learner/. Verifies the
- * expected layout exists and throws with a clear message if not.
+ * The build step copies plugins/learner → dist/plugins/learner so the
+ * compiled code and the plugin runtime assets are always version-locked
+ * within the same dist/ tree — for published npm installs and source
+ * checkouts alike (after `yarn build`).
+ *
+ * Source-checkout / tsx path (e.g. vitest): this file lives at
+ *   <package>/src/harnesses/impls/learner/plugin-path.ts
+ * Walking up four directories reaches the package root, and the plugin
+ * lives at <package>/plugins/learner/.
+ *
+ * Detection: if the resolved path three levels up ends with `/dist` the
+ * runtime is inside a compiled dist tree; otherwise fall back to the
+ * package-root layout (four levels up).
  */
+function requireAsset(pluginRoot, relative, hint) {
+    if (!existsSync(join(pluginRoot, relative))) {
+        const suffix = hint ? ` — ${hint}` : '';
+        throw new Error(`learner plugin at ${pluginRoot} is missing ${relative}${suffix}`);
+    }
+}
 export function resolvePluginRoot() {
     const here = dirname(fileURLToPath(import.meta.url));
-    const packageRoot = resolve(here, '..', '..', '..', '..');
-    const pluginRoot = join(packageRoot, 'plugins', 'learner');
+    const threeUp = resolve(here, '..', '..', '..');
+    // Compiled dist layout: dist/harnesses/impls/learner → dist/ → dist/plugins/learner
+    const isDistTree = threeUp.endsWith('/dist');
+    const pluginRoot = isDistTree
+        ? join(threeUp, 'plugins', 'learner')
+        : join(resolve(here, '..', '..', '..', '..'), 'plugins', 'learner');
     if (!existsSync(pluginRoot)) {
         throw new Error(`learner plugin not found at expected path: ${pluginRoot}. ` +
-            `Resolved from impl dir: ${here}.`);
-    }
-    if (!existsSync(join(pluginRoot, 'skills', 'learn', 'SKILL.md'))) {
-        throw new Error(`learner plugin at ${pluginRoot} is missing skills/learn/SKILL.md`);
+            `Resolved from impl dir: ${here}. ` +
+            `Run \`yarn build\` to copy plugins/learner into dist/plugins/learner.`);
     }
+    requireAsset(pluginRoot, 'skills/learn/SKILL.md');
+    requireAsset(pluginRoot, 'hooks/session-start', 'plugin assets may be stale or incomplete; rebuild the plugin');
+    requireAsset(pluginRoot, 'hooks/hooks.json', 'plugin assets may be stale or incomplete; rebuild the plugin');
     return pluginRoot;
 }
 //# sourceMappingURL=plugin-path.js.map

package/dist/harnesses/impls/learner/plugin-path.js.map

@@ -1 +1 @@
-{"version":3,"file":"plugin-path.js","sourceRoot":"","sources":["../../../../src/harnesses/impls/learner/plugin-path.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACrC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACnD,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAEzC;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,iBAAiB;IAC/B,MAAM,IAAI,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;IACrD,MAAM,WAAW,GAAG,OAAO,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;IAC1D,MAAM,UAAU,GAAG,IAAI,CAAC,WAAW,EAAE,SAAS,EAAE,SAAS,CAAC,CAAC;IAE3D,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,EAAE,CAAC;QAC5B,MAAM,IAAI,KAAK,CACb,8CAA8C,UAAU,IAAI;YAC1D,2BAA2B,IAAI,GAAG,CACrC,CAAC;IACJ,CAAC;IACD,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,UAAU,EAAE,QAAQ,EAAE,OAAO,EAAE,UAAU,CAAC,CAAC,EAAE,CAAC;QACjE,MAAM,IAAI,KAAK,CACb,qBAAqB,UAAU,mCAAmC,CACnE,CAAC;IACJ,CAAC;IACD,OAAO,UAAU,CAAC;AACpB,CAAC"}
+{"version":3,"file":"plugin-path.js","sourceRoot":"","sources":["../../../../src/harnesses/impls/learner/plugin-path.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACrC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACnD,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAEzC;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,SAAS,YAAY,CAAC,UAAkB,EAAE,QAAgB,EAAE,IAAa;IACvE,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAC,EAAE,CAAC;QAC5C,MAAM,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACxC,MAAM,IAAI,KAAK,CAAC,qBAAqB,UAAU,eAAe,QAAQ,GAAG,MAAM,EAAE,CAAC,CAAC;IACrF,CAAC;AACH,CAAC;AAED,MAAM,UAAU,iBAAiB;IAC/B,MAAM,IAAI,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;IACrD,MAAM,OAAO,GAAG,OAAO,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;IAEhD,oFAAoF;IACpF,MAAM,UAAU,GAAG,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;IAC7C,MAAM,UAAU,GAAG,UAAU;QAC3B,CAAC,CAAC,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,SAAS,CAAC;QACrC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,EAAE,SAAS,EAAE,SAAS,CAAC,CAAC;IAEtE,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,EAAE,CAAC;QAC5B,MAAM,IAAI,KAAK,CACb,8CAA8C,UAAU,IAAI;YAC1D,2BAA2B,IAAI,IAAI;YACnC,uEAAuE,CAC1E,CAAC;IACJ,CAAC;IACD,YAAY,CAAC,UAAU,EAAE,uBAAuB,CAAC,CAAC;IAClD,YAAY,CAAC,UAAU,EAAE,qBAAqB,EAAE,8DAA8D,CAAC,CAAC;IAChH,YAAY,CAAC,UAAU,EAAE,kBAAkB,EAAE,8DAA8D,CAAC,CAAC;IAC7G,OAAO,UAAU,CAAC;AACpB,CAAC"}

