@farmslot/agent-runtime 0.20.1 → 0.21.0

package/CHANGELOG.md

@@ -4,6 +4,11 @@
 
 - Active-development baseline; add user-facing changes here before release or package publication.
 
+## 0.21.0 - 2026-10-14
+
+- **BREAKING:** Make acceptance commands read-only, transfer only schema-2 engine claims, and refuse retained schema-1 verdicts for current completion.
+- Resolve completion's evidence-manifest requirement through the selected engine attempt, matching artifact validation without a root copy or symlink.
+
 ## 0.20.1 - 2026-10-14
 
 - Use the coordinated recipe release for content-based dependency readiness and engine-owned evidence manifests.

package/bin/farmslot-agent.mjs

@@ -16,7 +16,7 @@ function usage(exitCode = 0) {
     '  mark <task-md> <signal-json> <args...>',
     '  mark <task-dir> sub <start <id> --step N --from src | <id> <n> | <id> complete | <id> blocked | <id> status>',
     '  artifact-check <task-dir> [args...]',
-    '  ac <set <AC-N> <verdict> [--proof-mode m] [--evidence path]... | list | render> [--task-dir path]',
+    '  ac <list | render> [--task-dir path]',
     '  task init <task-dir> --flow f --platform p --template id --title t [--run-mode m] [options]',
     '  recipe-quality build [--input input.json] [--output artifacts/recipe-quality.json]',
     '    (flags override top-level input fields; training fields are merged)',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@farmslot/agent-runtime",
-  "version": "0.20.1",
+  "version": "0.21.0",
   "type": "module",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
@@ -38,7 +38,7 @@
   },
   "scripts": {
     "typecheck": "yarn workspace @farmslot/protocol build && tsc --noEmit --project tsconfig.json",
-    "test": "node --test test/claim-review-gate.test.mjs test/deterministic-checks.test.cjs && node test/mark-checklist-step.test.cjs && node test/mark-subtask.test.cjs && node test/acceptance-ledger.test.cjs && node test/checklist-target-sync.test.mjs && node test/acceptance-ledger-sync.test.mjs && node test/package-exports.test.cjs && node test/check-task-artifact-contract.test.cjs && node test/recipe-quality-builder.test.cjs && node test/task-init.test.mjs && node --import tsx ../../scripts/e2e-subtask-mark.mts && node --import tsx ../../scripts/e2e-acceptance-ledger.mts && node ../../scripts/quality/run-tsx-tests.mjs --cwd . --tsconfig tsconfig.json src/execution-template && node ../../scripts/quality/run-tsx-tests.mjs --cwd . --tsconfig tsconfig.json src/task-init && node ../../scripts/quality/run-tsx-tests.mjs --cwd . --tsconfig tsconfig.json src/native",
+    "test": "node --test test/claim-review-gate.test.mjs test/deterministic-checks.test.cjs && node test/mark-checklist-step.test.cjs && node test/mark-subtask.test.cjs && node --test test/acceptance-ledger.test.cjs test/acceptance-ledger-cutover.test.cjs && node test/checklist-target-sync.test.mjs && node test/acceptance-ledger-sync.test.mjs && node test/package-exports.test.cjs && node test/check-task-artifact-contract.test.cjs && node test/recipe-quality-builder.test.cjs && node test/task-init.test.mjs && node --import tsx ../../scripts/e2e-subtask-mark.mts && node ../../scripts/quality/run-tsx-tests.mjs --cwd . --tsconfig tsconfig.json src/execution-template && node ../../scripts/quality/run-tsx-tests.mjs --cwd . --tsconfig tsconfig.json src/task-init && node ../../scripts/quality/run-tsx-tests.mjs --cwd . --tsconfig tsconfig.json src/native",
     "format": "prettier --write --ignore-path ../../.prettierignore .",
     "format:check": "prettier --check --ignore-path ../../.prettierignore .",
     "lint": "eslint \"src/**/*.ts\"",
@@ -46,7 +46,8 @@
     "quality": "prettier --check --ignore-path ../../.prettierignore . && eslint \"src/**/*.ts\" && yarn typecheck && yarn build && yarn test",
     "build": "yarn workspace @farmslot/protocol build && rm -rf dist && tsc --project tsconfig.build.json",
     "prepack": "cd ../.. && node scripts/quality/check-farmslot-package-readiness.mjs --require-changelog --packages @farmslot/agent-runtime && yarn workspace @farmslot/agent-runtime build",
-    "prepublishOnly": "node ../../scripts/quality/check-farmslot-package-readiness.mjs --publish --packages @farmslot/agent-runtime"
+    "prepublishOnly": "node ../../scripts/quality/check-farmslot-package-readiness.mjs --publish --packages @farmslot/agent-runtime",
+    "test:acceptance-live": "node --import tsx ../../scripts/e2e-acceptance-ledger.mts"
   },
   "dependencies": {
     "@farmslot/adapter-sdk": "0.12.0",

