fix(background-task): clarify task id contracts

src/agents/anti-duplication.test.ts

@@ -51,7 +51,7 @@ describe("buildAntiDuplicationSection", () => {
     expect(result).toContain("Wait for Results Properly")
     expect(result).toContain("End your response")
     expect(result).toContain("Wait for the completion notification")
-    expect(result).toContain("background_output")
+    expect(result).toContain('background_output(task_id="bg_...")')
   })
 
   it("#given no arguments #when building #then explains why this matters", () => {

src/agents/atlas/atlas-prompt.test.ts

@@ -126,6 +126,13 @@ describe("Atlas prompts use task_id (not session_id) for retries", () => {
       expect(prompt, `${name}: missing task_id retry reference`).toMatch(/task_id/)
     }
   })
+
+  test("all variants should separate background ids from continuation task ids", () => {
+    for (const [name, prompt] of ALL_VARIANTS) {
+      expect(prompt, `${name}: missing bg result collection contract`).toContain('background_output(task_id="bg_...")')
+      expect(prompt, `${name}: missing ses continuation contract`).toContain('task(task_id="ses_..."')
+    }
+  })
 })
 
 describe("Atlas prompts no-excuses retry policy", () => {

src/agents/atlas/default-prompt-sections.ts

@@ -243,6 +243,6 @@ export const DEFAULT_ATLAS_CRITICAL_RULES = `<critical_overrides>
 - Run lsp_diagnostics after every delegation
 - Pass inherited wisdom to every subagent
 - Verify with your own tools
-- **Store task_id from every delegation output**
+- **Store continuation task_id (\`ses_...\`) from every delegation output**
-- **Use \`task_id="{task_id}"\` for retries, fixes, and follow-ups**
+- **Use \`task(task_id="ses_...", prompt="...")\` for retries, fixes, and follow-ups**
 </critical_overrides>`

src/agents/atlas/kimi-prompt-sections.ts

@@ -216,6 +216,6 @@ export const KIMI_ATLAS_CRITICAL_RULES = `<critical_overrides>
 - Run lsp_diagnostics after every delegation
 - Pass inherited wisdom to every subagent
 - Verify with your own tools
-- **Store task_id from every delegation output**
+- **Store continuation task_id (\`ses_...\`) from every delegation output**
-- **Use \`task_id="{task_id}"\` for retries, fixes, and follow-ups**
+- **Use \`task(task_id="ses_...", prompt="...")\` for retries, fixes, and follow-ups**
 </critical_overrides>`

src/agents/atlas/opus-4-7-prompt-sections.ts

@@ -232,6 +232,6 @@ export const OPUS_47_ATLAS_CRITICAL_RULES = `<critical_overrides>
 - Run lsp_diagnostics after every delegation
 - Pass inherited wisdom to every subagent
 - Verify with your own tools
-- **Store task_id from every delegation output**
+- **Store continuation task_id (\`ses_...\`) from every delegation output**
-- **Use \`task_id="{task_id}"\` for retries, fixes, and follow-ups**
+- **Use \`task(task_id="ses_...", prompt="...")\` for retries, fixes, and follow-ups**
 </critical_overrides>`

src/agents/atlas/shared-prompt.ts

@@ -120,7 +120,8 @@ task(category="quick", load_skills=[], run_in_background=false, prompt="...task
 - **Task execution** (\`category="..."\`): \`run_in_background=false\` — blocks for verification
 
 **Background management:**
-- Collect: \`background_output(task_id="...")\`
+- Collect with background task IDs (\`bg_...\`): \`background_output(task_id="bg_...")\`
+- Continue follow-ups with continuation task IDs (\`ses_...\`): \`task(task_id="ses_...")\`
 - Cancel DISPOSABLE background tasks individually before final answer: \`background_cancel(taskId="bg_explore_xxx")\`
 - **NEVER \`background_cancel(all=true)\`** — it kills tasks whose output you have not collected.
 </parallel_by_default>`

src/agents/dynamic-agent-policy-sections.ts

@@ -148,7 +148,7 @@ When you need the delegated results but they're not ready:
 
 1. **End your response** - do NOT continue with work that depends on those results
 2. **Wait for the completion notification** - the system will trigger your next turn
-3. **Then** collect results via \`background_output(task_id="...")\`
+3. **Then** collect results via \`background_output(task_id="bg_...")\`
 4. **Do NOT** impatiently re-search the same topics while waiting
 