package/dist/harnesses/impls/learner/types.d.ts

@@ -121,4 +121,19 @@ export interface LearnerHarnessConfig {
      * Defaults to 'bare'.
      */
     runtimeMode?: 'bare' | 'container' | 'docker-compose';
+    /**
+     * Path to the `codex` executable. Used by `isReady()` when this
+     * `LearnerHarness` is the Codex variant (`name === CODEX_HARNESS`) — it is
+     * passed to `probeCodexDoctor()`. Defaults to 'codex' (from PATH).
+     */
+    codexPath?: string;
+    /** Default Codex model when a SolverNet does not specify one. */
+    codexModel?: string;
+    /** Local OpenAI-compatible Codex provider base URL. */
+    codexBaseUrl?: string;
+    /**
+     * Timeout (ms) for the `codex --version` probe in the Codex variant's
+     * `isReady()`. Defaults to 30s.
+     */
+    codexDoctorTimeoutMs?: number;
 }

package/dist/harnesses/impls/stub.d.ts

@@ -0,0 +1,58 @@
+/**
+ * Env-gated stub harness for the T2.2 producer/evaluator gate.
+ *
+ * When JINN_HARNESS_STUB_INSTANCE is set, the canned patch at
+ * <fixturesDir>/<instanceMatcher>.patch is returned as a SWE-rebench v2
+ * restoration solution. Never calls an LLM; never accepts tasks whose
+ * spec.instance_id differs from the configured matcher.
+ *
+ * PRODUCTION SAFETY — two-env-var requirement.
+ * This is a *fake* harness: it produces canned, non-genuine work. If it ever
+ * entered a real operator run it would generate fraudulent on-chain activity.
+ * To make accidental activation impossible, the factory requires BOTH:
+ *   JINN_HARNESS_STUB_INSTANCE     — instance ID this stub responds to
+ *   JINN_TEST_MODE === '1'         — explicit test-mode sentinel
+ * If JINN_HARNESS_STUB_INSTANCE is set but JINN_TEST_MODE is not '1', the
+ * factory THROWS rather than silently registering the stub. A single stray
+ * exported env var in an operator's shell can no longer activate it.
+ *
+ * Activated by environment variables:
+ *   JINN_HARNESS_STUB_INSTANCE     — instance ID this stub responds to (required to activate)
+ *   JINN_TEST_MODE                 — must equal '1' (defense-in-depth; required to activate)
+ *   JINN_HARNESS_STUB_FIXTURES_DIR — dir containing <instanceMatcher>.patch files
+ *                                    (default: client/test/release/tier-2/fixtures)
+ */
+import type { Harness, HarnessContext, ReadyStatus, Solution } from '../types.js';
+export interface StubHarnessConfig {
+    /** Directory containing <instanceMatcher>.patch files. */
+    fixturesDir: string;
+    /** The instance ID this stub will accept. Tasks with other instance IDs are rejected. */
+    instanceMatcher: string;
+}
+/**
+ * A zero-LLM Harness that returns a canned patch for a specific SWE-rebench v2
+ * instance. Intended exclusively for T2.2 release-gate automation.
+ */
+export declare class StubHarness implements Harness {
+    readonly name = "harness:stub";
+    readonly version = "0.1.0-stub";
+    private readonly fixturesDir;
+    private readonly instanceMatcher;
+    constructor(config: StubHarnessConfig);
+    supports(ctx: {
+        solverType: string;
+        role?: 'restoration' | 'evaluation';
+    }): boolean;
+    isReady(): Promise<ReadyStatus>;
+    run(ctx: HarnessContext): Promise<Solution>;
+}
+/**
+ * Factory that reads JINN_HARNESS_STUB_INSTANCE and JINN_HARNESS_STUB_FIXTURES_DIR
+ * from the environment and returns a configured StubHarness, or null if the env
+ * var is absent (allowing the registry to skip registration silently).
+ *
+ * Defense-in-depth: if JINN_HARNESS_STUB_INSTANCE is set but JINN_TEST_MODE is
+ * not exactly '1', this THROWS rather than returning a harness — a real
+ * operator run must never silently pick up the fake stub harness.
+ */
+export declare function maybeCreateStubHarnessFromEnv(): StubHarness | null;

