refactor: remove experimental.task_system feature flag — task system is always on

Remove the task_system feature flag and all conditional branches, making
the task system unconditionally active. Strip taskSystemEnabled parameter
threading from 27 files (~580 lines removed). Add explicit guardrail in
Loom prompt to prevent direct plan execution (root cause of checkbox
regression). Remove dead todo finalization safety-net code from plugin
interface. Update all tests — 1371 pass, 0 fail.

Author: pgermishuys

.weave/plans/task-system-cleanup.md

@@ -28,8 +28,6 @@ export interface CreateBuiltinAgentsOptions {
   fingerprint?: ProjectFingerprint | null
   /** Custom agent metadata for Loom's dynamic delegation prompt */
   customAgentMetadata?: AvailableAgent[]
-  /** Whether the atomic task system is enabled (controls prompt content) */
-  taskSystemEnabled?: boolean
 }
 
 const AGENT_FACTORIES: Record<WeaveAgentName, AgentFactory> = {
@@ -183,7 +181,6 @@ export function createBuiltinAgents(options: CreateBuiltinAgentsOptions = {}): R
     resolveSkills,
     fingerprint,
     customAgentMetadata,
-    taskSystemEnabled,
   } = options
 
   const disabledSet = new Set(disabledAgents)
@@ -208,9 +205,9 @@ export function createBuiltinAgents(options: CreateBuiltinAgentsOptions = {}): R
     // so their prompts conditionally omit references to disabled agents
     let built: AgentConfig
     if (name === "loom") {
-      built = createLoomAgentWithOptions(resolvedModel, disabledSet, fingerprint, customAgentMetadata, taskSystemEnabled)
+      built = createLoomAgentWithOptions(resolvedModel, disabledSet, fingerprint, customAgentMetadata)
     } else if (name === "tapestry") {
-      built = createTapestryAgentWithOptions(resolvedModel, disabledSet, taskSystemEnabled)
+      built = createTapestryAgentWithOptions(resolvedModel, disabledSet)
     } else {
       built = buildAgent(factory, resolvedModel, {
         categories,

src/agents/builtin-agents.ts

@@ -223,17 +223,4 @@ describe("createLoomAgentWithOptions", () => {
     }])
     expect(config.mode).toBe("primary")
   })
-
-  it("uses task system prompt when taskSystemEnabled is true", () => {
-    const config = createLoomAgentWithOptions("claude-opus-4", undefined, null, undefined, true)
-    expect(config.prompt).toContain("TASK TRACKING")
-    expect(config.prompt).toContain("task_create")
-    expect(config.prompt).not.toContain("TODO OBSESSION")
-  })
-
-  it("returns composed prompt (not default) when only taskSystemEnabled is set", () => {
-    const config = createLoomAgentWithOptions("claude-opus-4", undefined, null, undefined, true)
-    // Should NOT fall through to LOOM_DEFAULTS since taskSystemEnabled triggers composition
-    expect(config.prompt).toContain("task_update")
-  })
 })

src/agents/loom/index.test.ts

@@ -16,14 +16,13 @@ export function createLoomAgentWithOptions(
   disabledAgents?: Set<string>,
   fingerprint?: ProjectFingerprint | null,
   customAgents?: AvailableAgent[],
-  taskSystemEnabled?: boolean,
 ): AgentConfig {
-  if ((!disabledAgents || disabledAgents.size === 0) && !fingerprint && (!customAgents || customAgents.length === 0) && !taskSystemEnabled) {
+  if ((!disabledAgents || disabledAgents.size === 0) && !fingerprint && (!customAgents || customAgents.length === 0)) {
     return { ...LOOM_DEFAULTS, model, mode: "primary" }
   }
   return {
     ...LOOM_DEFAULTS,
-    prompt: composeLoomPrompt({ disabledAgents, fingerprint, customAgents, taskSystemEnabled }),
+    prompt: composeLoomPrompt({ disabledAgents, fingerprint, customAgents }),
     model,
     mode: "primary",
   }

src/agents/loom/index.ts

@@ -50,6 +50,11 @@ describe("composeLoomPrompt", () => {
     expect(prompt).toContain("Tapestry runs Weft and Warp")
   })
 
