feat: remove LLM extract fallback, require YAML frontmatter

Agent output must contain valid YAML frontmatter matching the role schema.
If frontmatter parsing fails, the step fails immediately with a clear error
instead of falling back to an LLM extraction that can fabricate values.

The extract module remains as a public API export but is no longer used
in the agent run loop.

Breaking change: agents that relied on LLM extraction to produce valid
output will now fail. They must output proper frontmatter.

.workflows/solve-issue.yaml

@@ -0,0 +1,167 @@
+name: "solve-issue"
+description: "TDD-driven issue resolution for small, focused changes. Loop protection relies on engine maxRounds."
+roles:
+  planner:
+    description: "Analyzes issue and outputs a TDD test spec"
+    goal: "You are a planning agent. You analyze Gitea issues and produce a TDD test specification that downstream roles will implement and verify."
+    capabilities:
+      - issue-analysis
+      - planning
+    procedure: |
+      On first run (no previous steps):
+      1. Read the issue and all comments from Gitea using `tea issues <number> -r <owner/repo>`
+      2. Read CLAUDE.md (or equivalent project conventions file) to understand coding standards
+      3. Assess whether the issue has enough information to produce a test spec
+      4. If insufficient info: comment on the issue via `echo "..." | tea comment <number> -r <owner/repo>` (skip if you already commented), then output status=insufficient_info and terminate
+      5. If sufficient: produce a detailed TDD test spec in markdown covering all scenarios
+
+      On subsequent runs (bounced back by tester with fix_spec):
+      1. Read the tester's output from the previous step to understand what's wrong with the spec
+      2. Revise the test spec accordingly
+
+      After producing the test spec:
+      1. Store it via `uwf cas put "<markdown content>"` and capture the returned hash
+      2. Put the hash in meta.plan (required when status=ready)
+    output: "Output a brief summary of the test spec. Frontmatter must include: status (ready or insufficient_info) and plan (CAS hash of the test spec, required when status=ready)."
+    frontmatter:
+      type: object
+      properties:
+        status:
+          type: string
+          enum: [ready, insufficient_info]
+        plan:
+          type: string
+      required: [status]
+  developer:
+    description: "TDD implementation per test spec"
+    goal: "You are a developer agent. You implement code changes following TDD — write tests first, then implementation."
+    capabilities:
+      - coding
+    procedure: |
+      1. Read the test spec from CAS: `uwf cas get <plan hash>` (find the hash from the latest planner step's meta.plan)
+      2. If bounced back from reviewer or tester: read the previous role's output to understand what needs fixing
+      3. Write tests first based on the spec
+      4. Implement the code to make tests pass
+      5. Ensure `bun run build` passes with no errors
+      6. Run `bun test` to verify all tests pass
+    output: "List all files changed and provide a summary. Frontmatter must include: status (done or failed)."
+    frontmatter:
+      type: object
+      properties:
+        status:
+          type: string
+          enum: [done, failed]
+      required: [status]
+  reviewer:
+    description: "Code standards compliance check"
+    goal: "You are a code reviewer. You verify code standards compliance — NOT functionality (that's the tester's job)."
+    capabilities:
+      - code-review
+      - static-analysis
+    procedure: |
+      Hard checks (must all pass):
+      1. `bun run build` — no build errors
+      2. `bunx biome check` — no lint violations
+      3. TypeScript strict mode — no type errors
+
+      Soft checks (review against CLAUDE.md conventions):
+      - Functional-first: `function` + `type`, not `class` + `interface`
+      - No optional properties (`?:`) — use `T | null`
+      - Naming conventions (kebab-case files, PascalCase types, camelCase functions)
+      - Module boundary discipline (folder exports via index.ts)
+      - No `console.log` (use structured logger)
+      - No dynamic imports in production code
+
+      Only review standards compliance. Do NOT test functionality.
+      If rejecting, you MUST explain the specific reason in your output.
+    output: "Explain your decision with specific file/line references. Frontmatter must include: approved (true or false)."
+    frontmatter:
+      type: object
+      properties:
+        approved:
+          type: boolean
+      required: [approved]
+  tester:
+    description: "Functional correctness verification"
+    goal: "You are a tester agent. You verify that the implementation correctly satisfies every scenario in the test spec."
+    capabilities:
+      - testing
+    procedure: |
+      1. Run `bun test` for automated test verification
+      2. Read the test spec from CAS: `uwf cas get <plan hash>` (find the hash from the latest planner step's meta.plan)
+      3. Verify each scenario in the spec is covered and passing
+      4. Determine outcome:
+         - passed: all scenarios verified, tests pass
+         - fix_code: tests fail or implementation doesn't match spec → send back to developer
+         - fix_spec: the spec itself is wrong or incomplete → send back to planner
+    output: "Report test results per scenario. Frontmatter must include: status (passed, fix_code, or fix_spec)."
+    frontmatter:
+      type: object
+      properties:
+        status:
+          type: string
+          enum: [passed, fix_code, fix_spec]
+      required: [status]
+  committer:
+    description: "Commits and creates PR"
+    goal: "You are a committer agent. You create a clean commit and push a PR linking the original issue."
+    capabilities: []
+    procedure: |
+      Note: You inherit the developer's worktree and branch. Do NOT create a new branch.
+      1. Stage all changes: `git add -A`
+      2. Commit with a descriptive message referencing the issue: `git commit -m "type: description\n\nFixes #N"`
+      3. Push the branch: `git push -u origin <branch-name>`
+         - If push hook fails: capture the error log in your output, mark hook_failed
+      4. On push success: create a PR via `tea pr create --title "..." --description "..."`
+         - PR description must follow the project template: What / Why / Changes / Ref sections, with `Fixes #N` in Ref
+    output: "Include PR URL on success or error log on failure. Frontmatter must include: success (true or false)."
+    frontmatter:
+      type: object
+      properties:
+        success:
+          type: boolean
+      required: [success]
+conditions:
+  insufficientInfo:
+    description: "Planner determined there's not enough info to proceed"
+    expression: "$last('planner').status = 'insufficient_info'"
+  devFailed:
+    description: "Developer failed to implement"
+    expression: "$last('developer').status = 'failed'"
+  rejected:
+    description: "Reviewer rejected the implementation"
+    expression: "$last('reviewer').approved = false"
+  fixCode:
+    description: "Tester found code issues"
+    expression: "$last('tester').status = 'fix_code'"
+  fixSpec:
+    description: "Tester found spec issues"
+    expression: "$last('tester').status = 'fix_spec'"
+  hookFailed:
+    description: "Push hook failed"
+    expression: "$last('committer').success = false"
+graph:
+  $START:
+    - role: "planner"
+  planner:
+    - role: "$END"
+      condition: "insufficientInfo"
+    - role: "developer"
+  developer:
+    - role: "$END"
+      condition: "devFailed"
+    - role: "reviewer"
+  reviewer:
+    - role: "developer"
+      condition: "rejected"
+    - role: "tester"
+  tester:
+    - role: "developer"
+      condition: "fixCode"
+    - role: "planner"
+      condition: "fixSpec"
+    - role: "committer"
+  committer:
+    - role: "developer"
+      condition: "hookFailed"
+    - role: "$END"