package/dist/harnesses/impls/stub.js

@@ -0,0 +1,89 @@
+/**
+ * Env-gated stub harness for the T2.2 producer/evaluator gate.
+ *
+ * When JINN_HARNESS_STUB_INSTANCE is set, the canned patch at
+ * <fixturesDir>/<instanceMatcher>.patch is returned as a SWE-rebench v2
+ * restoration solution. Never calls an LLM; never accepts tasks whose
+ * spec.instance_id differs from the configured matcher.
+ *
+ * PRODUCTION SAFETY — two-env-var requirement.
+ * This is a *fake* harness: it produces canned, non-genuine work. If it ever
+ * entered a real operator run it would generate fraudulent on-chain activity.
+ * To make accidental activation impossible, the factory requires BOTH:
+ *   JINN_HARNESS_STUB_INSTANCE     — instance ID this stub responds to
+ *   JINN_TEST_MODE === '1'         — explicit test-mode sentinel
+ * If JINN_HARNESS_STUB_INSTANCE is set but JINN_TEST_MODE is not '1', the
+ * factory THROWS rather than silently registering the stub. A single stray
+ * exported env var in an operator's shell can no longer activate it.
+ *
+ * Activated by environment variables:
+ *   JINN_HARNESS_STUB_INSTANCE     — instance ID this stub responds to (required to activate)
+ *   JINN_TEST_MODE                 — must equal '1' (defense-in-depth; required to activate)
+ *   JINN_HARNESS_STUB_FIXTURES_DIR — dir containing <instanceMatcher>.patch files
+ *                                    (default: client/test/release/tier-2/fixtures)
+ */
+import * as fs from 'node:fs/promises';
+import * as path from 'node:path';
+/**
+ * A zero-LLM Harness that returns a canned patch for a specific SWE-rebench v2
+ * instance. Intended exclusively for T2.2 release-gate automation.
+ */
+export class StubHarness {
+    name = 'harness:stub';
+    version = '0.1.0-stub';
+    fixturesDir;
+    instanceMatcher;
+    constructor(config) {
+        this.fixturesDir = config.fixturesDir;
+        this.instanceMatcher = config.instanceMatcher;
+    }
+    supports(ctx) {
+        if (ctx.role === 'evaluation')
+            return false;
+        return ctx.solverType === 'swe-rebench-v2.v1';
+    }
+    async isReady() {
+        return { ready: true };
+    }
+    async run(ctx) {
+        const taskInstanceId = ctx.task.spec?.['instance_id'];
+        if (taskInstanceId !== this.instanceMatcher) {
+            throw new Error(`stub harness: task.spec.instance_id=${String(taskInstanceId)} does not match configured instanceMatcher=${this.instanceMatcher}`);
+        }
+        const patchPath = path.join(this.fixturesDir, `${this.instanceMatcher}.patch`);
+        const patch = await fs.readFile(patchPath, 'utf-8');
+        return {
+            venueRef: { name: this.name },
+            gating: {},
+            solutionPayload: {
+                schemaVersion: 'swe-rebench-v2-solution.v1',
+                patch,
+            },
+        };
+    }
+}
+/**
+ * Factory that reads JINN_HARNESS_STUB_INSTANCE and JINN_HARNESS_STUB_FIXTURES_DIR
+ * from the environment and returns a configured StubHarness, or null if the env
+ * var is absent (allowing the registry to skip registration silently).
+ *
+ * Defense-in-depth: if JINN_HARNESS_STUB_INSTANCE is set but JINN_TEST_MODE is
+ * not exactly '1', this THROWS rather than returning a harness — a real
+ * operator run must never silently pick up the fake stub harness.
+ */
+export function maybeCreateStubHarnessFromEnv() {
+    const instanceMatcher = process.env['JINN_HARNESS_STUB_INSTANCE'];
+    if (!instanceMatcher)
+        return null;
+    if (process.env['JINN_TEST_MODE'] !== '1') {
+        throw new Error('stub harness must never activate in a real operator run: ' +
+            'JINN_HARNESS_STUB_INSTANCE is set but JINN_TEST_MODE is not "1". ' +
+            'The stub harness produces canned, non-genuine work and would generate ' +
+            'fraudulent on-chain activity. Set JINN_TEST_MODE=1 if this is a Tier 2 ' +
+            'test; otherwise unset JINN_HARNESS_STUB_INSTANCE.');
+    }
+    const fixturesDir = process.env['JINN_HARNESS_STUB_FIXTURES_DIR'] ??
+        path.resolve(process.cwd(), 'test', 'release', 'tier-2', 'fixtures');
+    return new StubHarness({ instanceMatcher, fixturesDir });
+}
+//# sourceMappingURL=stub.js.map