 ### Why This Matters:

src/agents/hephaestus-id-contract.test.ts

@@ -0,0 +1,30 @@
+/// <reference types="bun-types" />
+
+import { describe, expect, test } from "bun:test"
+import { buildHephaestusPrompt as buildGptHephaestusPrompt } from "./hephaestus/gpt"
+import { buildHephaestusPrompt as buildGpt53CodexHephaestusPrompt } from "./hephaestus/gpt-5-3-codex"
+import { buildHephaestusPrompt as buildGpt54HephaestusPrompt } from "./hephaestus/gpt-5-4"
+import { buildGpt55HephaestusPrompt } from "./hephaestus/gpt-5-5"
+
+describe("Hephaestus background task ID guidance", () => {
+  const promptBuilders = [
+    ["gpt", () => buildGptHephaestusPrompt()],
+    ["gpt-5.3-codex", () => buildGpt53CodexHephaestusPrompt()],
+    ["gpt-5.4", () => buildGpt54HephaestusPrompt()],
+    ["gpt-5.5", () => buildGpt55HephaestusPrompt([])],
+  ] as const
+
+  for (const [name, buildPrompt] of promptBuilders) {
+    test(`#given ${name} prompt #when describing task follow-ups #then bg ids and continuation ids are disambiguated`, () => {
+      // given, when
+      const prompt = buildPrompt()
+
+      // then
+      expect(prompt).toContain("background task IDs (`bg_...`)")
+      expect(prompt).toContain("continuation IDs (`ses_...`)")
+      expect(prompt).toContain("background_output(task_id=\"bg_...\")")
+      expect(prompt).toContain("task(task_id=\"ses_...\")")
+      expect(prompt).not.toContain("returns a task_id")
+    })
+  }
+})

src/agents/hephaestus/gpt-5-3-codex.ts

@@ -299,7 +299,7 @@ Prompt structure for each agent:
 - Parallelize independent file reads - don't read files one at a time
 - NEVER use \`run_in_background=false\` for explore/librarian
 - Continue only with non-overlapping work after launching background agents
-- Collect results with \`background_output(task_id="...")\` when needed
+- Keep IDs separate: collect results with background task IDs (\`bg_...\`) via \`background_output(task_id="bg_...")\`; continue follow-up sessions with continuation IDs (\`ses_...\`) via \`task(task_id="ses_...")\`
 - BEFORE final answer, cancel DISPOSABLE tasks individually: \`background_cancel(taskId="bg_explore_xxx")\`, \`background_cancel(taskId="bg_librarian_xxx")\`
 - **NEVER use \`background_cancel(all=true)\`** - it kills tasks whose results you haven't collected yet
 
@@ -407,11 +407,11 @@ After delegation, ALWAYS verify: works as expected? follows codebase pattern? MU
 
 ### Session Continuity
 
-Every \`task()\` output includes a session_id. **USE IT for follow-ups.**
+Every \`task()\` output includes a continuation ID (\`ses_...\`). **USE IT for follow-ups.**
 
-- **Task failed/incomplete** - \`task_id="{id}", prompt="Fix: {error}"\`
+- **Task failed/incomplete** - \`task(task_id="ses_...", prompt="Fix: {error}")\`
-- **Follow-up on result** - \`task_id="{id}", prompt="Also: {question}"\`
+- **Follow-up on result** - \`task(task_id="ses_...", prompt="Also: {question}")\`
-- **Verification failed** - \`task_id="{id}", prompt="Failed: {error}. Fix."\`
+- **Verification failed** - \`task(task_id="ses_...", prompt="Failed: {error}. Fix.")\`
 
 ${
   oracleSection

src/agents/hephaestus/gpt-5-4.ts

@@ -111,6 +111,8 @@ export function buildHephaestusPrompt(
   const identityBlock = `<identity>
 You are Hephaestus, an autonomous deep worker for software engineering.
 
+ID contract: background task IDs (\`bg_...\`) use \`background_output(task_id="bg_...")\`; continuation IDs (\`ses_...\`) use \`task(task_id="ses_...")\`.
+
 You communicate warmly and directly, like a senior colleague walking through a problem together. You explain the why behind decisions, not just the what. You stay concise in volume but generous in clarity - every sentence carries meaning.
 
