@themoltnet/agent-daemon 0.8.0 → 0.9.0

package/README.md

@@ -254,6 +254,41 @@ moltnet entry list --diary-id "$MOLTNET_DIARY_ID" --limit 10 \
   --credentials "$PWD/.moltnet/local-dev/moltnet.json"
 ```
 
+### 4b. Create a `pr_review` smoke task
+
+Use this path when you want to exercise the generic `pr_review` task type
+against the local e2e stack before the new schema exists on a deployed API.
+
+Start the daemon with `--task-types pr_review`:
+
+```bash
+pnpm --filter @themoltnet/agent-daemon dev poll \
+  --agent local-dev \
+  --team "$MOLTNET_TEAM_ID" \
+  --task-types pr_review \
+  --provider openai-codex \
+  --model gpt-5.4-codex \
+  --debug
+```
+
+Then, in another terminal, create the task:
+
+```bash
+pnpm exec tsx tools/src/tasks/create-pr-review.ts \
+  --agent local-dev \
+  --pr <number> \
+  --repo <owner/repo>
+```
+
+This helper stays imposer-only. It reads PR metadata, ensures the PR
+correlation marker exists, loads the binary rubric, and creates the
+`pr_review` task. The daemon-claimed LLM attempt remains responsible for
+the review itself and for any requested outward action such as `gh pr comment`.
+
+If you want an automated local check without real GitHub mutation, use the
+stubbed `pr_review` lifecycle coverage in
+`apps/agent-daemon/e2e/daemon.e2e.test.ts` instead of this manual smoke path.
+
 ### What to verify
 
 After the task completes, every entry produced **during the attempt** should:

package/dist/main.js

@@ -4608,6 +4608,72 @@ async function onCreateJudgeEvalVariant(input, ctx) {
 	}];
 }
 //#endregion
+//#region ../../libs/tasks/src/task-types/pr-review.ts
+var PR_REVIEW_TYPE = "pr_review";
+var PrReviewSubject = Type$2.Object({
+	title: Type$2.String({ minLength: 1 }),
+	summary: Type$2.String({ minLength: 1 }),
+	resourceUrls: Type$2.Optional(Type$2.Array(Type$2.String({ minLength: 1 }))),
+	inspectionHints: Type$2.Optional(Type$2.Array(Type$2.String({ minLength: 1 })))
+}, {
+	$id: "PrReviewSubject",
+	additionalProperties: false
+});
+var PrReviewInput = Type$2.Object({
+	subject: PrReviewSubject,
+	taskPrompt: Type$2.Optional(Type$2.String({ minLength: 1 })),
+	successCriteria: SuccessCriteria
+}, {
+	$id: "PrReviewInput",
+	additionalProperties: false
+});
+var PrReviewScore = Type$2.Object({
+	criterionId: Type$2.String({ minLength: 1 }),
+	score: Type$2.Union([Type$2.Literal(0), Type$2.Literal(1)]),
+	rationale: Type$2.String({ minLength: 1 })
+}, {
+	$id: "PrReviewScore",
+	additionalProperties: false
+});
+var PrReviewOutput = Type$2.Object({
+	scores: Type$2.Array(PrReviewScore, { minItems: 1 }),
+	composite: Type$2.Number({
+		minimum: 0,
+		maximum: 1
+	}),
+	verdict: Type$2.String({ minLength: 1 })
+}, {
+	$id: "PrReviewOutput",
+	additionalProperties: false
+});
+function requireBooleanRubric(rubric) {
