FE-1525: Establish the runbook-to-workpiece control and expose the construction gap

.github/actions/prune-repository/prune.py

@@ -40,24 +40,11 @@
     "@rust/hash-graph-types": ["@rust/hash-graph-test-data"],
 }
 
-# Extras that must not fire on a transitive or prefix match. Brunch core's
-# architecture and contract tests inspect the app and shipped plugins, but a job
-# whose requested scope is only a sibling or a consumer of core must not pull
-# those fixtures.
-REQUESTED_DEPENDENCIES: dict[str, list[str]] = {
-    "@hashintel/brunch-agent": [
-        "@apps/brunch-agent",
-        "@hashintel/brunch-agent-plugin-gherkin",
-        "@hashintel/brunch-agent-plugin-sdcpn",
-    ],
-}
-
 # Non-workspace paths required by packages in the *requested* scope.
 # `turbo prune` copies workspace directories and root manifests only.
 REQUESTED_PATHS: dict[str, list[str]] = {
-    # The Brunch context root is deliberately not a workspace, but the
-    # architecture tests in packages/core read its docs, scripts, and agent
-    # contract files
+    # Core's shipped-definition and baseline tests read the non-workspace
+    # context root alongside their plugin task dependencies.
     "@hashintel/brunch-agent": [
         ".config/oxlint/brunch",
         "libs/@hashintel/brunch-agent/AGENTS.md",
@@ -66,9 +53,18 @@
         "libs/@hashintel/brunch-agent/evaluations",
         "libs/@hashintel/brunch-agent/scripts",
     ],
-    # The app's condition-5 test executes the evaluation runner as a child
-    # process; the context root is not a workspace and must be copied explicitly.
-    "@apps/brunch-agent": ["libs/@hashintel/brunch-agent/evaluations"],
+    # The app's tests execute evaluation runners and govern the complete Brunch
+    # composition. Its context root is not a workspace, so copy the docs,
+    # scripts, and agent contract files explicitly.
+    "@apps/brunch-agent": [
+        ".config/oxlint/brunch",
+        "libs/@hashintel/brunch-agent/AGENTS.md",
+        "libs/@hashintel/brunch-agent/CONTEXT.md",
+        "libs/@hashintel/brunch-agent/docs",
+        "libs/@hashintel/brunch-agent/evaluations",
+        "libs/@hashintel/brunch-agent/scripts",
+        "libs/@hashintel/petrinaut/docs",
+    ],
 }
 
 TURBO_QUERY = """
@@ -129,20 +125,6 @@ def turbo_dependency_map() -> dict[str, frozenset[str]]:
     return dep_map
 
 
-def extras_for_requested(requested: Iterable[str]) -> frozenset[str]:
-    """Return extras implied by the job's requested scopes only.
-
-    Exact identity: a name that is a prefix of its siblings must not match them.
-    """
-
-    names = set(requested)
-    extras: set[str] = set()
-    for trigger, additions in REQUESTED_DEPENDENCIES.items():
-        if trigger in names:
-            extras.update(additions)
-    return frozenset(extras)
-
-
 def extra_paths_for_requested(requested: Iterable[str]) -> list[str]:
     """Return non-workspace paths implied by the job's requested scopes only."""
 
@@ -297,7 +279,7 @@ def main(argv: list[str] | None = None) -> None:
     }
 
     dependencies = turbo_dependency_map()
-    scopes = fixpoint_expand(initial | extras_for_requested(initial), dependencies)
+    scopes = fixpoint_expand(initial, dependencies)
     turbo_prune(scopes, dry_run=args.dry_run)
     copy_extra_paths(initial, dry_run=args.dry_run)
     stub_missing_members(dry_run=args.dry_run)

.github/actions/prune-repository/prune_test.py

@@ -8,7 +8,6 @@
 from prune import (
     expand_scopes,
     extra_paths_for_requested,
-    extras_for_requested,
     fixpoint_expand,
 )
 