 You build context by examining the codebase first without assumptions. You think through the nuances of the code you encounter. You persist until the task is fully handled end-to-end, even when tool calls fail. You only end your turn when the problem is solved and verified.
@@ -234,7 +236,7 @@ Agent prompt structure:
 - [REQUEST]: What to find, format to return, what to skip
 
 Background task management:
-- Collect results with \`background_output(task_id="...")\` when completed
+- Keep IDs separate: collect results with background task IDs (\`bg_...\`) via \`background_output(task_id="bg_...")\`; continue follow-up sessions with continuation IDs (\`ses_...\`) via \`task(task_id="ses_...")\`
 - Before final answer, cancel disposable tasks individually: \`background_cancel(taskId="...")\`
 - Never use \`background_cancel(all=true)\` - it kills tasks whose results you have not collected yet
 
@@ -312,10 +314,10 @@ Every delegation prompt needs these 6 sections:
 After delegation, verify by reading every file the subagent touched. Check: works as expected? follows codebase pattern? Do not trust self-reports.
 
 <session_continuity>
-Every \`task()\` returns a task_id. Use it for all follow-ups:
+Every \`task()\` output includes a continuation ID (\`ses_...\`). Use it for all follow-ups:
-- Task failed/incomplete: \`task_id="{id}", prompt="Fix: {error}"\`
+- Task failed/incomplete: \`task(task_id="ses_...", prompt="Fix: {error}")\`
-- Follow-up on result: \`task_id="{id}", prompt="Also: {question}"\`
+- Follow-up on result: \`task(task_id="ses_...", prompt="Also: {question}")\`
-- Verification failed: \`task_id="{id}", prompt="Failed: {error}. Fix."\`
+- Verification failed: \`task(task_id="ses_...", prompt="Failed: {error}. Fix.")\`
 
 This preserves full context, avoids repeated exploration, saves 70%+ tokens.
 </session_continuity>

src/agents/hephaestus/gpt-5-5.ts

@@ -22,6 +22,8 @@ function buildTaskSystemGuide(useTaskSystem: boolean): string {
 
 const HEPHAESTUS_GPT_5_5_TEMPLATE = `You are Hephaestus, an autonomous deep worker based on GPT-5.5. You and the user share one workspace. You receive goals, not step-by-step instructions, and execute them end-to-end.
 
+ID contract: background task IDs (\`bg_...\`) use \`background_output(task_id="bg_...")\`; continuation IDs (\`ses_...\`) use \`task(task_id="ses_...")\`.
+
 # Tone
 
 Warm but spare. Communicate efficiently - enough context for the user to trust the work, then stop. No flattery, no narration, no padding. Acknowledge real progress briefly; never invent it.
@@ -172,7 +174,7 @@ AGENTS.md files in your context carry directory-scoped conventions. Obey them fo
 **\`task()\`** for both research sub-agents and category-based delegation. Allowed: \`subagent_type="explore"\`, \`"librarian"\`, \`"oracle"\`, or \`category="..."\`.
 
 - Every \`task()\` call needs \`load_skills\` (an empty array \`[]\` is valid).
-- Reuse \`task_id\` for follow-ups; never start a fresh session on a continuation. Saves 70%+ of tokens and preserves the sub-agent's full context.
+- Reuse continuation IDs (\`ses_...\`) for follow-ups via \`task(task_id="ses_...")\`; never pass background task IDs (\`bg_...\`) to \`task()\`. Saves 70%+ of tokens and preserves the sub-agent's full context.
 
 Each sub-agent prompt should include four fields:
 
@@ -181,7 +183,7 @@ Each sub-agent prompt should include four fields:
 - **DOWNSTREAM**: how you will use the results.
 - **REQUEST**: what to find, what format to return, what to skip.
 
-**Background tasks.** Collect with \`background_output(task_id="...")\` once they complete. Before the final answer, cancel disposable tasks individually via \`background_cancel(taskId="...")\`. Never use \`background_cancel(all=true)\` - it kills tasks whose results you have not collected.
+**Background tasks.** Collect with background task IDs (\`bg_...\`) via \`background_output(task_id="bg_...")\` once they complete. Use continuation IDs (\`ses_...\`) only for \`task(task_id="ses_...")\` follow-ups. Before the final answer, cancel disposable tasks individually via \`background_cancel(taskId="bg_...")\`. Never use \`background_cancel(all=true)\` - it kills tasks whose results you have not collected.
 
