fix(dag): make YAML authoring explicit

packages/core/src/plugin/command/orchestration-policy.md

@@ -159,18 +159,16 @@ continue QA, reduce scope, use `standard`, or explicitly waive. A `WAIVED`
 start is informed only when both `waiver_reason` and `acknowledged_risks` are
 non-empty; preserve them for audit.
 
-Do not supply `protocol_version`, `state`, or `fingerprint` in the admission
-input. Those are durable audit fields owned by the workflow boundary:
-it sets protocol version 1, initializes state from the verdict, normalizes the
-Brief for fingerprint computation, and computes the lowercase hexadecimal
-SHA-256 hash. A successful deep start alone transitions the durable record to
-`CONSUMED`.
+The author-written admission input accepts only `brief_revision`, `qa_mode`,
+`verdict`, `brief`, and, for an informed waiver, `waiver_reason` and
+`acknowledged_risks`. Do not copy any additional fields from a persisted
+workflow or tool response. The workflow boundary creates and advances its
+durable audit record.
 
 Material changes to goal, scope, constraints, assumptions, or acceptance
-criteria create a new brief revision, invalidate the prior fingerprint, and
-return admission to questioning. The workflow boundary generates the
-replacement fingerprint from the revised Brief. Do not replay QA from a
-consumed record after recovery.
+criteria create a new brief revision, invalidate the prior admission record,
+and return admission to questioning. Do not replay QA from a consumed record
+after recovery.
 
 ## Role Resolution
 
@@ -185,8 +183,8 @@ If a required capability has no eligible role, report the missing capability and
 
 ## Model Assignment
 
-Never emit `node.model` or `config.node_defaults.model`. Model assignment belongs
-to runtime configuration, not the workflow graph:
+Workflow YAML has no model-selection field. Model assignment belongs to
+runtime configuration, not the workflow graph:
 
 `dag.jsonc` tier → configured agent model → parent session model
 

packages/core/src/plugin/command/workflow-blocks.md

@@ -4,12 +4,23 @@ Blocks are the high-level interface for assembling a one-off workflow YAML
 file. The tool compiles them into ordinary durable DAG nodes before validation
 and persistence. Existing node-based YAML remains compatible.
 
-## Shape
+## Authoring contract
 
-Use `objective` and `blocks` inside `config` for **start**, or alongside
-`blocks` for **extend**. A replan uses the same fields inside `fragment`.
+Never infer or invent a YAML field. Copy the envelope for the intended action
+and change values only. Unknown fields are errors; a field from a tool call,
+runtime result, low-level node, or another action does not belong here unless
+it is explicitly listed below.
+
+### Start file
+
+The author-written top-level fields are optional `title`, optional `mode`,
+optional `admission`, and required `config`. For the block route, `config`
+contains required `name`, `objective`, and `blocks`; its only optional fields
+are `node_defaults`, `max_concurrency`, `max_node_replan_attempts`, and
+`max_total_nodes`.
 
 ```yaml
+title: Implement session recovery
 config:
   name: implement-session-recovery
   objective: Implement session recovery with focused tests and evidence-backed review.
@@ -33,6 +44,69 @@ config:
       depends_on: [verify]
 ```
 