examples/analyze-topic.yaml

@@ -19,7 +19,7 @@ roles:
     output: |
       Provide your analysis as markdown under the frontmatter.
       The frontmatter must include your structured findings.
-    meta:
+    frontmatter:
       type: object
       properties:
         thesis:

examples/solve-issue.yaml

@@ -9,7 +9,7 @@ roles:
       - planning
     procedure: "Analyze the issue and create a detailed, actionable implementation plan."
     output: "Output the plan summary and list of concrete steps."
-    meta:
+    frontmatter:
       type: object
       properties:
         plan:
@@ -28,7 +28,7 @@ roles:
       - testing
     procedure: "Implement the plan. Write code, tests, and ensure existing tests pass."
     output: "List all files changed and provide a summary of the implementation."
-    meta:
+    frontmatter:
       type: object
       properties:
         filesChanged:
@@ -46,7 +46,7 @@ roles:
       - static-analysis
     procedure: "Review the implementation against the plan. Check for bugs, edge cases, and style."
     output: "Approve or reject with detailed comments explaining your decision."
-    meta:
+    frontmatter:
       type: object
       properties:
         approved:
@@ -57,7 +57,7 @@ roles:
 conditions:
   notApproved:
     description: "Reviewer rejected the implementation"
-    expression: "steps[-1].output.approved = false"
+    expression: "$last('reviewer').approved = false"
 graph:
   $START:
     - role: "planner"

packages/cli-workflow/src/commands/workflow.ts

@@ -2,7 +2,12 @@ import { readFile } from "node:fs/promises";
 
 import type { JSONSchema } from "@uncaged/json-cas";
 import { putSchema, validate } from "@uncaged/json-cas";
-import type { CasRef, RoleDefinition, WorkflowPayload } from "@uncaged/workflow-protocol";
+import type {
+  CasRef,
+  RoleDefinition,
+  Transition,
+  WorkflowPayload,
+} from "@uncaged/workflow-protocol";
 import { parse } from "yaml";
 
 import {
@@ -46,11 +51,28 @@ function isJsonSchema(value: unknown): value is JSONSchema {
   return typeof value === "object" && value !== null && !Array.isArray(value);
 }
 