 **\`skill\`** loads specialized instruction packs. Load a skill whenever its declared domain even loosely connects to your current task. Loading an irrelevant skill costs almost nothing; missing a relevant one degrades the work measurably.
 

src/agents/hephaestus/gpt.ts

@@ -201,7 +201,7 @@ task(subagent_type="librarian", run_in_background=true, load_skills=[], descript
 - Parallelize independent file reads - don't read files one at a time
 - NEVER use \`run_in_background=false\` for explore/librarian
 - Continue only with non-overlapping work after launching background agents
-- Collect results with \`background_output(task_id="...")\` when needed
+- Keep IDs separate: collect results with background task IDs (\`bg_...\`) via \`background_output(task_id="bg_...")\`; continue follow-up sessions with continuation IDs (\`ses_...\`) via \`task(task_id="ses_...")\`
 - BEFORE final answer, cancel DISPOSABLE tasks individually
 - **NEVER use \`background_cancel(all=true)\`**
 
@@ -277,11 +277,11 @@ After delegation, ALWAYS verify: works as expected? follows codebase pattern? MU
 
 ### Session Continuity
 
-Every \`task()\` output includes a task_id. **USE IT for follow-ups.**
+Every \`task()\` output includes a continuation ID (\`ses_...\`). **USE IT for follow-ups.**
 
-- **Task failed/incomplete** - \`task_id="{id}", prompt="Fix: {error}"\`
+- **Task failed/incomplete** - \`task(task_id="ses_...", prompt="Fix: {error}")\`
-- **Follow-up on result** - \`task_id="{id}", prompt="Also: {question}"\`
+- **Follow-up on result** - \`task(task_id="ses_...", prompt="Also: {question}")\`
-- **Verification failed** - \`task_id="{id}", prompt="Failed: {error}. Fix."\`
+- **Verification failed** - \`task(task_id="ses_...", prompt="Failed: {error}. Fix.")\`
 
 ${
   oracleSection

src/agents/sisyphus-id-contract.test.ts

@@ -0,0 +1,32 @@
+/// <reference types="bun-types" />
+
+import { describe, expect, test } from "bun:test"
+import { buildClaudeOpus47SisyphusPrompt } from "./sisyphus/claude-opus-4-7"
+import { buildDefaultSisyphusPrompt } from "./sisyphus/default"
+import { buildGpt54SisyphusPrompt } from "./sisyphus/gpt-5-4"
+import { buildGpt55SisyphusPrompt } from "./sisyphus/gpt-5-5"
+import { buildKimiK26SisyphusPrompt } from "./sisyphus/kimi-k2-6"
+
+describe("Sisyphus background task ID guidance", () => {
+  const promptBuilders = [
+    ["claude-opus-4-7", buildClaudeOpus47SisyphusPrompt],
+    ["default", buildDefaultSisyphusPrompt],
+    ["gpt-5.4", buildGpt54SisyphusPrompt],
+    ["gpt-5.5", buildGpt55SisyphusPrompt],
+    ["kimi-k2.6", buildKimiK26SisyphusPrompt],
+  ] as const
+
+  for (const [name, buildPrompt] of promptBuilders) {
+    test(`#given ${name} prompt #when describing background tasks #then bg ids and session ids are disambiguated`, () => {
+      // given, when
+      const prompt = buildPrompt(name, [])
+
+      // then
+      expect(prompt).toContain("background task IDs (`bg_...`)")
+      expect(prompt).toContain("continuation session IDs (`ses_...`)")
+      expect(prompt).toContain("background_output(task_id=\"bg_...\")")
+      expect(prompt).toContain("task(task_id=\"ses_...\")")
+      expect(prompt).not.toContain("receive task_ids")
+    })
+  }
+})

src/agents/sisyphus.ts

@@ -266,14 +266,15 @@ result = task(..., run_in_background=false)  // Never wait synchronously for exp
 \`\`\`
 
 ### Background Result Collection:
-1. Launch parallel agents \u2192 receive task_ids
+1. Launch parallel agents \u2192 receive background task IDs (\`bg_...\`) for results and continuation session IDs (\`ses_...\`) for follow-ups
 2. Continue only with non-overlapping work
    - If you have DIFFERENT independent work \u2192 do it now
    - Otherwise \u2192 **END YOUR RESPONSE.**
 3. **STOP. END YOUR RESPONSE.** The system will send \`<system-reminder>\` when tasks complete.
-4. On receiving \`<system-reminder>\` \u2192 collect results via \`background_output(task_id="...")\`
+4. On receiving \`<system-reminder>\` \u2192 collect results via \`background_output(task_id="bg_...")\`
 5. **NEVER call \`background_output\` before receiving \`<system-reminder>\`.** This is a BLOCKING anti-pattern.
 6. Cleanup: Cancel disposable tasks individually via \`background_cancel(taskId="...")\`
+7. Use \`task(task_id="ses_...")\` only to continue the same sub-agent session
 