+### Extend file
+
+An extend file contains exactly `objective` and `blocks`. It has no `config`,
+`name`, or `fragment` wrapper.
+
+```yaml
+objective: Add regression coverage for the newly confirmed recovery edge case.
+blocks:
+  - id: recovery-fix
+    kind: coding
+    instruction: Implement only the confirmed edge-case fix and its regression test.
+  - id: recovery-verify
+    kind: verify
+    depends_on: [recovery-fix]
+```
+
+### Replan file
+
+A replan file contains only `fragment`. For the block route, `fragment` has the
+same fields as start's `config`: required `name`, `objective`, and `blocks`,
+plus the four optional config fields listed above.
+
+```yaml
+fragment:
+  name: recover-session-recovery
+  objective: Replace the invalid route with a diagnosed, verified repair path.
+  blocks:
+    - id: diagnose
+      kind: debug
+      instruction: Reproduce the failure and identify the evidence-backed root cause.
+    - id: repair
+      kind: coding
+      depends_on: [diagnose]
+      instruction: Apply the smallest repair supported by the diagnosis.
+    - id: verify
+      kind: verify
+      depends_on: [repair]
+```
+
+Every block accepts only `id`, `kind`, `depends_on`, `instruction`,
+`worker_type`, `required`, and `report_to_parent`. `id` and `kind` are required.
+Allowed `kind` values are `explore`, `plan`, `prototype`, `debug`, `coding`,
+`verify`, `review`, and `synthesize`. Omit `worker_type` unless the exact
+configured agent name is already known; never invent one.
+
+For optional `node_defaults`, the only fields are `required`,
+`report_to_parent`, and `worker_config`; `worker_config` accepts only
+`timeout_ms`.
+
+`action`, `workflow_id`, `operation`, and `spec_path` are tool-call fields, not
+YAML fields. `profile` is also a tool-call field used only by validate. The
+listed YAML fields are exhaustive: do not copy extra fields from read/status
+output or persisted runtime records. Deep admission is the only exception to
+the minimal start envelope; load `guide(topic=policy)` and copy its
+author-written admission shape instead of guessing fields.
+
+Use these exact calls after writing the file:
+
+- validate: `{ action: "validate", spec_path: "workflow.yaml", profile: "portable" }`
+- start: `{ action: "start", spec_path: "workflow.yaml" }`
+- extend: `{ action: "extend", workflow_id: "dag_...", spec_path: "extend.yaml" }`
+- replan: `{ action: "control", operation: "replan", workflow_id: "dag_...", spec_path: "replan.yaml" }`
+
 This guide owns the author-written block fields and semantics. The action
 schema stays shallow and accepts only `spec_path`; the YAML validator rejects
 unknown or missing graph fields by name and reports each error with its path.

packages/core/src/plugin/command/workflow.md

@@ -44,10 +44,10 @@ file and retry with the same path.
 
 Before a deep start, qualify the request interactively in the parent session.
 The start spec places `mode: deep`, a versioned `READY` or informed `WAIVED`
-admission input, and `config` at the same level. The admission input contains
-`brief_revision`, `qa_mode`, `verdict`, `brief`, and waiver audit fields when
-applicable; the workflow boundary owns `protocol_version`, `state`, and
-`fingerprint`. Do not put admission QA inside the graph: its answers define the
+admission input, and `config` at the same level. The admission input accepts
+only `brief_revision`, `qa_mode`, `verdict`, `brief`, and waiver audit fields
+when applicable. Do not copy additional fields from persisted records or tool
+responses. Do not put admission QA inside the graph: its answers define the
 graph. Use the orchestration policy below for QA modes, round budgets, verdict
 recovery, revision invalidation, and waiver audit fields.
 