-async function resolveMetaRef(uwf: UwfStore, roleName: string, meta: unknown): Promise<CasRef> {
-  if (!isJsonSchema(meta)) {
-    fail(`role "${roleName}": meta must be a JSON Schema object`);
+/** Normalize graph transitions: ensure condition is null (not undefined) for fallback entries. */
+function normalizeGraph(graph: Record<string, Transition[]>): Record<string, Transition[]> {
+  const result: Record<string, Transition[]> = {};
+  for (const [node, transitions] of Object.entries(graph)) {
+    result[node] = transitions.map((t) => ({
+      role: t.role,
+      condition: t.condition ?? null,
+    }));
   }
-  const schema: JSONSchema = meta.title === undefined ? { ...meta, title: roleName } : meta;
+  return result;
+}
+
+async function resolveFrontmatterRef(
+  uwf: UwfStore,
+  roleName: string,
+  frontmatter: unknown,
+): Promise<CasRef> {
+  if (!isJsonSchema(frontmatter)) {
+    fail(`role "${roleName}": frontmatter must be a JSON Schema object`);
+  }
+  const schema: JSONSchema =
+    frontmatter.title === undefined ? { ...frontmatter, title: roleName } : frontmatter;
   return putSchema(uwf.store, schema);
 }
 
@@ -60,14 +82,18 @@ export async function materializeWorkflowPayload(
 ): Promise<WorkflowPayload> {
   const roles: Record<string, RoleDefinition> = {};
   for (const [roleName, role] of Object.entries(raw.roles)) {
-    const meta = await resolveMetaRef(uwf, `${raw.name}.${roleName}`, role.meta);
+    const frontmatter = await resolveFrontmatterRef(
+      uwf,
+      `${raw.name}.${roleName}`,
+      role.frontmatter,
+    );
     roles[roleName] = {
       description: role.description,
       goal: role.goal,
       capabilities: role.capabilities,
       procedure: role.procedure,
       output: role.output,
-      meta,
+      frontmatter,
     };
   }
   return {
@@ -75,7 +101,7 @@ export async function materializeWorkflowPayload(
     description: raw.description,
     roles,
     conditions: raw.conditions,
-    graph: raw.graph,
+    graph: normalizeGraph(raw.graph),
   };
 }
 

packages/cli-workflow/src/validate.ts

@@ -15,8 +15,8 @@ function isRoleDefinition(value: unknown): boolean {
   if (!isRecord(value)) {
     return false;
   }
-  const meta = value.meta;
-  const metaOk = isRecord(meta) && typeof meta.type === "string";
+  const frontmatter = value.frontmatter;
+  const frontmatterOk = isRecord(frontmatter) && typeof frontmatter.type === "string";
   const capabilities = value.capabilities;
   const capabilitiesOk =
     Array.isArray(capabilities) && capabilities.every((c) => typeof c === "string");
@@ -26,7 +26,7 @@ function isRoleDefinition(value: unknown): boolean {
     capabilitiesOk &&
     typeof value.procedure === "string" &&
     typeof value.output === "string" &&
-    metaOk
+    frontmatterOk
   );
 }
 
@@ -42,7 +42,10 @@ function isTransition(value: unknown): boolean {
     return false;
   }
   const condition = value.condition;
-  return typeof value.role === "string" && (condition === null || typeof condition === "string");
+  return (
+    typeof value.role === "string" &&
+    (condition === null || condition === undefined || typeof condition === "string")
+  );
 }
 
 function isStringRecord(value: unknown, itemCheck: (item: unknown) => boolean): boolean {

packages/workflow-agent-kit/src/run.ts

@@ -3,10 +3,9 @@ import type { CasRef, StepNodePayload, ThreadId } from "@uncaged/workflow-protoc
 import { config as loadDotenv } from "dotenv";
 import { buildOutputFormatInstruction } from "./build-output-format-instruction.js";
 import { buildContextWithMeta } from "./context.js";
-import { extract } from "./extract.js";
 import { tryFrontmatterFastPath } from "./frontmatter.js";
 import type { AgentStore } from "./storage.js";
-import { getEnvPath, loadWorkflowConfig, resolveStorageRoot } from "./storage.js";
+import { getEnvPath, resolveStorageRoot } from "./storage.js";
 import type { AgentContext, AgentOptions, AgentRunResult } from "./types.js";
 