package/scripts/acceptance-cli.cjs

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 /**
- * `farmslot-agent ac`: record and read the acceptance-criteria ledger
- * (`artifacts/acceptance-status.json`), the one writer of that file.
+ * `farmslot-agent ac`: read the engine-owned acceptance-criteria projection
+ * (`artifacts/acceptance-status.json`). Retained schema-1 data is display only.
  *
  * The ids come from `inputs/handoff.json`, which task init fills from the
  * criteria it received; this CLI never invents an id or a criterion text.
@@ -10,14 +10,11 @@ const fs = require('node:fs');
 const path = require('node:path');
 
 const {
-  ACCEPTANCE_PROOF_MODES,
-  ACCEPTANCE_VERDICTS,
   AcceptanceRefusal,
   acceptanceStatusList,
   handoffAcceptanceCriteria,
   renderAcceptanceCoverage,
   readAcceptanceLedger,
-  setAcceptanceVerdict,
 } = require('./acceptance-ledger.cjs');
 
 const HANDOFF_INPUT = path.join('inputs', 'handoff.json');
@@ -25,12 +22,6 @@ const HANDOFF_INPUT = path.join('inputs', 'handoff.json');
 const USAGE = [
   'usage: farmslot-agent ac <command> [--task-dir <path>]',
   '',
-  `  ac set <AC-N> <${ACCEPTANCE_VERDICTS.join('|')}> [options]`,
-  `      --proof-mode <${ACCEPTANCE_PROOF_MODES.join('|')}>   how the criterion is proven`,
-  '      --evidence <task-dir relative path>   repeatable; the file must exist',
-  '      --recipe-node <id>                    repeatable; recipe node proving it',
-  '      --actor <identity>                     required for an automatic-ledger override',
-  '      --note "..."                          why, especially for untestable',
   '  ac list      every criterion with its current verdict (JSON, verdict null when unrecorded)',
   '  ac render    the coverage table, ending with the Overall recipe coverage line',
   '',
@@ -52,20 +43,10 @@ function parseArgs(argv) {
   const opts = {
     taskDir: null,
     positional: [],
-    proofMode: undefined,
-    evidence: [],
-    recipeNodes: [],
-    note: undefined,
-    actor: undefined,
   };
   for (let i = 0; i < argv.length; i += 1) {
     const arg = argv[i];
     if (arg === '--task-dir') opts.taskDir = takeValue(argv, i++, arg);
-    else if (arg === '--proof-mode') opts.proofMode = takeValue(argv, i++, arg);
-    else if (arg === '--evidence') opts.evidence.push(takeValue(argv, i++, arg));
-    else if (arg === '--recipe-node') opts.recipeNodes.push(takeValue(argv, i++, arg));
-    else if (arg === '--actor') opts.actor = takeValue(argv, i++, arg);
-    else if (arg === '--note') opts.note = takeValue(argv, i++, arg);
     else if (arg === '-h' || arg === '--help') usage(0);
     else if (arg.startsWith('-')) throw new AcceptanceRefusal(`unknown option ${arg}`, 2);
     else opts.positional.push(arg);
@@ -86,27 +67,14 @@ function resolveTaskDir(explicit) {
 function main(argv) {
   const [command, ...rest] = argv;
   if (!command || command === '-h' || command === '--help') usage(command ? 0 : 2);
+  if (command === 'set') {
+    throw new AcceptanceRefusal(
+      'Verdicts are engine-owned. Run eligible recipe assertions; a verified person uses farmslot run gate --action mark-untestable for an exception.',
+    );
+  }
   const opts = parseArgs(rest);
   const taskDir = resolveTaskDir(opts.taskDir);
 