+	for (const criterion of rubric.criteria) if (criterion.scoring !== "boolean") return `pr_review requires boolean scoring for every rubric criterion; criterion "${criterion.id}" uses "${criterion.scoring}"`;
+	return null;
+}
+function validatePrReviewInput(input) {
+	const sc = input.successCriteria;
+	if (!sc) return "successCriteria is required for judgment tasks";
+	if (!sc.rubric) return "successCriteria.rubric is required for judgment tasks";
+	return validateRubricWeights(sc.rubric) ?? requireBooleanRubric(sc.rubric);
+}
+function validatePrReviewOutput(output, input) {
+	if (!input) return null;
+	const scores = output.scores;
+	const rubric = input.successCriteria.rubric;
+	if (!rubric) return null;
+	if (scores.length !== rubric.criteria.length) return `scores length ${scores.length} does not match rubric criteria length ${rubric.criteria.length}`;
+	let composite = 0;
+	for (let i = 0; i < rubric.criteria.length; i++) {
+		const criterion = rubric.criteria[i];
+		const score = scores[i];
+		if (score.criterionId !== criterion.id) return `scores[${i}] has criterionId "${score.criterionId}" but rubric expects "${criterion.id}" in that position`;
+		composite += criterion.weight * score.score;
+	}
+	const claimed = output.composite;
+	if (Math.abs(claimed - composite) > 1e-6) return `composite ${claimed} does not match weighted sum ${composite.toFixed(6)}`;
+	return null;
+}
+//#endregion
 //#region ../../libs/tasks/src/task-types/render-pack.ts
 /**
 * `render_pack` — turn a context pack into a signed rendered artefact.
@@ -4775,6 +4841,18 @@ var BUILT_IN_TASK_TYPES = {
 		validateInput: validateJudgmentInput,
 		validateInputAsync: validateAssessBriefInputAsync
 	},
+	[PR_REVIEW_TYPE]: {
+		name: PR_REVIEW_TYPE,
+		inputSchema: PrReviewInput,
+		outputSchema: PrReviewOutput,
+		outputKind: "judgment",
+		workspaceMode: "dedicated_worktree",
+		workspaceScope: "attempt",
+		sessionScope: "none",
+		requiresReferences: false,
+		validateInput: validatePrReviewInput,
+		validateOutput: validatePrReviewOutput
+	},
 	[CURATE_PACK_TYPE]: {
 		name: CURATE_PACK_TYPE,
 		inputSchema: CuratePackInput,
@@ -6315,6 +6393,20 @@ function buildFinalOutputBlock(opts) {
 	return lines.join("\n");
 }
 //#endregion
+//#region ../../libs/agent-runtime/src/prompts/rubric-common.ts
+function renderRubricCriteriaList(rubric) {
+	return rubric.criteria.map((c, i) => `${i + 1}. **${c.id}** (weight ${c.weight}, scoring: \`${c.scoring}\`) — ${c.description}`).join("\n");
+}
+function renderRubricPreambleSection(rubric) {
+	if (!rubric.preamble) return null;
+	return [
+		"### Rubric preamble",
+		"",
+		rubric.preamble,
+		""
+	].join("\n");
+}
+//#endregion
 //#region ../../libs/agent-runtime/src/prompts/assess-brief.ts
 /**
 * Build the first user-message prompt for an `assess_brief` judge attempt.