package/dist/harnesses/impls/stub.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"stub.js","sourceRoot":"","sources":["../../../src/harnesses/impls/stub.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAEH,OAAO,KAAK,EAAE,MAAM,kBAAkB,CAAC;AACvC,OAAO,KAAK,IAAI,MAAM,WAAW,CAAC;AAUlC;;;GAGG;AACH,MAAM,OAAO,WAAW;IACb,IAAI,GAAG,cAAc,CAAC;IACtB,OAAO,GAAG,YAAY,CAAC;IAEf,WAAW,CAAS;IACpB,eAAe,CAAS;IAEzC,YAAY,MAAyB;QACnC,IAAI,CAAC,WAAW,GAAG,MAAM,CAAC,WAAW,CAAC;QACtC,IAAI,CAAC,eAAe,GAAG,MAAM,CAAC,eAAe,CAAC;IAChD,CAAC;IAED,QAAQ,CAAC,GAAgE;QACvE,IAAI,GAAG,CAAC,IAAI,KAAK,YAAY;YAAE,OAAO,KAAK,CAAC;QAC5C,OAAO,GAAG,CAAC,UAAU,KAAK,mBAAmB,CAAC;IAChD,CAAC;IAED,KAAK,CAAC,OAAO;QACX,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;IACzB,CAAC;IAED,KAAK,CAAC,GAAG,CAAC,GAAmB;QAC3B,MAAM,cAAc,GAAI,GAAG,CAAC,IAAI,CAAC,IAA4C,EAAE,CAAC,aAAa,CAAC,CAAC;QAC/F,IAAI,cAAc,KAAK,IAAI,CAAC,eAAe,EAAE,CAAC;YAC5C,MAAM,IAAI,KAAK,CACb,uCAAuC,MAAM,CAAC,cAAc,CAAC,8CAA8C,IAAI,CAAC,eAAe,EAAE,CAClI,CAAC;QACJ,CAAC;QACD,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,GAAG,IAAI,CAAC,eAAe,QAAQ,CAAC,CAAC;QAC/E,MAAM,KAAK,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;QACpD,OAAO;YACL,QAAQ,EAAE,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE;YAC7B,MAAM,EAAE,EAAE;YACV,eAAe,EAAE;gBACf,aAAa,EAAE,4BAA4B;gBAC3C,KAAK;aACN;SACF,CAAC;IACJ,CAAC;CACF;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,6BAA6B;IAC3C,MAAM,eAAe,GAAG,OAAO,CAAC,GAAG,CAAC,4BAA4B,CAAC,CAAC;IAClE,IAAI,CAAC,eAAe;QAAE,OAAO,IAAI,CAAC;IAClC,IAAI,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC,KAAK,GAAG,EAAE,CAAC;QAC1C,MAAM,IAAI,KAAK,CACb,2DAA2D;YACzD,mEAAmE;YACnE,wEAAwE;YACxE,yEAAyE;YACzE,mDAAmD,CACtD,CAAC;IACJ,CAAC;IACD,MAAM,WAAW,GACf,OAAO,CAAC,GAAG,CAAC,gCAAgC,CAAC;QAC7C,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,QAAQ,EAAE,UAAU,CAAC,CAAC;IACvE,OAAO,IAAI,WAAW,CAAC,EAAE,eAAe,EAAE,WAAW,EAAE,CAAC,CAAC;AAC3D,CAAC"}