-  if (command === 'set') {
-    const [id, verdict, ...extra] = opts.positional;
-    if (!id || !verdict || extra.length > 0) {
-      throw new AcceptanceRefusal('usage: ac set <AC-N> <verdict> [options]', 2);
-    }
-    const ledger = setAcceptanceVerdict(taskDir, {
-      id,
-      verdict,
-      proofMode: opts.proofMode,
-      evidence: opts.evidence,
-      recipeNodes: opts.recipeNodes,
-      note: opts.note,
-      actor: opts.actor,
-    });
-    const entry = ledger.criteria.find((criterion) => criterion.id === id);
-    console.log(`${id}: ${entry.verdict}${entry.proofMode ? ` (${entry.proofMode})` : ''}`);
-    return;
-  }
   if (opts.positional.length > 0) {
     throw new AcceptanceRefusal(`ac ${command} takes no positional arguments`, 2);
   }
@@ -118,11 +86,15 @@ function main(argv) {
     const ledger = readAcceptanceLedger(taskDir);
     if (!ledger) {
       throw new AcceptanceRefusal(
-        'no acceptance ledger yet — record verdicts with `farmslot-agent ac set <id> <verdict>`',
+        'no acceptance ledger yet. Run the task recipe to derive engine verdicts.',
       );
     }
     // The registered criteria, not just the recorded rows: a partial ledger must
     // not report 1/1 when the task has three criteria.
+    if (ledger.schemaVersion === 1)
+      process.stdout.write(
+        'Legacy schema-1 ledger: retained display only; no measured recipe proof.\n\n',
+      );
     process.stdout.write(renderAcceptanceCoverage(ledger, handoffAcceptanceCriteria(taskDir)));
     return;
   }

package/scripts/acceptance-ledger.cjs

@@ -6,11 +6,10 @@ const { atomicWrite, readJson } = require('./mark-io.cjs');
 // Acceptance-criteria ledger (ADR-060; contract in docs/reference/agent-runtime.md
 // and docs/reference/task-directory-contract.md).
 //
-// This module is the only writer of artifacts/acceptance-status.json, the way
-// `mark` is the only writer of SIGNAL.json. Two commands call it: `farmslot-agent ac`
-// and a recipe run with a task dir, which records its proof targets
-// (@farmslot/recipe-cli). The ids come from task init, which records the criteria in
-// inputs/handoff.json; this module never invents one.
+// The engine derives schema-2 verdicts. This module transfers the selected engine
+// ledger into the task namespace; `farmslot-agent ac` only reads it. Retained
+// schema-1 ledgers remain display input, never current completion proof. The ids
+// come from task init in inputs/handoff.json; this module never invents one.
 //
 // Portable contract helpers are shared with protocol clients, including unbuilt slot scripts.
 