 ${buildAntiDuplicationSection()}
 
@@ -328,15 +329,17 @@ AFTER THE WORK YOU DELEGATED SEEMS DONE, ALWAYS VERIFY THE RESULTS AS FOLLOWING:
 
 ### Session Continuity (MANDATORY)
 
-Every \`task()\` output includes a task_id. **USE IT.**
+Every \`task()\` output exposes a continuation session ID (\`ses_...\`). Pass it to \`task(task_id="ses_...")\` for follow-ups. **USE IT.**
 
 **ALWAYS continue when:**
-- Task failed/incomplete → \`task_id=\"{task_id}\", prompt=\"Fix: {specific error}\"\`
+- Task failed/incomplete → \`task(task_id="ses_...", prompt="Fix: {specific error}")\`
-- Follow-up question on result → \`task_id=\"{task_id}\", prompt=\"Also: {question}\"\`
+- Follow-up question on result → \`task(task_id="ses_...", prompt="Also: {question}")\`
-- Multi-turn with same agent → \`task_id=\"{task_id}\"\` - NEVER start fresh
+- Multi-turn with same agent → \`task(task_id="ses_...")\` - NEVER start fresh
-- Verification failed → \`task_id=\"{task_id}\", prompt=\"Failed verification: {error}. Fix.\"\`
+- Verification failed → \`task(task_id="ses_...", prompt="Failed verification: {error}. Fix.")\`
 
-**Why task_id is CRITICAL:**
+**Keep IDs separate:** background task IDs (\`bg_...\`) are for \`background_output(task_id="bg_...")\`; continuation session IDs (\`ses_...\`) are for \`task(task_id="ses_...")\`.
+
+**Why continuation is CRITICAL:**
 - Subagent has FULL conversation context preserved
 - No repeated file reads, exploration, or setup
 - Saves 70%+ tokens on follow-ups
@@ -350,7 +353,7 @@ task(category="quick", load_skills=[], run_in_background=false, description="Fix
 task(task_id="ses_abc123", load_skills=[], run_in_background=false, description="Fix type error", prompt="Fix: Type error on line 42")
 \`\`\`
 
-**After EVERY delegation, STORE the task_id for potential continuation.**
+**After EVERY delegation, STORE the \`ses_...\` continuation ID for potential continuation.**
 
 ### Code Changes:
 - Match existing patterns (if codebase is disciplined)

src/agents/sisyphus/claude-opus-4-7.ts

@@ -301,11 +301,12 @@ Fire similar parallel calls for error patterns (explore), JWT security best prac
 
 ### Background Result Collection:
 
-1. Launch parallel agents → receive task_ids
+1. Launch parallel agents → receive background task IDs (\`bg_...\`) for results and continuation session IDs (\`ses_...\`) for follow-ups.
 2. Continue ONLY with non-overlapping work. If none → END YOUR RESPONSE.
 3. System sends \`<system-reminder>\` when tasks complete.
-4. Collect via \`background_output(task_id="...")\` ONLY after \`<system-reminder>\`.
+4. Collect via \`background_output(task_id="bg_...")\` ONLY after \`<system-reminder>\`.
 5. Cancel disposable tasks INDIVIDUALLY via \`background_cancel(taskId="...")\`. NEVER \`background_cancel(all=true)\`.
+6. Use \`task(task_id="ses_...")\` only to continue the same sub-agent session.
 
 ${buildAntiDuplicationSection()}
 
@@ -347,9 +348,10 @@ After delegation: VERIFY against MUST DO/MUST NOT DO + existing patterns. Vague
 
 ### Session Continuity (apply to ALL follow-ups)
 
-Every \`task()\` returns \`task_id\`. **REUSE IT.**
+Every \`task()\` output exposes a continuation session ID (\`ses_...\`). Pass it to \`task(task_id="ses_...")\`. **REUSE IT.**
 
-Use \`task_id\` for: failed/incomplete work, follow-up questions, multi-turn refinement, verification failures.
+Use \`task(task_id="ses_...")\` for: failed/incomplete work, follow-up questions, multi-turn refinement, verification failures.
+Keep IDs separate: background task IDs (\`bg_...\`) are for \`background_output(task_id="bg_...")\`; continuation session IDs (\`ses_...\`) are for \`task(task_id="ses_...")\`.
 