@@ -142,8 +142,9 @@ config:
       timeout_ms: 600000
 ```
 
-Never emit `node.model` or `config.node_defaults.model`. Model selection is
-configuration-owned: critical nodes (`required: true` and review workers) use
+The listed node and default fields are exhaustive; workflow YAML has no
+model-selection field. Model selection is configuration-owned: critical nodes
+(`required: true` and review workers) use
 the `advanced` tier in `dag.jsonc`, other nodes use `standard`, then resolution
 falls back to the selected agent model and the parent-session model. If no
 source provides a model, the workflow tool starts parent-session QA and leaves
@@ -467,11 +468,10 @@ is believed to be running. Two disciplines close the gap:
 
 ## Model Assignment Strategy
 
-Workflow definitions MUST NOT specify `node.model` or
-`config.node_defaults.model`. Resolution follows the `dag.jsonc` tier, then the
-configured agent model, then the parent-session model. If all three are absent,
-the workflow tool asks the user to configure a model and does not create the
-workflow.
+Workflow YAML has no model-selection field. Resolution follows the `dag.jsonc`
+tier, then the configured agent model, then the parent-session model. If all
+three are absent, the workflow tool asks the user to configure a model and does
+not create the workflow.
 
 - Expensive models for planning, review, and arbitration — high-stakes decisions where reasoning quality matters.
 - Fast models for mechanical implementation — well-specified edits where speed and cost matter.

packages/core/test/plugin/command.test.ts

@@ -124,6 +124,32 @@ describe("CommandPlugin.Plugin", () => {
     }),
   )
 
+  it.effect("publishes an exact block YAML authoring contract", () =>
+    Effect.sync(() => {
+      expect(CommandPlugin.WorkflowBlocksContent).toContain("Never infer or invent a YAML field")
+      expect(CommandPlugin.WorkflowBlocksContent).toContain("### Start file")
+      expect(CommandPlugin.WorkflowBlocksContent).toContain("### Extend file")
+      expect(CommandPlugin.WorkflowBlocksContent).toContain("### Replan file")
+      expect(CommandPlugin.WorkflowBlocksContent).toMatch(
+        /`id`, `kind`, `depends_on`, `instruction`,\s+`worker_type`, `required`, and `report_to_parent`/,
+      )
+      expect(CommandPlugin.WorkflowBlocksContent).toMatch(
+        /`action`, `workflow_id`, `operation`, and `spec_path` are tool-call fields/,
+      )
+      const authoringContract = CommandPlugin.WorkflowBlocksContent.slice(
+        CommandPlugin.WorkflowBlocksContent.indexOf("## Authoring contract"),
+        CommandPlugin.WorkflowBlocksContent.indexOf("This guide owns"),
+      )
+      expect(authoringContract).toMatch(/[Tt]he\s+listed YAML fields are exhaustive/)
+      expect(authoringContract).not.toMatch(
+        /session_id|project_id|protocol_version|fingerprint|node_defaults\.model|node\.model/,
+      )
+      expect(CommandPlugin.WorkflowBlocksContent).toContain(
+        '{ action: "control", operation: "replan", workflow_id: "dag_...", spec_path: "replan.yaml" }',
+      )
+    }),
+  )
+
   it.effect("preserves opt-outs read-only scope and explicit role assignments", () =>
     Effect.sync(() => {
       expect(CommandPlugin.OrchestrationPolicyContent).toContain("single agent")
@@ -137,9 +163,7 @@ describe("CommandPlugin.Plugin", () => {
 
   it.effect("documents config-first model fallback without invented identifiers", () =>
     Effect.sync(() => {
-      expect(CommandPlugin.OrchestrationPolicyContent).toContain(
-        "Never emit `node.model` or `config.node_defaults.model`",
-      )
+      expect(CommandPlugin.OrchestrationPolicyContent).toContain("Workflow YAML has no model-selection field")
       expect(CommandPlugin.OrchestrationPolicyContent).toContain(
         "`dag.jsonc` tier → configured agent model → parent session model",
       )
@@ -338,14 +362,17 @@ describe("CommandPlugin.Plugin", () => {
       expect(CommandPlugin.OrchestrationPolicyContent).toContain("waiver_reason")
       expect(CommandPlugin.OrchestrationPolicyContent).toContain("acknowledged_risks")
       expect(CommandPlugin.OrchestrationPolicyContent).toContain("Material changes")
-      expect(CommandPlugin.OrchestrationPolicyContent).toContain("invalidate the prior fingerprint")