@@ -114,6 +113,11 @@ function readAcceptanceLedger(taskDir) {
 }
 
 function writeAcceptanceLedger(taskDir, ledger) {
+  if (ledger?.schemaVersion !== 2) {
+    throw new AcceptanceRefusal(
+      'Only schema-2 engine claims may be written; schema-1 ledgers are retained display only.',
+    );
+  }
   const issues = validateAcceptanceStatusLedger(ledger);
   if (issues.length > 0) {
     throw new AcceptanceRefusal(
@@ -138,70 +142,6 @@ function assertEvidencePath(taskDir, rawPath) {
   return normalized;
 }
 
-/**
- * Record one verdict. The criterion must be one task init registered: the ledger
- * carries exactly the handoff ids, never an id the worker made up.
- */
-function setAcceptanceVerdict(taskDir, input) {
-  const criteria = handoffAcceptanceCriteria(taskDir);
-  if (criteria.length === 0) {
-    throw new AcceptanceRefusal(
-      `${HANDOFF_INPUT} lists no acceptance criteria; there is no ledger to write`,
-    );
-  }
-  const criterion = criteria.find((entry) => entry.id === input.id);
-  if (!criterion) {
-    throw new AcceptanceRefusal(
-      `unknown acceptance criterion ${input.id}; ${HANDOFF_INPUT} lists ${criteria
-        .map((entry) => entry.id)
-        .join(', ')}`,
-    );
-  }
-  if (!ACCEPTANCE_VERDICTS.includes(input.verdict)) {
-    throw new AcceptanceRefusal(
-      `unknown verdict ${input.verdict}; expected one of ${ACCEPTANCE_VERDICTS.join(', ')}`,
-    );
-  }
-  if (input.proofMode !== undefined && !ACCEPTANCE_PROOF_MODES.includes(input.proofMode)) {
-    throw new AcceptanceRefusal(
-      `unknown proof mode ${input.proofMode}; expected one of ${ACCEPTANCE_PROOF_MODES.join(', ')}`,
-    );
-  }
-  const evidence = (input.evidence ?? []).map((entry) => assertEvidencePath(taskDir, entry));
-  const stored = readAcceptanceLedger(taskDir);
-  if (stored) {
-    const issues = validateAcceptanceStatusLedger(stored);
-    if (issues.length > 0) {
-      throw new AcceptanceRefusal(
-        `${ACCEPTANCE_STATUS_ARTIFACT} does not match the ledger contract:\n- ${issues.join('\n- ')}`,
-      );
-    }
-  }
-  if (stored?.schemaVersion === 2) {
-    throw new AcceptanceRefusal(
-      'Automatic verdicts are engine-owned. Request human review; a verified person uses farmslot run gate --action mark-untestable.',
-    );
-  }
-  const byId = new Map((stored?.criteria ?? []).map((entry) => [entry.id, entry]));
-  byId.set(criterion.id, {
-    id: criterion.id,
-    text: criterion.text,
-    verdict: input.verdict,
-    ...(input.proofMode ? { proofMode: input.proofMode } : {}),
-    evidence,
-    recipeNodes: [...(input.recipeNodes ?? [])],
-    ...(input.note ? { note: input.note } : {}),
-    updatedAt: input.now ?? new Date().toISOString(),
-  });
-  // Handoff order, so the ledger and the rendered table always read like TASK.md.
-  const ledger = {
-    schemaVersion: 1,
-    criteria: criteria.map((entry) => byId.get(entry.id)).filter(Boolean),
-  };
-  writeAcceptanceLedger(taskDir, ledger);
-  return ledger;
-}
-
 /** Transfer one engine-owned execution. Never aggregate sibling attempts. */
 function setDerivedClaims(taskDir, claims) {
