humanish 0.32.0 → 0.34.0

package/dist/program.js

@@ -11,6 +11,7 @@ import { runInit } from "./init.js";
 import { inspectLabManifest, listLabManifests, resolveLabManifest } from "./labs.js";
 import { runLabPreflight } from "./lab-preflight.js";
 import { runLab, resolveLabDryRun, selectLabBackend } from "./lab-engine.js";
+import { loadAdapterScorer } from "./adapter-scorer-loader.js";
 import { CUA_ACTOR_LAB_SCHEMA } from "./cua-actor-lab.js";
 import { openTarget, renderObserver, serveObserver } from "./observer.js";
 import { serveObserverStatic } from "./observer-static.js";
@@ -481,6 +482,7 @@ function registerWatchCommand(parent, io) {
         .option("--redact-repos", "Lab only: redact repo labels in durable artifacts.")
         .option("--no-redact-repos", "Lab only: persist repo labels. Use only for public-safe runs.")
         .option("--keep", "Lab only: keep disposable clone sandbox for debugging.")
+        .option("--scorer <path>", "Terminal/computer-use/shared-world labs only: repo-relative adopter scorer module (.mjs). Overrides review.scorer.ref. Executable code — review it as code.")
         .option("--run-id <id>", "Explicit run id for deterministic fixture tests.")
         .option("--cwd <path>", "Target project directory.", ".")
         .option("--env-file <path>", "Load a local env file for this watch without persisting values.")
@@ -574,6 +576,7 @@ function registerWatchCommand(parent, io) {
                     repo: options.repo,
                     ...(options.repos === undefined ? {} : { repos: options.repos }),
                     ...(options.runId === undefined ? {} : { runId: options.runId }),
+                    ...(options.scorer === undefined ? {} : { scorer: options.scorer }),
                     ...(options.sims === undefined ? {} : { sims: options.sims }),
                     ...(options.expose === undefined ? {} : { expose: options.expose }),
                     ...(options.tunnel === undefined ? {} : { tunnel: options.tunnel }),
@@ -1238,6 +1241,7 @@ function registerLabCommands(parent, io) {
         .option("--redact-repos", "Meta only: redact repo labels in durable lab artifacts.")
         .option("--no-redact-repos", "Meta only: persist repo labels in durable lab artifacts. Use only for public-safe runs.")
         .option("--keep", "Smoke labs only: keep disposable clone sandbox for debugging.")
+        .option("--scorer <path>", "Terminal/computer-use/shared-world labs only: repo-relative adopter scorer module (.mjs). Overrides review.scorer.ref. Executable code — review it as code.")
         .option("--json", JSON_OPTION_DESCRIPTION)
         .addHelpText("after", [
         "",
@@ -1245,6 +1249,7 @@ function registerLabCommands(parent, io) {
         "  humanish lab run first-run",
         "  humanish lab run fanout-demo --rerun-failed-from latest --lanes lane-02,lane-04",
         "  humanish lab run oss --dry-run --json --no-open",
+        "  humanish lab run my-terminal-lab --scorer scorers/product.mjs",
         "  humanish lab run .humanish/labs/private-dogfood.yaml --env-file .humanish/local/provider.env",
         "",
         "Human watch path:",
@@ -1473,6 +1478,39 @@ async function runOssSmokeAction(args) {
     writeResult(args.command, args.io, result, formatOssLabHuman);
     args.io.setExitCode(result.ok ? 0 : 2);
 }
+/**
+ * Resolve `review.scorer.ref` (or the `--scorer` override) to a loaded adopter scorer, fail-closed
+ * (typed error) PRE-SPEND. Precedence: CLI `--scorer` overrides the manifest; `source` records which
+ * won. No scorer declared → `{ ok: true }` with no scorer. A declared scorer on an unsupported
+ * backend, or a bad/unreadable/broken ref, → `{ ok: false }` so the caller aborts with exit 2.
+ */
+async function maybeLoadAdapterScorer(args) {
+    const ref = args.flag ?? args.config.review?.scorer?.ref;
+    if (ref === undefined)
+        return { ok: true };
+    const source = args.flag !== undefined ? "cli-flag" : "manifest";
+    const loaded = await loadAdapterScorer({ cwd: args.cwd, ref, backend: args.backend, source });
+    if (!loaded.ok)
+        return { ok: false, error: loaded.error };
+    return { ok: true, scorer: { hooks: loaded.hooks, provenance: loaded.provenance } };
+}
+/** Terminal route hooks bag from a loaded scorer (deriveArtifacts is browser-only, dropped here). */
+function terminalScorerHooks(scorer) {
+    const { hooks } = scorer;
+    return {
+        ...(hooks.score ? { score: hooks.score } : {}),
+        ...(hooks.deriveFeedback ? { deriveFeedback: hooks.deriveFeedback } : {})
+    };
+}
+/** Browser route hooks bag from a loaded scorer (score + deriveFeedback + deriveArtifacts). */
+function browserScorerHooks(scorer) {
+    const { hooks } = scorer;
+    return {
+        ...(hooks.score ? { score: hooks.score } : {}),
+        ...(hooks.deriveFeedback ? { deriveFeedback: hooks.deriveFeedback } : {}),
+        ...(hooks.deriveArtifacts ? { deriveArtifacts: hooks.deriveArtifacts } : {})
+    };
+}
 async function runLabCommand(args) {
     const resolved = await resolveLabManifest(args.options.cwd, args.lab);
     if (!resolved.ok) {
@@ -1508,6 +1546,33 @@ async function runLabCommand(args) {
         args.io.setExitCode(2);
         return;
     }
+    // #316: resolve + load a config-declared/CLI-flagged adopter scorer FAIL-CLOSED, before any spend.