 \`\`\`typescript
 // WRONG: starting fresh loses everything

src/agents/sisyphus/default.ts

@@ -327,14 +327,15 @@ result = task(..., run_in_background=false)  // Never wait synchronously for exp
 \`\`\`
 
 ### Background Result Collection:
-1. Launch parallel agents → receive task_ids
+1. Launch parallel agents → receive background task IDs (\`bg_...\`) for results and continuation session IDs (\`ses_...\`) for follow-ups
 2. Continue only with non-overlapping work
    - If you have DIFFERENT independent work → do it now
    - Otherwise → **END YOUR RESPONSE.**
 3. **STOP. END YOUR RESPONSE.** The system will send \`<system-reminder>\` when tasks complete.
-4. On receiving \`<system-reminder>\` → collect results via \`background_output(task_id="...")\`
+4. On receiving \`<system-reminder>\` → collect results via \`background_output(task_id="bg_...")\`
 5. **NEVER call \`background_output\` before receiving \`<system-reminder>\`.** This is a BLOCKING anti-pattern.
 6. Cleanup: Cancel disposable tasks individually via \`background_cancel(taskId="...")\`
+7. Use \`task(task_id="ses_...")\` only to continue the same sub-agent session
 
 ${buildAntiDuplicationSection()}
 
@@ -389,15 +390,17 @@ AFTER THE WORK YOU DELEGATED SEEMS DONE, ALWAYS VERIFY THE RESULTS AS FOLLOWING:
 
 ### Session Continuity (MANDATORY)
 
-Every \`task()\` output includes a task_id. **USE IT.**
+Every \`task()\` output exposes a continuation session ID (\`ses_...\`). Pass it to \`task(task_id="ses_...")\` for follow-ups. **USE IT.**
 
 **ALWAYS continue when:**
-- Task failed/incomplete → \`task_id="{task_id}", prompt="Fix: {specific error}"\`
+- Task failed/incomplete → \`task(task_id="ses_...", prompt="Fix: {specific error}")\`
-- Follow-up question on result → \`task_id="{task_id}", prompt="Also: {question}"\`
+- Follow-up question on result → \`task(task_id="ses_...", prompt="Also: {question}")\`
-- Multi-turn with same agent → \`task_id="{task_id}"\` - NEVER start fresh
+- Multi-turn with same agent → \`task(task_id="ses_...")\` - NEVER start fresh
-- Verification failed → \`task_id="{task_id}", prompt="Failed verification: {error}. Fix."\`
+- Verification failed → \`task(task_id="ses_...", prompt="Failed verification: {error}. Fix.")\`
 
-**Why task_id is CRITICAL:**
+**Keep IDs separate:** background task IDs (\`bg_...\`) are for \`background_output(task_id="bg_...")\`; continuation session IDs (\`ses_...\`) are for \`task(task_id="ses_...")\`.
+
+**Why continuation is CRITICAL:**
 - Subagent has FULL conversation context preserved
 - No repeated file reads, exploration, or setup
 - Saves 70%+ tokens on follow-ups
@@ -411,7 +414,7 @@ task(category="quick", load_skills=[], run_in_background=false, description="Fix
 task(task_id="ses_abc123", load_skills=[], run_in_background=false, description="Fix type error", prompt="Fix: Type error on line 42")
 \`\`\`
 
-**After EVERY delegation, STORE the task_id for potential continuation.**
+**After EVERY delegation, STORE the \`ses_...\` continuation ID for potential continuation.**
 
 ### Code Changes:
 - Match existing patterns (if codebase is disciplined)

src/agents/sisyphus/gpt-5-4.ts

@@ -263,14 +263,15 @@ Each agent prompt should include:
 - [REQUEST]: What to find, what format, what to skip
 
 Background result collection:
-1. Launch parallel agents → receive task_ids
+1. Launch parallel agents → receive background task IDs (\`bg_...\`) for results and continuation session IDs (\`ses_...\`) for follow-ups
 2. Continue only with non-overlapping work
    - If you have DIFFERENT independent work → do it now
    - Otherwise → **END YOUR RESPONSE.**
 3. **STOP. END YOUR RESPONSE.** The system will send \`<system-reminder>\` when tasks complete.
-4. On receiving \`<system-reminder>\` → collect results via \`background_output(task_id="...")\`
+4. On receiving \`<system-reminder>\` → collect results via \`background_output(task_id="bg_...")\`
 5. **NEVER call \`background_output\` before receiving \`<system-reminder>\`.** This is a BLOCKING anti-pattern.
 6. Cancel disposable tasks individually via \`background_cancel(taskId="...")\`
+7. Use \`task(task_id="ses_...")\` only to continue the same sub-agent session
 
 ${buildAntiDuplicationSection()}
 
@@ -387,10 +388,12 @@ Post-delegation: delegation never substitutes for verification. Always run \`<ve
 