package/dist/harnesses/impls/swe-rebench-v2-evaluator/eval-runner.d.ts

@@ -37,6 +37,36 @@ export declare class EvalCouldNotGradeError extends Error {
     readonly logExcerpt: string;
     constructor(reason: string, logExcerpt?: string);
 }
+/**
+ * Thrown by `runEval` when the disk cannot be brought above the eval
+ * disk-floor even after a broad prune. A clean abort — the caller stops
+ * gracefully; no instance is graded, nothing is marked. Distinct from
+ * `EvalCouldNotGradeError`: this is operator-environment, retryable, and must
+ * never be turned into a `scorable: false` admission (#476).
+ */
+export declare class InsufficientDiskError extends Error {
+    readonly freeBytes: number;
+    readonly floorBytes: number;
+    constructor(freeBytes: number, floorBytes: number);
+}
+/**
+ * Default free-disk floor required before an eval round: 20 GB. A single
+ * SWE-rebench eval image was observed to peak transiently at ~12.6 GB, so the
+ * floor clears the worst observed instance with real margin. Override with
+ * `JINN_EVAL_DISK_FLOOR_GB` on constrained hosts.
+ */
+export declare const DEFAULT_EVAL_DISK_FLOOR_BYTES = 20000000000;
+/** Resolve the disk floor: explicit option > `JINN_EVAL_DISK_FLOOR_GB` env > default. */
+export declare function resolveDiskFloorBytes(opt: number | undefined): number;
+/**
+ * Default wall-clock limit for one upstream eval.py invocation: 2 hours. Some
+ * linux/amd64 SWE-rebench images can wedge indefinitely under Apple Silicon
+ * emulation after a native crash, so the subprocess gets a hard guardrail.
+ * Override with `JINN_SWE_REBENCH_EVAL_TIMEOUT_MS`; set `0` to disable.
+ */
+export declare const DEFAULT_EVAL_TIMEOUT_MS: number;
+/** Resolve the eval timeout: explicit option > env > default. */
+export declare function resolveEvalTimeoutMs(opt: number | undefined): number;
 export interface PythonEvalRunnerOptions {
     /** Path to the cloned SWE-rebench-V2 repo (cached locally). */
     upstreamRepoDir: string;
@@ -45,66 +75,55 @@ export interface PythonEvalRunnerOptions {
     /** Workers for parallel eval (defaults to 1; we run one task at a time). */
     maxWorkers?: number;
     /**
-     * Max number of distinct eval images to keep in the local Docker cache.
-     * The runner tracks an in-process LRU keyed by image tag; once usage exceeds
-     * this cap, the least-recently-used images are removed via
-     * {@link PythonEvalRunnerOptions.cleanupImage}.
-     *
-     * The leaderboard pool has hundreds of unique instances at ~3 GB/image, so
-     * an unbounded cache fills operator disks in days (jinn-mono-uy6v.11).
+     * Removes a completed round's entire Docker footprint — the round's image,
+     * stopped containers, and build cache — so eval disk usage never
+     * accumulates across instances (#476). Called once per `runEval`, in a
+     * `finally`, even when the eval threw.
      *
-     * Default: `process.env.JINN_EVAL_IMAGE_CACHE_MAX` parsed as an integer, or
-     * `DEFAULT_EVAL_IMAGE_CACHE_MAX` (20) if unset/invalid.
+     * Defaults to {@link defaultPruneRound}. Implementations MUST NOT throw —
+     * `runEval` guards defensively, but cleanup failures should be swallowed
+     * (logged elsewhere if desired) so a flaky `docker` never escapes `runEval`.
      */
-    imageCacheMax?: number;
+    pruneRound?: (image: string) => Promise<void>;
     /**
-     * Removes an image from the local Docker cache (or no-ops if the operator
-     * has chosen not to GC). Called for each eviction from the LRU.
-     *
-     * Defaults to `docker rmi <image>` via the system `docker` binary. Test
-     * suites inject a stub to capture the eviction order without shelling out.