 function fail(message: string): never {
@@ -73,24 +72,19 @@ async function runAgent(options: AgentOptions, ctx: AgentContext): Promise<Agent
 async function extractOutput(
   rawOutput: string,
   outputSchema: CasRef,
-  storageRoot: string,
   ctx: Awaited<ReturnType<typeof buildContextWithMeta>>,
 ): Promise<CasRef> {
-  const fastPath = await runWithMessage("frontmatter fast path", () =>
-    tryFrontmatterFastPath(rawOutput, outputSchema, ctx.meta.store),
-  ).catch(() => null);
+  const fastPath = await tryFrontmatterFastPath(rawOutput, outputSchema, ctx.meta.store);
 
   if (fastPath !== null) {
     return fastPath.outputHash;
   }
 
-  const config = await runWithMessage("failed to load config", () =>
-    loadWorkflowConfig(storageRoot),
+  fail(
+    "Agent output does not contain valid YAML frontmatter matching the role schema.\n" +
+      "The agent must output a YAML frontmatter block (--- delimited) as the first thing in its response.\n" +
+      `Raw output (first 500 chars): ${rawOutput.slice(0, 500)}`,
   );
-  const extracted = await runWithMessage("extract failed", () =>
-    extract(rawOutput, outputSchema, config),
-  );
-  return extracted.hash;
 }
 
 async function persistStep(options: {
@@ -130,13 +124,13 @@ export function createAgent(options: AgentOptions): () => Promise<void> {
       fail(`unknown role: ${role}`);
     }
 
-    const metaSchema = getSchema(ctx.meta.store, roleDef.meta);
-    if (metaSchema !== null) {
-      ctx.outputFormatInstruction = buildOutputFormatInstruction(metaSchema);
+    const frontmatterSchema = getSchema(ctx.meta.store, roleDef.frontmatter);
+    if (frontmatterSchema !== null) {
+      ctx.outputFormatInstruction = buildOutputFormatInstruction(frontmatterSchema);
     }
 
     const agentResult = await runAgent(options, ctx);
-    const outputHash = await extractOutput(agentResult.output, roleDef.meta, storageRoot, ctx);
+    const outputHash = await extractOutput(agentResult.output, roleDef.frontmatter, ctx);
     const stepHash = await persistStep({
       ctx,
       outputHash,

packages/workflow-moderator/__tests__/evaluate.test.ts

@@ -35,11 +35,11 @@ const solveIssueWorkflow: WorkflowPayload = {
   conditions: {
     needsClarification: {
       description: "Planner requests clarification from user",
-      expression: "$exists(steps[-1].output.needsClarification)",
+      expression: "$exists($last('planner').needsClarification)",
     },
-    notApproved: {
+    rejected: {
       description: "Reviewer rejected the implementation",
-      expression: "steps[-1].output.approved = false",
+      expression: "$last('reviewer').approved = false",
     },
   },
   graph: {
@@ -50,7 +50,7 @@ const solveIssueWorkflow: WorkflowPayload = {
     ],
     developer: [{ role: "reviewer", condition: null }],
     reviewer: [
-      { role: "developer", condition: "notApproved" },
+      { role: "developer", condition: "rejected" },
       { role: "$END", condition: null },
     ],
   },
@@ -72,7 +72,7 @@ describe("evaluate", () => {
     expect(result).toEqual({ ok: true, value: "planner" });
   });
 