@@ -21,54 +20,47 @@
 
 
 class BrunchRequestedExtras(unittest.TestCase):
-    def test_core_job_adds_the_app_plugins_and_context_paths(self) -> None:
-        expected_workspaces = frozenset({APP, PLUGIN_GHERKIN, PLUGIN_SDCPN})
-        self.assertEqual(extras_for_requested({CORE}), expected_workspaces)
+    def test_app_task_adds_the_core_plugins_and_context_paths(self) -> None:
+        expected_workspaces = frozenset({CORE, PLUGIN_GHERKIN, PLUGIN_SDCPN})
         expanded = fixpoint_expand(
-            {CORE} | extras_for_requested({CORE}),
+            {APP},
             {
-                CORE: frozenset(),
-                APP: frozenset({CORE}),
+                APP: expected_workspaces,
                 PLUGIN_GHERKIN: frozenset({CORE}),
                 PLUGIN_SDCPN: frozenset({CORE}),
             },
         )
         self.assertTrue(expected_workspaces.issubset(expanded))
         self.assertEqual(
-            extra_paths_for_requested({CORE}),
+            extra_paths_for_requested({APP}),
             [
                 ".config/oxlint/brunch",
                 "libs/@hashintel/brunch-agent/AGENTS.md",
                 "libs/@hashintel/brunch-agent/CONTEXT.md",
                 "libs/@hashintel/brunch-agent/docs",
                 "libs/@hashintel/brunch-agent/evaluations",
                 "libs/@hashintel/brunch-agent/scripts",
+                "libs/@hashintel/petrinaut/docs",
             ],
         )
 
-    def test_app_job_adds_the_baseline_evaluation_paths(self) -> None:
+    def test_core_adds_its_non_workspace_test_fixtures(self) -> None:
         self.assertEqual(
-            extra_paths_for_requested({APP}),
-            ["libs/@hashintel/brunch-agent/evaluations"],
+            extra_paths_for_requested({CORE}),
+            [
+                ".config/oxlint/brunch",
+                "libs/@hashintel/brunch-agent/AGENTS.md",
+                "libs/@hashintel/brunch-agent/CONTEXT.md",
+                "libs/@hashintel/brunch-agent/docs",
+                "libs/@hashintel/brunch-agent/evaluations",
+                "libs/@hashintel/brunch-agent/scripts",
+            ],
         )
 
-    def test_sibling_or_website_job_does_not_add_brunch_extras(self) -> None:
-        self.assertEqual(extras_for_requested({TRANSPORT}), frozenset())
-        self.assertEqual(extras_for_requested({WEBSITE}), frozenset())
+    def test_sibling_or_website_job_does_not_add_context_paths(self) -> None:
         self.assertEqual(extra_paths_for_requested({TRANSPORT}), [])
         self.assertEqual(extra_paths_for_requested({WEBSITE}), [])
 
-    def test_core_in_the_dependency_closure_does_not_add_the_app(self) -> None:
-        dependencies = {
-            WEBSITE: frozenset({CORE, TRANSPORT}),
-            TRANSPORT: frozenset(),
-            CORE: frozenset(),
-        }
-        expanded = fixpoint_expand({WEBSITE}, dependencies)
-        self.assertNotIn(APP, expanded)
-        self.assertEqual(extras_for_requested({WEBSITE}), frozenset())
-
-
 class DarwinPrefix(unittest.TestCase):
     def test_child_crate_still_triggers_the_prefix_family(self) -> None:
         extras = expand_scopes({"@rust/darwin-kperf-sys"})