-     *
-     * Implementations MUST NOT throw — failures should be swallowed (logged
-     * elsewhere if desired) so a missing/failed `docker rmi` never escapes
-     * `runEval`. The runner enforces this defensively too.
+     * Resolves the eval image digest while the image is still local, before
+     * per-round pruning removes it. Defaults to `docker image inspect`.
+     */
+    resolveImageDigest?: (image: string) => Promise<string | null>;
+    /**
+     * Required free disk (bytes) before an eval round starts. Explicit value >
+     * `JINN_EVAL_DISK_FLOOR_GB` env > {@link DEFAULT_EVAL_DISK_FLOOR_BYTES}.
      */
-    cleanupImage?: (image: string) => Promise<void>;
+    diskFloorBytes?: number;
+    /** Probe of free disk (bytes). Defaults to a `statfs` on the temp dir. */
+    freeDiskBytes?: () => Promise<number>;
+    /**
+     * Broad reclaim invoked when free disk is below the floor. Defaults to
+     * `docker system prune -f`. MUST NOT throw.
+     */
+    systemPrune?: () => Promise<void>;
+    /**
+     * Wall-clock timeout (ms) for one upstream eval.py invocation. Explicit value
+     * > `JINN_SWE_REBENCH_EVAL_TIMEOUT_MS` env > {@link DEFAULT_EVAL_TIMEOUT_MS}.
+     * Set to 0 to disable.
+     */
+    evalTimeoutMs?: number;
 }
-/**
- * Default cap on the per-instance Docker image cache when no explicit
- * `imageCacheMax` and no `JINN_EVAL_IMAGE_CACHE_MAX` env var are configured.
- *
- * 20 images × ~3 GB/image ≈ 60 GB working set — small enough that even a
- * 256 GB disk has headroom, large enough that the steady-state loop on a
- * frequently-repeating subset of the pool rarely re-pulls.
- */
-export declare const DEFAULT_EVAL_IMAGE_CACHE_MAX = 20;
-export declare function resolveImageCacheMax(opt: number | undefined): number;
 export declare function matchInfraSignature(log: string): string | null;
 export declare class PythonEvalRunner implements EvalRunner {
     private readonly opts;
-    /**
-     * LRU of image tags whose Docker layers may be cached locally. Stored as a
-     * `Set<string>` because `Set` preserves insertion order; we delete-then-add
-     * to refresh recency and `next()` on the keys iterator to find the
-     * least-recently-used entry.
-     */
-    private readonly imageLru;
-    private readonly imageCacheMax;
-    private readonly cleanupImage;
+    private readonly pruneRound;
+    private readonly diskFloorBytes;
+    private readonly freeDiskBytes;
+    private readonly systemPrune;
+    private readonly resolveImageDigest;
+    private readonly evalTimeoutMs;
     constructor(opts: PythonEvalRunnerOptions);
-    runEval(args: Parameters<EvalRunner['runEval']>[0]): ReturnType<EvalRunner['runEval']>;
     /**
-     * Move `image` to the most-recently-used slot of the in-process LRU; if the
-     * set now exceeds {@link imageCacheMax}, evict the oldest entries via
-     * {@link cleanupImage}. Eviction failures are swallowed so a flaky
-     * `docker rmi` cannot escape `runEval`.
-     *
-     * The cap is enforced after the just-used image is inserted: the
-     * just-evaluated image is the *most* recent, so repeat-evals of recently
-     * used instances never re-pull. Only when more than N distinct images have
-     * been used does the oldest get rmi'd.
+     * Ensure enough free disk for an eval round. Below the floor → broad prune →
+     * re-probe; still below → `InsufficientDiskError` (clean abort). (#476)
      */
-    private recordImageUsage;
+    private ensureDiskHeadroom;
+    runEval(args: Parameters<EvalRunner['runEval']>[0]): ReturnType<EvalRunner['runEval']>;
     private runEvalImpl;
 }