+  it("contains delegation guardrail in PlanWorkflow section", () => {
+    const prompt = composeLoomPrompt()
+    expect(prompt).toContain("NEVER execute plan tasks directly")
+  })
+
   it("does not include ProjectContext with no fingerprint", () => {
     const prompt = composeLoomPrompt()
     expect(prompt).not.toContain("<ProjectContext>")
@@ -134,6 +139,12 @@ describe("buildDelegationSection", () => {
 })
 
 describe("buildPlanWorkflowSection", () => {
+  it("contains delegation guardrail at the top", () => {
+    const section = buildPlanWorkflowSection(new Set())
+    expect(section).toContain("NEVER execute plan tasks directly")
+    expect(section).toContain("/start-work")
+  })
+
   it("includes Pattern, Weft, Warp, and Tapestry by default", () => {
     const section = buildPlanWorkflowSection(new Set())
     expect(section).toContain("Pattern")
@@ -219,20 +230,30 @@ describe("individual section builders", () => {
     expect(buildRoleSection()).toContain("orchestrator")
   })
 
-  it("buildDisciplineSection contains TODO OBSESSION", () => {
-    expect(buildDisciplineSection()).toContain("TODO OBSESSION")
+  it("buildDisciplineSection contains TASK TRACKING", () => {
+    expect(buildDisciplineSection()).toContain("TASK TRACKING")
+  })
+
+  it("buildDisciplineSection contains task_create and task_update", () => {
+    expect(buildDisciplineSection()).toContain("task_create")
+    expect(buildDisciplineSection()).toContain("task_update")
+  })
+
+  it("buildDisciplineSection contains plan delegation guardrail", () => {
+    expect(buildDisciplineSection()).toContain("PLANS: Never execute plan tasks directly")
+    expect(buildDisciplineSection()).toContain("/start-work")
   })
 
   it("buildSidebarTodosSection contains format rules", () => {
     expect(buildSidebarTodosSection()).toContain("35 chars")
   })
 
-  it("buildSidebarTodosSection contains BEFORE FINISHING mandatory block", () => {
+  it("buildSidebarTodosSection uses task_create and task_update", () => {
     const section = buildSidebarTodosSection()
-    expect(section).toContain("BEFORE FINISHING (MANDATORY)")
-    expect(section).toContain("NON-NEGOTIABLE")
-    expect(section).toContain("final todowrite")
-    expect(section).not.toContain("sidebar hides")
+    expect(section).toContain("task_create")
+    expect(section).toContain("task_update")
+    expect(section).toContain("task_list")
+    expect(section).toContain("atomically")
   })
 
   it("buildDelegationNarrationSection contains duration hints", () => {
@@ -256,49 +277,6 @@ describe("individual section builders", () => {
   })
 })
 
-describe("task system enabled — Loom prompt sections", () => {
-  it("buildDisciplineSection uses task_create language when enabled", () => {
-    const section = buildDisciplineSection(true)
-    expect(section).toContain("TASK TRACKING")
-    expect(section).toContain("task_create")
-    expect(section).toContain("task_update")
-    expect(section).not.toContain("TODO OBSESSION")
-    expect(section).not.toContain("todowrite")
-  })
-
-  it("buildSidebarTodosSection uses task tools when enabled", () => {
-    const section = buildSidebarTodosSection(true)
-    expect(section).toContain("task_create")
-    expect(section).toContain("task_update")
-    expect(section).toContain("task_list")
-    expect(section).toContain("atomically")
-    expect(section).not.toContain("todowrite")
-    expect(section).not.toContain("BEFORE FINISHING")
-  })
-
-  it("buildDelegationNarrationSection uses task tools when enabled", () => {
-    const section = buildDelegationNarrationSection(new Set(), true)
-    expect(section).toContain("task_create")
-    expect(section).toContain("task_update")
-    expect(section).not.toContain("todowrite")
-  })
-
-  it("composeLoomPrompt with taskSystemEnabled replaces todowrite references", () => {
-    const prompt = composeLoomPrompt({ taskSystemEnabled: true })
-    expect(prompt).toContain("TASK TRACKING")
-    expect(prompt).toContain("task_create")
-    expect(prompt).not.toContain("TODO OBSESSION")
-    expect(prompt).not.toContain("BEFORE FINISHING (MANDATORY)")
-  })
-
-  it("composeLoomPrompt without taskSystemEnabled uses legacy todowrite", () => {
-    const prompt = composeLoomPrompt()
-    expect(prompt).toContain("TODO OBSESSION")
-    expect(prompt).toContain("todowrite")
-    expect(prompt).not.toContain("TASK TRACKING")
-  })
-})
-
 describe("buildCustomAgentDelegationSection", () => {
   const makeCustomAgent = (name: string, domain: string, trigger: string): AvailableAgent => ({
     name,

src/agents/loom/prompt-composer.test.ts

@@ -18,8 +18,6 @@ export interface LoomPromptOptions {
   fingerprint?: ProjectFingerprint | null
   /** Custom agent metadata for dynamic delegation sections */
   customAgents?: AvailableAgent[]
-  /** Whether the atomic task system is enabled (default false for backward compat) */
-  taskSystemEnabled?: boolean
 }
 
 export function buildRoleSection(): string {
@@ -30,32 +28,22 @@ You are the team lead. Understand the request, break it into tasks, delegate int
 </Role>`
 }
 