+    // A declared gate that cannot load (bad ref, not found, load failure, no hooks, unsupported backend)
+    // aborts with exit 2 rather than green-passing.
+    const scorerLoad = await maybeLoadAdapterScorer({
+        cwd: args.options.cwd,
+        config,
+        backend,
+        flag: args.options.scorer
+    });
+    if (!scorerLoad.ok) {
+        const result = {
+            schema: "humanish.run-result.v1",
+            ok: false,
+            cwd: resolve(args.options.cwd),
+            warnings: [],
+            error: scorerLoad.error
+        };
+        writeResult(args.command, args.io, result, formatRunHuman);
+        args.io.setExitCode(2);
+        return;
+    }
+    const scorer = scorerLoad.scorer;
+    if (scorer) {
+        // Cross-repo guardrail: `humanish lab run` now import()s host JS named in the manifest. Surface it
+        // visibly so the invoker (who may not be the manifest author) knows executable code just ran.
+        args.io.writeErr(`warning: review scorer ${scorer.provenance.ref} (${scorer.provenance.source}) is executable host code loaded and run in-process — review it as code, not config.\n`);
+    }
     switch (backend) {
         case "synthetic":
             await runSyntheticBackend({ ...args, config });
@@ -1519,19 +1584,19 @@ async function runLabCommand(args) {
             await runSmokeBackend({ ...args, config });
             return;
         case "cua":
-            await runCuaBackend({ ...args, config });
+            await runCuaBackend({ ...args, config, ...(scorer ? { scorer } : {}) });
             return;
         case "scripted":
             await runScriptedBackend({ ...args, config });
             return;
         case "terminal":
-            await runTerminalBackend({ ...args, config });
+            await runTerminalBackend({ ...args, config, ...(scorer ? { scorer } : {}) });
             return;
         case "shared-world":
-            await runSharedWorldBackend({ ...args, config });
+            await runSharedWorldBackend({ ...args, config, ...(scorer ? { scorer } : {}) });
             return;
         case "concurrent-shared-world":
-            await runConcurrentSharedWorldBackend({ ...args, config });
+            await runConcurrentSharedWorldBackend({ ...args, config, ...(scorer ? { scorer } : {}) });
             return;
         default:
             // Compile-time exhaustiveness: a future backend must be handled here, not silently no-op.
@@ -1718,6 +1783,7 @@ async function runCuaBackend(args) {
                 }
                 : {}),
             ...(args.options.runId === undefined ? {} : { runId: args.options.runId }),
+            ...(args.scorer ? { cuaHooks: browserScorerHooks(args.scorer), scorerProvenance: args.scorer.provenance } : {}),
             ...(args.options.rerunFailedFrom === undefined
                 ? {}
                 : {
@@ -1853,7 +1919,8 @@ async function runTerminalBackend(args) {
         cwd: args.options.cwd,
         open: args.mode === "watch" ? false : shouldOpen,
         ...(args.options.dryRun === undefined ? {} : { dryRun: args.options.dryRun }),
-        ...(args.options.runId === undefined ? {} : { runId: args.options.runId })
+        ...(args.options.runId === undefined ? {} : { runId: args.options.runId }),
+        ...(args.scorer ? { terminalHooks: terminalScorerHooks(args.scorer), scorerProvenance: args.scorer.provenance } : {})
     });
     if (outcome.backend !== "terminal") {
         throw new Error(`Expected terminal backend, got ${outcome.backend}.`);
@@ -1885,7 +1952,8 @@ async function runSharedWorldBackend(args) {
         cwd: args.options.cwd,
         open: args.mode === "watch" ? false : shouldOpen,
         ...(args.options.dryRun === undefined ? {} : { dryRun: args.options.dryRun }),
-        ...(args.options.runId === undefined ? {} : { runId: args.options.runId })
+        ...(args.options.runId === undefined ? {} : { runId: args.options.runId }),
+        ...(args.scorer ? { sharedWorldHooks: browserScorerHooks(args.scorer), scorerProvenance: args.scorer.provenance } : {})
     });
     if (outcome.backend !== "shared-world") {
         throw new Error(`Expected shared-world backend, got ${outcome.backend}.`);
@@ -1975,7 +2043,8 @@ async function runConcurrentSharedWorldBackend(args) {
                     }
                 }
                 : {}),
-            ...(args.options.runId === undefined ? {} : { runId: args.options.runId })
+            ...(args.options.runId === undefined ? {} : { runId: args.options.runId }),
+            ...(args.scorer ? { sharedWorldHooks: browserScorerHooks(args.scorer), scorerProvenance: args.scorer.provenance } : {})
         });
     }
     catch (error) {