apps/brunch-agent/README.md

@@ -10,13 +10,25 @@ yarn dev:brunch
 
 The first step builds the Petrinaut libraries the panel imports (`dist/` and design-system
 codegen). Then it starts the Brunch server at `http://127.0.0.1:4321` and the real Petrinaut
-website at `http://127.0.0.1:4915`. The website proxies `/api/chat` to Brunch. The panel talks to one plain
-Flue chat agent: streamed text and reasoning, one server `ping` tool, one stub
-skill (`confirm-path`, activated via `activate_skill`), and the existing Petrinaut
-`readPetrinautDoc` client tool. There is no elicitation loop, sweep tool, or
-`brunch_ask` on this path. Capture is a harness-side pipe: an explicit settled
-range of Flue history is applied into a JSON store beside the conversation
-database, not by the interviewer.
+website at `http://127.0.0.1:4915`. The website proxies `/api/chat` to Brunch. The panel talks to one Flue
+chat agent: streamed text and reasoning, one server `ping` tool, one
+modelling runbook skill (`sdcpn-modelling`, activated via `activate_skill`,
+with supporting resources via `read_skill_resource`), and the existing
+Petrinaut `readPetrinautDoc` client tool. There is no elicitation loop,
+sweep tool, or `brunch_ask` on this path. Capture is a harness-side pipe:
+an explicit settled range of Flue history is applied into a JSON store
+beside the conversation database, not by the interviewer.
+
+A headless Mission 3 drive (simulated expert, same `ChatAgent` door):
+
+```sh
+yarn workspace @apps/brunch-agent runbook:headless
+```
+
+`ANTHROPIC_API_KEY` is required. `BRUNCH_CHAT_MODEL` selects the interviewer
+(default `claude-sonnet-4-5` for this script only). Artifacts write under
+`libs/@hashintel/brunch-agent/docs/evidence/evaluations/vestera-runbook-headless/`
+unless `BRUNCH_RUNBOOK_OUTPUT_DIR` is set.
 
 Conversations persist in `apps/brunch-agent/.data-wipe-me/conversations.db`. `BRUNCH_DEV_DB_PATH`
 overrides that local path. Capture envelopes for one Flue conversation sit beside that sqlite

apps/brunch-agent/docs/task-dependencies.json

@@ -40,7 +40,10 @@
     ],
     "test:unit": [
       "@apps/brunch-agent#build",
+      "@hashintel/brunch-agent#build",
       "@hashintel/brunch-agent-binding-flue#build",
+      "@hashintel/brunch-agent-plugin-gherkin#build",
+      "@hashintel/brunch-agent-plugin-sdcpn#build",
       "@hashintel/brunch-agent-transport-aisdk#build",
       "@hashintel/petrinaut-core#build"
     ]

apps/brunch-agent/package.json

@@ -12,6 +12,8 @@
     "lint:eslint": "oxlint --type-aware --type-check --report-unused-disable-directives-severity=error .",
     "lint:tsc": "tsgo --noEmit",
     "petrinaut:dev": "vite dev --config petrinaut-local.vite.config.ts",
+    "runbook:elicit": "vite build && node --experimental-strip-types src/runbook-elicitation-run.ts",
+    "runbook:headless": "vite build && node --experimental-strip-types src/runbook-headless-run.ts",
     "test:unit": "vitest run --config vitest.config.ts",
     "transcript": "node --experimental-strip-types src/transcript-cli.ts"
   },

apps/brunch-agent/src/agents/chat-agent.ts

@@ -1,46 +1,68 @@
 "use agent";
 /**
- * One plain Flue chat agent for the Petrinaut panel throughline.
+ * One Flue chat agent for the Petrinaut panel throughline.
  *
- * Capture is a harness-side pipe, not an interviewer tool. One stub skill is
- * mounted so activation can appear in Flue history.
+ * Capture is a harness-side pipe, not an interviewer tool. One runbook skill
+ * carries the modelling lifecycle and its supporting resources.
  */
 
-import { defineSkill, useModel, useSkill, useTool } from "@flue/runtime";
+import { useInitialData, useModel, useSkill, useTool } from "@flue/runtime";
+import * as v from "valibot";
 
+import sdcpnModellingSkill from "../skills/sdcpn-modelling/SKILL.md";
+import {
+  petrinautConstructionTools,
+  VALIDATED_CONSTRUCTION_MODE,
+} from "../tools/petrinaut-construction.ts";
 import { ping } from "../tools/ping.ts";
 import { readPetrinautDoc } from "../tools/read-petrinaut-doc.ts";
 
 export const CHAT_MODEL_ID =
   process.env["BRUNCH_CHAT_MODEL"] || "claude-haiku-4-5";
 
-export const STUB_SKILL_NAME = "confirm-path";
+export const RUNBOOK_SKILL_NAME = sdcpnModellingSkill.name;
 
 export const ACTIVATE_SKILL_TOOL_NAME = "activate_skill";
 