-export function buildDisciplineSection(taskSystemEnabled = false): string {
-  if (taskSystemEnabled) {
-    return `<Discipline>
+export function buildDisciplineSection(): string {
+  return `<Discipline>
 TASK TRACKING (NON-NEGOTIABLE):
 - 2+ steps → task_create FIRST, one task per step
 - task_update to in_progress before starting (ONE at a time)
 - task_update to completed IMMEDIATELY after each step
 - NEVER batch completions
 
-No tasks on multi-step work = INCOMPLETE WORK.
-</Discipline>`
-  }
-  return `<Discipline>
-TODO OBSESSION (NON-NEGOTIABLE):
-- 2+ steps → todowrite FIRST, atomic breakdown
-- Mark in_progress before starting (ONE at a time)
-- Mark completed IMMEDIATELY after each step
-- NEVER batch completions
+PLANS: Never execute plan tasks directly — always /start-work → Tapestry.
 
-No todos on multi-step work = INCOMPLETE WORK.
+No tasks on multi-step work = INCOMPLETE WORK.
 </Discipline>`
 }
 
-export function buildSidebarTodosSection(taskSystemEnabled = false): string {
-  if (taskSystemEnabled) {
-    return `<SidebarTodos>
+export function buildSidebarTodosSection(): string {
+  return `<SidebarTodos>
 The user sees a Todo sidebar (~35 char width). Use task_create and task_update to manage it:
 
 - task_create: creates a task and syncs to sidebar automatically
@@ -70,36 +58,6 @@ WORKFLOW:
 3. task_update({ id: "T-xxx", status: "completed" }) — when done
 
 FORMAT: Keep subjects under 35 chars. One in_progress at a time.
-</SidebarTodos>`
-  }
-  return `<SidebarTodos>
-The user sees a Todo sidebar (~35 char width). Use todowrite strategically:
-
-WHEN PLANNING (multi-step work):
-- Create "in_progress": "Planning: [brief desc]"
-- When plan ready: mark completed, add "Plan ready — /start-work"
-
-WHEN DELEGATING TO AGENTS:
-- FIRST: Create "in_progress": "[agent]: [task]" (e.g. "thread: scan models")
-- The todowrite call MUST come BEFORE the Task/call_weave_agent tool call in your response
-- Mark "completed" AFTER summarizing what the agent returned
-- If multiple delegations: one todo per active agent
-
-WHEN DOING QUICK TASKS (no plan needed):
-- One "in_progress" todo for current step
-- Mark "completed" immediately when done
-
-FORMAT RULES:
-- Max 35 chars per todo content
-- Max 5 visible todos at any time
-- in_progress = yellow highlight — use for ACTIVE work only
-- Prefix delegations with agent name
-
-BEFORE FINISHING (MANDATORY):
-- ALWAYS issue a final todowrite before your last response
-- Mark ALL in_progress items → "completed" (or "cancelled")
-- Never leave in_progress items when done
-- This is NON-NEGOTIABLE — skipping it breaks the UI
 </SidebarTodos>`
 }
 
@@ -141,7 +99,7 @@ ${lines.join("\n")}
 </Delegation>`
 }
 
-export function buildDelegationNarrationSection(disabled: Set<string> = new Set(), taskSystemEnabled = false): string {
+export function buildDelegationNarrationSection(disabled: Set<string> = new Set()): string {
   const hints: string[] = []
   if (isAgentEnabled("pattern", disabled)) {
     hints.push('- Pattern (planning): "This may take a moment — Pattern is researching the codebase and writing a detailed plan..."')
@@ -159,18 +117,6 @@ export function buildDelegationNarrationSection(disabled: Set<string> = new Set(
     ? `\nDURATION HINTS — tell the user when something takes time:\n${hints.join("\n")}`
     : ""
 