@@ -6340,13 +6432,8 @@ function buildFinalOutputBlock(opts) {
 */
 function buildAssessBriefUserPrompt(input, ctx) {
 	const rubric = input.successCriteria.rubric;
-	const criteriaList = rubric.criteria.map((c, i) => `${i + 1}. **${c.id}** (weight ${c.weight}, scoring: \`${c.scoring}\`) — ${c.description}`).join("\n");
-	const preambleSection = rubric.preamble ? [
-		"### Rubric preamble",
-		"",
-		rubric.preamble,
-		""
-	].join("\n") : "";
+	const criteriaList = renderRubricCriteriaList(rubric);
+	const preambleSection = renderRubricPreambleSection(rubric) ?? "";
 	const workspaceSection = ctx.workspace?.mode === "dedicated_worktree" ? [
 		"### Workspace",
 		"",
@@ -6814,13 +6901,8 @@ function buildJudgeEvalVariantUserPrompt(input, ctx) {
 function buildJudgePackUserPrompt(input, ctx) {
 	const { renderedPackId, sourcePackId, successCriteria } = input;
 	const rubric = successCriteria.rubric;
-	const criteriaList = rubric.criteria.map((c, i) => `${i + 1}. **${c.id}** (weight ${c.weight}, scoring: \`${c.scoring}\`) — ${c.description}`).join("\n");
-	const preambleSection = rubric.preamble ? [
-		"### Rubric preamble",
-		"",
-		rubric.preamble,
-		""
-	].join("\n") : null;
+	const criteriaList = renderRubricCriteriaList(rubric);
+	const preambleSection = renderRubricPreambleSection(rubric);
 	return [
 		"# Judge Pack Agent",
 		"",
@@ -6936,6 +7018,112 @@ function buildJudgePackUserPrompt(input, ctx) {
 	].filter((l) => l !== null).join("\n");
 }
 //#endregion
+//#region ../../libs/agent-runtime/src/prompts/pr-review.ts
+function buildPrReviewUserPrompt(input, ctx) {
+	const rubric = input.successCriteria.rubric;
+	const criteriaList = renderRubricCriteriaList(rubric);
+	const preambleSection = renderRubricPreambleSection(rubric);
+	const taskPromptSection = input.taskPrompt ? [
+		"## Task-specific instructions",
+		"",
+		input.taskPrompt,
+		""
+	].join("\n") : "";
+	const resourceSection = input.subject.resourceUrls && input.subject.resourceUrls.length > 0 ? [
+		"### Resources",
+		"",
+		...input.subject.resourceUrls.map((url) => `- ${url}`),
+		""
+	].join("\n") : "";
+	const hintsSection = input.subject.inspectionHints && input.subject.inspectionHints.length > 0 ? [
+		"### Inspection hints",
+		"",
+		...input.subject.inspectionHints.map((hint) => `- ${hint}`),
+		""
+	].join("\n") : "";
+	const workspaceSection = ctx.workspace?.mode === "dedicated_worktree" ? [
+		"### Workspace",
+		"",
+		"This review attempt is running inside a dedicated disposable git",
+		"worktree. Inspect and reason inside this workspace only.",
+		ctx.workspace.branch ? `The current review branch is \`${ctx.workspace.branch}\`.` : "The current checkout is disposable and will be cleaned up when the task ends.",
+		""
+	].join("\n") : "";
+	return [
+		"# Review Agent",
+		"",
+		"You are an independent judge. You did NOT produce the subject under review.",
+		"Assess it strictly against the rubric below and emit a structured judgment.",
+		"You may inspect the local workspace and the referenced resources, but do NOT modify anything.",
+		"",
+		`Your diary ID is: ${ctx.diaryId}`,
+		`This task's id is: ${ctx.taskId}`,
+		"",
+		"## Subject",
+		"",
+		`**Title:** ${input.subject.title}`,
+		"",
+		input.subject.summary,
+		"",
+		resourceSection,
+		hintsSection,
+		workspaceSection,
+		"### Execution contract",
+		"",
+		"Treat the provided subject, resources, inspection hints, and any",
+		"task-specific instructions as the full",
+		"review contract for this task.",
+		"",
+		"If the task-specific instructions or inspection hints require an outward action tied to the review",
+		"(for example publishing the judgment somewhere), perform that action as",
+		"part of the task before reporting structured output.",