-  test("condition match (notApproved → developer)", async () => {
+  test("condition match (rejected → developer)", async () => {
     const context = makeContext([
       {
         role: "reviewer",
@@ -126,4 +126,116 @@ describe("evaluate", () => {
     const result = await evaluate(solveIssueWorkflow, context);
     expect(result).toEqual({ ok: true, value: "developer" });
   });
+
+  test("$last returns most recent matching role's frontmatter", async () => {
+    const workflow: WorkflowPayload = {
+      ...solveIssueWorkflow,
+      conditions: {
+        devFailed: {
+          description: "Developer failed",
+          expression: "$last('developer').status = 'failed'",
+        },
+      },
+      graph: {
+        $START: [{ role: "developer", condition: null }],
+        developer: [
+          { role: "$END", condition: "devFailed" },
+          { role: "reviewer", condition: null },
+        ],
+      },
+    };
+    const context = makeContext([
+      {
+        role: "developer",
+        output: { status: "done" },
+        detail: "1VPBG9SM5E7WK",
+        agent: "uwf-hermes",
+      },
+      {
+        role: "reviewer",
+        output: { approved: false },
+        detail: "2MXBG6PN4A8JR",
+        agent: "uwf-hermes",
+      },
+      {
+        role: "developer",
+        output: { status: "failed" },
+        detail: "3QNTH7WK8D2PA",
+        agent: "uwf-hermes",
+      },
+    ]);
+    const result = await evaluate(workflow, context);
+    expect(result).toEqual({ ok: true, value: "$END" });
+  });
+
+  test("$first returns earliest matching role's frontmatter", async () => {
+    const workflow: WorkflowPayload = {
+      ...solveIssueWorkflow,
+      conditions: {
+        firstPlanReady: {
+          description: "First planner run was ready",
+          expression: "$first('planner').status = 'ready'",
+        },
+      },
+      graph: {
+        $START: [{ role: "planner", condition: null }],
+        planner: [
+          { role: "$END", condition: "firstPlanReady" },
+          { role: "developer", condition: null },
+        ],
+      },
+    };
+    const context = makeContext([
+      {
+        role: "planner",
+        output: { status: "ready", plan: "ABC123" },
+        detail: "7BQST3VW9F2MA",
+        agent: "uwf-hermes",
+      },
+      {
+        role: "developer",
+        output: { status: "done" },
+        detail: "1VPBG9SM5E7WK",
+        agent: "uwf-hermes",
+      },
+      {
+        role: "planner",
+        output: { status: "revised", plan: "DEF456" },
+        detail: "4RNMK6PX8B3WQ",
+        agent: "uwf-hermes",
+      },
+    ]);
+    const result = await evaluate(workflow, context);
+    expect(result).toEqual({ ok: true, value: "$END" });
+  });
+
+  test("$last returns undefined for unmatched role", async () => {
+    const workflow: WorkflowPayload = {
+      ...solveIssueWorkflow,
+      conditions: {
+        hasReviewer: {
+          description: "Reviewer has run",
+          expression: "$exists($last('reviewer'))",
+        },
+      },
+      graph: {
+        $START: [{ role: "planner", condition: null }],
+        planner: [
+          { role: "$END", condition: "hasReviewer" },
+          { role: "developer", condition: null },
+        ],
+      },
+    };
+    const context = makeContext([
+      {
+        role: "planner",
+        output: { status: "ready" },
+        detail: "7BQST3VW9F2MA",
+        agent: "uwf-hermes",
+      },
+    ]);
+    const result = await evaluate(workflow, context);
+    // no reviewer step → $exists returns false → fallback to developer
+    expect(result).toEqual({ ok: true, value: "developer" });
+  });
 });

packages/workflow-moderator/src/evaluate.ts

@@ -21,12 +21,44 @@ function isTruthy(value: unknown): boolean {
   return true;
 }
 
+function findByRole(
+  steps: ModeratorContext["steps"],
+  role: string,
+  direction: "first" | "last",
+): unknown {
+  if (direction === "last") {
+    for (let i = steps.length - 1; i >= 0; i--) {
+      if (steps[i].role === role) {
+        return steps[i].output;
+      }
+    }
+  } else {
+    for (const step of steps) {
+      if (step.role === role) {
+        return step.output;
+      }
+    }
+  }
+  return undefined;
+}
+
 async function evaluateJsonata(
   expression: string,
   context: ModeratorContext,
 ): Promise<Result<unknown, Error>> {
   try {
-    const result = await jsonata(expression).evaluate(context);
+    const expr = jsonata(expression);
+    expr.registerFunction(
+      "first",
+      (role: string) => findByRole(context.steps, role, "first"),
+      "<s:x>",
+    );
+    expr.registerFunction(
+      "last",
+      (role: string) => findByRole(context.steps, role, "last"),
+      "<s:x>",
+    );
+    const result = await expr.evaluate(context);
     return { ok: true, value: result };
   } catch (error) {
     return {

packages/workflow-protocol/src/schemas.ts

@@ -2,14 +2,14 @@ import type { JSONSchema } from "@uncaged/json-cas";
 
 const ROLE_DEFINITION: JSONSchema = {
   type: "object",
-  required: ["description", "goal", "capabilities", "procedure", "output", "meta"],
+  required: ["description", "goal", "capabilities", "procedure", "output", "frontmatter"],
   properties: {
     description: { type: "string" },
     goal: { type: "string" },
     capabilities: { type: "array", items: { type: "string" } },
     procedure: { type: "string" },
     output: { type: "string" },
-    meta: { type: "string", format: "cas_ref" },
+    frontmatter: { type: "string", format: "cas_ref" },
   },
   additionalProperties: false,
 };

packages/workflow-protocol/src/types.ts

@@ -22,7 +22,7 @@ export type RoleDefinition = {
   capabilities: string[];
   procedure: string;
   output: string;
-  meta: CasRef;
+  frontmatter: CasRef;
 };
 
 export type Transition = {