-  const trackingTool = taskSystemEnabled ? "task_create/task_update" : "todowrite"
-  const trackingInstr = taskSystemEnabled
-    ? `2. BEFORE the Task tool call: Create a task (in_progress) for the delegation using task_create.
-   This ensures the sidebar updates immediately, not after the subagent finishes.`
-    : `2. BEFORE the Task tool call: Create/update a sidebar todo (in_progress) for the delegation.
-   The todowrite call MUST appear BEFORE the Task tool call in your response.
-   This ensures the sidebar updates immediately, not after the subagent finishes.`
-
-  const completeInstr = taskSystemEnabled
-    ? `4. Mark the delegation task as "completed" using task_update after summarizing results.`
-    : `4. Mark the delegation todo as "completed" after summarizing results.`
-
   return `<DelegationNarration>
 EVERY delegation MUST follow this pattern — no exceptions:
 
@@ -179,14 +125,15 @@ EVERY delegation MUST follow this pattern — no exceptions:
    - "Asking Pattern to create an implementation plan for the new feature..."
    - "Sending to Spindle to research the library's API docs..."
 
-${trackingInstr}
+2. BEFORE the Task tool call: Create a task (in_progress) for the delegation using task_create.
+   This ensures the sidebar updates immediately, not after the subagent finishes.
 
 3. AFTER the agent returns: Write a brief summary of what was found/produced:
    - "Thread found 3 files related to auth: src/auth/login.ts, src/auth/session.ts, src/auth/middleware.ts"
    - "Pattern saved the plan to .weave/plans/feature-x.md with 7 tasks"
    - "Spindle confirmed the library supports streaming — docs at [url]"
 
-${completeInstr}
+4. Mark the delegation task as "completed" using task_update after summarizing results.
 ${hintsBlock}
 
 The user should NEVER see a blank pause with no explanation. If you're about to call Task, WRITE SOMETHING FIRST.
@@ -250,6 +197,10 @@ export function buildPlanWorkflowSection(disabled: Set<string>): string {
 - SKIP plan workflow: Quick fixes, single-file changes, simple questions`)
 
   return `<PlanWorkflow>
+CRITICAL: NEVER execute plan tasks directly. ALWAYS delegate to Tapestry via /start-work.
+Only Tapestry has the plan execution protocol (checkbox marking, verification, post-execution review).
+If you execute plan tasks yourself, checkboxes won't be marked and progress will be lost.
+
 For complex tasks that benefit from structured planning before execution:
 
 ${steps.join("\n")}
@@ -351,15 +302,14 @@ export function composeLoomPrompt(options: LoomPromptOptions = {}): string {
   const disabled = options.disabledAgents ?? new Set()
   const fingerprint = options.fingerprint
   const customAgents = options.customAgents ?? []
-  const taskSystemEnabled = options.taskSystemEnabled ?? false
 
   const sections = [
     buildRoleSection(),
     buildProjectContextSection(fingerprint),
-    buildDisciplineSection(taskSystemEnabled),
-    buildSidebarTodosSection(taskSystemEnabled),
+    buildDisciplineSection(),
+    buildSidebarTodosSection(),
     buildDelegationSection(disabled),
-    buildDelegationNarrationSection(disabled, taskSystemEnabled),
+    buildDelegationNarrationSection(disabled),
     buildCustomAgentDelegationSection(customAgents, disabled),
     buildPlanWorkflowSection(disabled),
     buildReviewWorkflowSection(disabled),

src/agents/loom/prompt-composer.ts

@@ -9,14 +9,14 @@ export type { TapestryPromptOptions } from "./prompt-composer"
 /**
  * Create a Tapestry agent config with optional disabled agents for prompt composition.
  */
-export function createTapestryAgentWithOptions(model: string, disabledAgents?: Set<string>, taskSystemEnabled?: boolean): AgentConfig {
-  if ((!disabledAgents || disabledAgents.size === 0) && !taskSystemEnabled) {
+export function createTapestryAgentWithOptions(model: string, disabledAgents?: Set<string>): AgentConfig {
+  if (!disabledAgents || disabledAgents.size === 0) {
     return { ...TAPESTRY_DEFAULTS, tools: { ...TAPESTRY_DEFAULTS.tools }, model, mode: "primary" }
   }
   return {
     ...TAPESTRY_DEFAULTS,
     tools: { ...TAPESTRY_DEFAULTS.tools },
-    prompt: composeTapestryPrompt({ disabledAgents, taskSystemEnabled }),
+    prompt: composeTapestryPrompt({ disabledAgents }),
     model,
     mode: "primary",
   }