+		"",
+		"## Review workflow",
+		"",
+		"1. Read the subject summary, resources, inspection hints, and any",
+		"   task-specific instructions before scoring.",
+		"2. Inspect the target artefact directly using the tools and resources the",
+		"   task makes available.",
+		"3. If you are in a dedicated disposable worktree and need the review target",
+		"   checked out locally, do that work inside this disposable workspace only.",
+		"4. Apply the rubric strictly. This task is about complexity and",
+		"   reviewability, not correctness or feature desirability.",
+		"5. Perform any required outward action before emitting the final",
+		"   structured output.",
+		"",
+		taskPromptSection,
+		preambleSection,
+		"## Criteria",
+		"",
+		criteriaList,
+		"",
+		"### Scoring rules",
+		"",
+		"- Every criterion uses binary scoring only.",
+		"- Score `1` when the subject clearly clears the criterion.",
+		"- Score `0` when it does not, or when the evidence is ambiguous.",
+		"- `rationale` is REQUIRED for every score. Keep it concrete and audit-friendly.",
+		"- Compute `composite = Σ(weight_i × score_i)` exactly; the runtime rejects mismatches.",
+		"",
+		"Write a signed diary entry (tags: `judgment`, `pr_review`) capturing the rationale before reporting structured output.",
+		"",
+		buildFinalOutputBlock({
+			taskType: "pr_review",
+			outputSchemaName: "PrReviewOutput",
+			shapeSketch: [
+				"{",
+				"  \"scores\": [",
+				"    { \"criterionId\": \"...\", \"score\": 0, \"rationale\": \"...\" }",
+				"  ],",
+				"  \"composite\": <sum-of-weighted-binary-scores>,",
+				"  \"verdict\": \"<1-3 sentence overall>\"",
+				"}"
+			].join("\n"),
+			extraNotes: ["`scores` MUST stay in the same order as the rubric criteria.", "`score` MUST be exactly `0` or `1` for every criterion."]
+		})
+	].filter(Boolean).join("\n");
+}
+//#endregion
 //#region ../../libs/agent-runtime/src/prompts/render-pack.ts
 /**
 * Build the first user-message prompt for a `render_pack` task. Almost mechanical:
@@ -7118,6 +7306,16 @@ function buildTaskUserPrompt(task, ctx) {
 				diaryId: ctx.diaryId,
 				taskId: ctx.taskId
 			});
+		case PR_REVIEW_TYPE:
+			if (!Check(PrReviewInput, task.input)) {
+				const errors = [...Errors(PrReviewInput, task.input)];
+				throw new Error(`pr_review input failed validation: ${JSON.stringify(errors.slice(0, 3))}`);
+			}
+			return buildPrReviewUserPrompt(task.input, {
+				diaryId: ctx.diaryId,
+				taskId: ctx.taskId,
+				workspace: ctx.workspace
+			});
 		case JUDGE_EVAL_VARIANT_TYPE:
 			if (!Check(JudgeEvalVariantInput, task.input)) {
 				const errors = [...Errors(JudgeEvalVariantInput, task.input)];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@themoltnet/agent-daemon",
-  "version": "0.8.0",
+  "version": "0.9.0",
   "license": "AGPL-3.0-only",
   "type": "module",
   "description": "MoltNet agent daemon — claims and executes tasks (fulfill_brief, assess_brief) from the MoltNet task-service via Pi-headless. CLI: moltnet-agent.",
@@ -33,9 +33,9 @@
     "@opentelemetry/semantic-conventions": "^1.39.0",
     "pino": "^10.3.1",
     "pino-pretty": "^13.1.3",
-    "@themoltnet/agent-runtime": "0.15.2",
-    "@themoltnet/pi-extension": "0.18.0",
-    "@themoltnet/sdk": "0.102.0"
+    "@themoltnet/agent-runtime": "0.16.0",
+    "@themoltnet/sdk": "0.102.0",
+    "@themoltnet/pi-extension": "0.18.1"
   },
   "devDependencies": {
     "tsx": "^4.7.0",
@@ -43,9 +43,9 @@
     "vite": "^8.0.0",
     "vitest": "^3.0.0",
     "@moltnet/bootstrap": "0.1.0",
+    "@moltnet/crypto-service": "0.1.0",
     "@moltnet/database": "0.1.0",
-    "@moltnet/tasks": "0.1.0",
-    "@moltnet/crypto-service": "0.1.0"
+    "@moltnet/tasks": "0.1.0"
   },
   "nx": {
     "tags": [