   if (claims.schemaVersion !== 2) throw new AcceptanceRefusal('Derived claims require schema 2.');
@@ -268,24 +208,26 @@ function acceptanceContractIssues(taskDir, options = {}) {
   if (!stored) {
     return [
       `${ACCEPTANCE_STATUS_ARTIFACT} is missing but ${HANDOFF_INPUT} lists ${criteria.length} ` +
-        'acceptance criteria — record a verdict for each with `farmslot-agent ac set <id> <verdict>`',
+        'acceptance criteria. Run the task recipe with an eligible assertion for each criterion.',
+    ];
+  }
+  if (stored.schemaVersion !== 2) {
+    return [
+      `${ACCEPTANCE_STATUS_ARTIFACT}: legacy schema-1 verdicts are display only. Run the task recipe to obtain schema-2 engine claims.`,
     ];
   }
   const issues = validateAcceptanceStatusLedger(stored);
   if (issues.length > 0) return issues;
   const known = new Set(criteria.map((criterion) => criterion.id));
   for (const entry of stored.criteria) {
-    if (
-      !known.has(entry.id) &&
-      (stored.schemaVersion === 1 || ACCEPTANCE_CRITERION_ID_PATTERN.test(entry.id))
-    ) {
+    if (!known.has(entry.id) && ACCEPTANCE_CRITERION_ID_PATTERN.test(entry.id)) {
       issues.push(
         `${ACCEPTANCE_STATUS_ARTIFACT}: ${entry.id} is not an acceptance criterion of this task`,
       );
     }
   }
   const byId = new Map(stored.criteria.map((entry) => [entry.id, entry]));
-  if (stored.schemaVersion === 2 && stored.criteria.some((entry) => entry.overrides !== undefined))
+  if (stored.criteria.some((entry) => entry.overrides !== undefined))
     issues.push(
       'Unverified local overrides require a canonical engine projection; they grant no review authority.',
     );
@@ -293,21 +235,18 @@ function acceptanceContractIssues(taskDir, options = {}) {
     const entry = byId.get(criterion.id);
     if (!entry) {
       issues.push(
-        `${criterion.id} has no verdict — run \`farmslot-agent ac set ${criterion.id} <verdict>\``,
+        `${criterion.id} has no engine verdict. Run the task recipe with an eligible assertion for this criterion.`,
       );
       continue;
     }
     if (
-      !(stored.schemaVersion === 2 && options.deferClaimVerdicts === true) &&
+      options.deferClaimVerdicts !== true &&
       (entry.verdict === 'failed' ||
         entry.verdict === 'awaiting-review' ||
         (!options.allowWeak && (entry.verdict === 'missing' || entry.verdict === 'weak')))
     ) {
       issues.push(
-        stored.schemaVersion === 2
-          ? `${criterion.id} is ${entry.verdict}: run an eligible recipe assertion or request human evidence review.`
-          : `${criterion.id} is ${entry.verdict}: prove it, or record \`untestable\` with a note ` +
-              '(a flow may waive this with worker_terminal.acceptance.allowWeak)',
+        `${criterion.id} is ${entry.verdict}: run an eligible recipe assertion or request human evidence review.`,
       );
     }
   }
@@ -328,7 +267,6 @@ module.exports = {
   requireHandoffAcceptanceCriteria,
   readAcceptanceLedger,
   renderAcceptanceCoverage,
-  setAcceptanceVerdict,
   setDerivedClaims,
   summarizeAcceptanceStatus,
   validateAcceptanceStatusLedger,