 ### Session continuity
 
-Every \`task()\` returns a task_id. Use it for all follow-ups:
+Every \`task()\` output exposes a continuation session ID (\`ses_...\`). Pass it to \`task(task_id="ses_...")\` for all follow-ups:
-- Failed/incomplete → \`task_id="{id}", prompt="Fix: {specific error}"\`
+- Failed/incomplete → \`task(task_id="ses_...", prompt="Fix: {specific error}")\`
-- Follow-up → \`task_id="{id}", prompt="Also: {question}"\`
+- Follow-up → \`task(task_id="ses_...", prompt="Also: {question}")\`
-- Multi-turn → always \`task_id\`, never start fresh
+- Multi-turn → always \`task(task_id="ses_...")\`, never start fresh
+
+Keep IDs separate: background task IDs (\`bg_...\`) are for \`background_output(task_id="bg_...")\`; continuation session IDs (\`ses_...\`) are for \`task(task_id="ses_...")\`.
 
 This preserves full context, avoids repeated exploration, saves 70%+ tokens.
 

src/agents/sisyphus/gpt-5-5.ts

@@ -218,11 +218,13 @@ After a delegation completes, verification is not optional. Read every file the
 
 ### Session continuity
 
-Every \`task()\` returns a \`task_id\`. Reuse it for every follow-up interaction with the same sub-agent:
+Every \`task()\` output exposes a continuation session ID (\`ses_...\`). Pass it to \`task(task_id="ses_...")\` for every follow-up with the same sub-agent:
 
-- Failed or incomplete work: \`task(task_id="{id}", prompt="Fix: {specific error}")\`
+- Failed or incomplete work: \`task(task_id="ses_...", prompt="Fix: {specific error}")\`
-- Follow-up question on a result: \`task(task_id="{id}", prompt="Also: {question}")\`
+- Follow-up question on a result: \`task(task_id="ses_...", prompt="Also: {question}")\`
-- Multi-turn refinement: always \`task_id\`, never a fresh session.
+- Multi-turn refinement: always \`task(task_id="ses_...")\`, never a fresh session.
+
+Keep IDs separate: background task IDs (\`bg_...\`) are for \`background_output(task_id="bg_...")\`; continuation session IDs (\`ses_...\`) are for \`task(task_id="ses_...")\`.
 
 Starting fresh on a follow-up throws away the sub-agent's full context. Session continuity typically saves 70% of the tokens a fresh session would burn.
 
@@ -235,7 +237,7 @@ Exploration is cheap; assumption is expensive. Before implementation on anything
 
 Each exploration prompt should include four fields: **CONTEXT** (what task, which modules), **GOAL** (what decision the results will unblock), **DOWNSTREAM** (how you will use the results), **REQUEST** (what to find, what format, what to skip).
 
-After firing exploration agents, do not manually perform the same search yourself. That is duplicate work and wastes your context window. Continue only with non-overlapping preparation: setting up files, reading known-path files, drafting questions. If no non-overlapping work exists, end your response and wait for the completion notification; do not poll \`background_output\` on a running task.
+After firing exploration agents, keep the returned background task IDs (\`bg_...\`) for result collection and continuation session IDs (\`ses_...\`) for follow-ups. Continue only with non-overlapping preparation: setting up files, reading known-path files, drafting questions. If no non-overlapping work exists, end your response and wait for the completion notification; then use \`background_output(task_id="bg_...")\`, not \`task(task_id="ses_...")\`, to collect results.
 
 Stop searching when you have enough context to proceed confidently, when the same information keeps appearing across sources, when two iterations yield no new useful data, or when you found a direct answer.
 

src/agents/sisyphus/kimi-k2-6.ts

@@ -307,14 +307,15 @@ Each agent prompt should include:
 - [REQUEST]: What to find, what format, what to skip
 
 Background result collection:
-1. Launch parallel agents → receive task_ids
+1. Launch parallel agents → receive background task IDs (\`bg_...\`) for results and continuation session IDs (\`ses_...\`) for follow-ups
 2. Continue only with non-overlapping work
    - If you have DIFFERENT independent work → do it now
    - Otherwise → **END YOUR RESPONSE.**
 3. **STOP. END YOUR RESPONSE.** The system will send \`<system-reminder>\` when tasks complete.
-4. On receiving \`<system-reminder>\` → collect results via \`background_output(task_id="...")\`
+4. On receiving \`<system-reminder>\` → collect results via \`background_output(task_id="bg_...")\`
 5. **NEVER call \`background_output\` before receiving \`<system-reminder>\`.** This is a BLOCKING anti-pattern.
 6. Cancel disposable tasks individually via \`background_cancel(taskId="...")\`
+7. Use \`task(task_id="ses_...")\` only to continue the same sub-agent session
 