-      expect(CommandPlugin.OrchestrationPolicyContent).toContain("SHA-256 hash")
-      expect(CommandPlugin.WorkflowFactsContent).toContain(
-        "The start spec places `mode: deep`, a versioned `READY` or informed `WAIVED`",
+      expect(CommandPlugin.OrchestrationPolicyContent).toContain("invalidate the prior admission record")
+      const admissionPolicy = CommandPlugin.OrchestrationPolicyContent.slice(
+        CommandPlugin.OrchestrationPolicyContent.indexOf("## Deep Admission QA"),
+        CommandPlugin.OrchestrationPolicyContent.indexOf("## Role Resolution"),
       )
+      expect(admissionPolicy).not.toMatch(/protocol_version|fingerprint/)
       expect(CommandPlugin.WorkflowFactsContent).toContain(
-        "the workflow boundary owns `protocol_version`, `state`, and\n`fingerprint`",
+        "The start spec places `mode: deep`, a versioned `READY` or informed `WAIVED`",
       )
+      expect(CommandPlugin.WorkflowFactsContent).toContain("The admission input accepts\nonly `brief_revision`")
+      expect(CommandPlugin.WorkflowFactsContent).not.toMatch(/protocol_version|node_defaults\.model|node\.model/)
       expect(CommandPlugin.WorkflowFactsContent).toContain("A one-off graph may use a")
       expect(CommandPlugin.WorkflowFactsContent).toContain("task-local file")
       expect(CommandPlugin.WorkflowFactsContent).not.toContain("`config.mode`")
@@ -388,11 +415,9 @@ describe("CommandPlugin.Plugin", () => {
       expect(CommandPlugin.WorkflowFactsContent).not.toContain("Gate failure cancels the workflow automatically")
       expect(CommandPlugin.WorkflowFactsContent).toContain("Static `prompt_template.input`")
       expect(CommandPlugin.WorkflowFactsContent).toContain("it must\nnever appear as `[object Object]`")
-      expect(CommandPlugin.WorkflowFactsContent).toContain(
-        "Workflow definitions MUST NOT specify `node.model` or\n`config.node_defaults.model`",
-      )
+      expect(CommandPlugin.WorkflowFactsContent).toContain("Workflow YAML has no model-selection field")
       expect(CommandPlugin.WorkflowFactsContent).toMatch(
-        /`dag\.jsonc` tier, then the\s+configured agent model, then the parent-session model/,
+        /`dag\.jsonc`\s+tier, then the\s+configured agent model, then the parent-session model/,
       )
       expect(CommandPlugin.WorkflowFactsContent).toContain("Propose-then-assemble")
       const reviewExample = CommandPlugin.WorkflowFactsContent.slice(

packages/opencode/test/dag/workflow-authoring.test.ts

@@ -1,6 +1,7 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
 import { CrossSpawnSpawner } from "@opencode-ai/core/cross-spawn-spawner"
+import { CommandPlugin } from "@opencode-ai/core/plugin/command"
 import { WorkflowAuthoring } from "../../src/dag/authoring"
 import { DagValidation } from "../../src/dag/validation"
 import { testEffect } from "../lib/effect"
@@ -23,6 +24,35 @@ const start = {
 }
 
 describe("WorkflowAuthoring source-to-graph seam", () => {
+  it.effect("keeps every block-guide YAML envelope executable", () =>
+    Effect.gen(function* () {
+      const guide = CommandPlugin.WorkflowBlocksContent
+      const example = (heading: string) => {
+        const section = guide.slice(guide.indexOf(`### ${heading}`))
+        const match = section.match(/```yaml\n([\s\S]*?)```/)
+        expect(match?.[1]).toBeDefined()
+        return match?.[1] ?? ""
+      }
+      const authoring = WorkflowAuthoring.make()
+      const inputs = [
+        { action: "start" as const, content: example("Start file") },
+        { action: "extend" as const, content: example("Extend file") },
+        { action: "replan" as const, content: example("Replan file") },
+      ]
+
+      for (const input of inputs) {
+        const result = yield* authoring.prepare({
+          action: input.action,
+          source: { kind: "yaml", source: `${input.action}.yaml`, content: input.content },
+          profile: "portable",
+        })
+        expect(result.errors).toEqual([])
+        expect(result.valid).toBe(true)
+        expect(result.prepared?.nodes.length).toBeGreaterThan(0)
+      }
+    }),
+  )
+
   it.effect("prepares start, extend, and replan through one action-aware interface", () =>
     Effect.gen(function* () {
       const authoring = WorkflowAuthoring.make()