package/scripts/check-task-artifact-contract.mjs

@@ -7,6 +7,7 @@ import { fileURLToPath } from 'node:url';
 const require = createRequire(import.meta.url);
 const {
   expandedArtifactsForCommand,
+  resolveWorkerTerminalArtifactPath,
   deterministicChecks,
 } = require('./worker-terminal-contract.cjs');
 const { SUBTASK_INDEX_REL, openSubtaskUnits } = require('./subtask-unit.cjs');
@@ -164,18 +165,8 @@ function resolveTaskArtifactPath(rel) {
 }
 
 function publicationArtifactPath(rel) {
-  if (rel !== 'artifacts/evidence-manifest.json') return rel;
-  const pointerPath = path.join(taskDir, 'artifacts/latest-valid-recipe-run.json');
-  if (!existsSync(pointerPath)) return rel;
-  const pointer = recipeProtocol?.sanitizeLatestValidRecipeRunPointer?.(
-    JSON.parse(readFileSync(pointerPath, 'utf8')),
-  );
-  if (!pointer) throw new Error('latest-valid-recipe-run.json: invalid pointer.');
-  const selected = `artifacts/${pointer.relativeArtifactRoot}/evidence-manifest.json`;
   const version = terminalContract ? (terminalContract.recordedEvidenceContractVersion ?? 0) : 1;
-  if (version !== 0 && version !== 1) throw new Error('Unsupported recorded evidence contract.');
-  if (version === 0 && !existsSync(path.resolve(taskDir, selected))) return rel;
-  return selected;
+  return resolveWorkerTerminalArtifactPath(taskDir, rel, version);
 }
 
 function isRecord(value) {

package/scripts/mark-checklist-step.cjs

@@ -6,6 +6,7 @@ const { spawnSync } = require('node:child_process');
 const {
   expandedArtifactsForCommand,
   resolveWorkerTerminalContract,
+  resolveWorkerTerminalArtifactPath,
   deterministicChecks,
 } = require('./worker-terminal-contract.cjs');
 const {
@@ -225,7 +226,14 @@ function assertRequiredArtifacts(taskDir, terminalCommand, contract) {
   );
   for (const storedPath of expanded.artifacts) {
     if (storedPath === LEARNINGS_ARTIFACT && opts['skip-learnings']) continue;
-    assertArtifactFile(taskDir, storedPath);
+    assertArtifactFile(
+      taskDir,
+      resolveWorkerTerminalArtifactPath(
+        taskDir,
+        storedPath,
+        contract.recordedEvidenceContractVersion ?? 0,
+      ),
+    );
   }
 }
 

package/scripts/worker-terminal-contract.cjs

@@ -174,6 +174,7 @@ const BUILTIN_FLOW_COMMANDS = {
 const DEFAULT_WHEN_PRESENT = [
   {
     path: 'artifacts/recipe.json',
+    // The manifest requirement resolves to the selected engine attempt, not a root copy.
     alsoRequire: ['artifacts/recipe-coverage.md', 'artifacts/evidence-manifest.json'],
     requireRecipeQuality: true,
     requireRecipeCoverage: true,
@@ -432,6 +433,21 @@ function expandedArtifactsForCommand(contract, command, fileExists) {
 
 const DETERMINISTIC_REFERENCE = /\[DETERMINISTIC:\s*([A-Za-z_][A-Za-z0-9_.]*)\s*\]/;
 
+/** Resolve the logical manifest requirement through the existing retained selection. */
+function resolveWorkerTerminalArtifactPath(taskDir, relPath, recordedEvidenceContractVersion = 1) {
+  if (relPath !== 'artifacts/evidence-manifest.json') return relPath;
+  if (!fs.existsSync(path.join(taskDir, 'artifacts/latest-valid-recipe-run.json'))) return relPath;
+  const { readRecipeAttemptSelection } = require('./retained-recipe-attempt.cjs');
+  const pointer = readRecipeAttemptSelection(taskDir);
+  if (!pointer) throw new Error('Invalid retained recipe attempt selection.');
+  const selected = `artifacts/${pointer.relativeArtifactRoot}/evidence-manifest.json`;
+  if (recordedEvidenceContractVersion !== 0 && recordedEvidenceContractVersion !== 1)
+    throw new Error('Unsupported recorded evidence contract.');
+  if (recordedEvidenceContractVersion === 0 && !fs.existsSync(path.resolve(taskDir, selected)))
+    return relPath;
+  return selected;
+}
+
 function deterministicDeclarations(markdown, checklist) {
   return enumerateChecklistCheckboxes(markdown).flatMap((item) => {
     const match = item.rawLabel.match(DETERMINISTIC_REFERENCE);
@@ -770,5 +786,6 @@ module.exports = {
   templateTerminalCommands,
   templateBodyScope,
   expandedArtifactsForCommand,
+  resolveWorkerTerminalArtifactPath,
   deterministicChecks,
 };