-const confirmPath = defineSkill({
-  name: STUB_SKILL_NAME,
-  description:
-    "Confirm how this assistant is mounted. Use when checking the server path or tool layout.",
-  instructions:
-    "Say that ping confirms the server tool path. Then continue helping the user.",
-});
+export const chatAgentInitialDataSchema = v.optional(
+  v.object({
+    mode: v.literal(VALIDATED_CONSTRUCTION_MODE),
+  }),
+);
+
+export type ChatAgentInitialData = v.InferOutput<
+  typeof chatAgentInitialDataSchema
+>;
 
 export function ChatAgent() {
+  const initialData = useInitialData<ChatAgentInitialData>();
   useModel(`anthropic/${CHAT_MODEL_ID}`);
-  useSkill(confirmPath);
+  useSkill(sdcpnModellingSkill);
   useTool(ping);
   useTool(readPetrinautDoc);
-  return [
-    "You are a concise assistant inside the Petrinaut editor.",
+  if (initialData?.mode === VALIDATED_CONSTRUCTION_MODE) {
+    for (const constructionTool of petrinautConstructionTools) {
+      useTool(constructionTool);
+    }
+  }
+  const instructions = [
+    "You are the Brunch modelling assistant inside the Petrinaut editor.",
+    `Activate the \`${RUNBOOK_SKILL_NAME}\` skill before interviewing or constructing a process model.`,
+    "The Markdown IR is the shared workpiece of one looping lifecycle.",
     "Call ping when you need to confirm the server tool path.",
-    `Activate the \`${STUB_SKILL_NAME}\` skill before calling ping.`,
     "When the user asks how Petrinaut's UI works, call readPetrinautDoc.",
     "A client-tool-result signal is JSON [{ toolCallId, toolName, output }]. Treat output as the browser's result for that call and continue helping the user.",
-  ].join("\n");
+  ];
+  if (initialData?.mode === VALIDATED_CONSTRUCTION_MODE) {
+    instructions.push(
+      "This is a construct-only headless conversation. Use only the supplied runbook IR as modelling input, do not interview, and build the net through the mounted Petrinaut tools instead of emitting net JSON.",
+    );
+  }
+  return instructions.join("\n");
 }
 
 /**
  * Pinned, and never to be edited: conversation storage keys on this literal.
  */
 ChatAgent.agentName = "brunch-chat-agent";
+ChatAgent.initialData = chatAgentInitialDataSchema;

apps/brunch-agent/src/capture-sweep.ts

@@ -34,15 +34,22 @@ export interface CaptureSweepResult {
 const conversationUrl = (instanceId: string): string =>
   `http://brunch.local/agents/${CHAT_AGENT_ROUTE}/${instanceId}`;
 
-const ownedTransport = (identity: ConversationIdentity): typeof fetch => {
+const sourceAppTransport: typeof fetch = async (input, init) => {
+  const { default: app } = await import("./app.ts");
+  return app.fetch(input instanceof Request ? input : new Request(input, init));
+};
+
+const ownedTransport = (
+  identity: ConversationIdentity,
+  transport: typeof fetch,
+): typeof fetch => {
   const ownership = agentOwnershipHeaders(identity);
   return async (input, init) => {
-    const { default: app } = await import("./app.ts");
     const headers = new Headers(init?.headers);
     for (const [key, value] of Object.entries(ownership)) {
       headers.set(key, value);
     }
-    return app.fetch(
+    return transport(
       input instanceof Request
         ? new Request(input, { headers })
         : new Request(input, { ...init, headers }),
@@ -53,14 +60,15 @@ const ownedTransport = (identity: ConversationIdentity): typeof fetch => {
 export const applyCaptureSweep = async (
   identity: ConversationIdentity,
   userEntryIds: readonly string[],
+  transport: typeof fetch = sourceAppTransport,
 ): Promise<CaptureSweepResult> => {
   const instanceId = flueConversationIdFrom(identity);
   const store = createLocalCaptureStore(captureStorePath(instanceId), {
     ownerKey: identity.principalKey,
   });
   const historyReader = createFlueHistoryReader({
     resolveConversationUrl: conversationUrl,
-    transport: ownedTransport(identity),
+    transport: ownedTransport(identity, transport),
     archive: store,
   });
   const snapshot = await historyReader.read(instanceId);