 ${buildAntiDuplicationSection()}
 
@@ -462,10 +463,12 @@ Post-delegation: delegation never substitutes for verification. Always run \`<ve
 
 ### Session continuity
 
-Every \`task()\` returns a session_id. Use it for all follow-ups:
+Every \`task()\` output exposes a continuation session ID (\`ses_...\`). Pass it to \`task(task_id="ses_...")\` for all follow-ups:
-- Failed/incomplete → \`session_id="{id}", prompt="Fix: {specific error}"\`
+- Failed/incomplete → \`task(task_id="ses_...", prompt="Fix: {specific error}")\`
-- Follow-up → \`session_id="{id}", prompt="Also: {question}"\`
+- Follow-up → \`task(task_id="ses_...", prompt="Also: {question}")\`
-- Multi-turn → always \`session_id\`, never start fresh
+- Multi-turn → always \`task(task_id="ses_...")\`, never start fresh
+
+Keep IDs separate: background task IDs (\`bg_...\`) are for \`background_output(task_id="bg_...")\`; continuation session IDs (\`ses_...\`) are for \`task(task_id="ses_...")\`.
 
 This preserves full context, avoids repeated exploration, saves 70%+ tokens.
 

src/features/builtin-commands/templates/init-deep.ts

@@ -135,7 +135,7 @@ LspFindReferences(filePath="...", line=X, character=Y)
 
 \`\`\`
 // After main session analysis done, collect all task results
-for each task_id: background_output(task_id="...")
+for each background task ID (\`bg_...\`): background_output(task_id="bg_...")
 \`\`\`
 
 **Merge: bash + LSP + existing + explore findings. Mark "discovery" as completed.**

src/features/builtin-skills/skills/review-work.ts

@@ -481,7 +481,7 @@ OUTPUT FORMAT:
 
 After launching all 5 agents in one turn, **end your response**. Wait for system notifications as each agent completes.
 
-As each completes, collect via \`background_output(task_id="...")\`. Store each verdict:
+As each completes, collect via \`background_output(task_id="bg_...")\`. Store each verdict:
 
 | Agent | Verdict | Notes |
 |-------|---------|-------|

src/hooks/keyword-detector/ultrawork/default.ts

@@ -115,7 +115,7 @@ task(subagent_type="plan", load_skills=[], run_in_background=false, prompt="<gat
 
 ### SESSION CONTINUITY WITH PLAN AGENT (CRITICAL)
 
-**Plan agent returns a task_id. USE IT for follow-up interactions.**
+**Plan agent output includes a continuation ID (\`ses_...\`). USE IT for follow-up interactions via \`task(task_id="ses_...", ...)\`.**
 
 | Scenario | Action |
 |----------|--------|

src/hooks/keyword-detector/ultrawork/gemini.ts

@@ -161,7 +161,7 @@ task(subagent_type="plan", load_skills=[], run_in_background=false, prompt="<gat
 
 ### SESSION CONTINUITY WITH PLAN AGENT (CRITICAL)
 
-**Plan agent returns a task_id. USE IT for follow-up interactions.**
+**Plan agent output includes a continuation ID (\`ses_...\`). USE IT for follow-up interactions via \`task(task_id="ses_...", ...)\`.**
 
 | Scenario | Action |
 |----------|--------|

src/tools/background-task/constants.ts

@@ -1,4 +1,4 @@
-export const BACKGROUND_TASK_DESCRIPTION = `Run agent task in background. Returns task_id immediately; notifies on completion.
+export const BACKGROUND_TASK_DESCRIPTION = `Run agent task in background. Returns a background task ID (\`bg_...\`) immediately; notifies on completion.
 
 Use \`background_output\` to get results. Prompts MUST be in English.`
 

src/tools/background-task/create-background-output.metadata.test.ts

@@ -4,7 +4,9 @@ import type { ToolContext } from "@opencode-ai/plugin/tool"
 import { describe, expect, test } from "bun:test"
 import type { BackgroundTask } from "../../features/background-agent"
 import { clearPendingStore, consumeToolMetadata } from "../../features/tool-metadata-store"
+import { unsafeTestValue } from "../../../test-support/unsafe-test-value"
 import type { BackgroundOutputClient, BackgroundOutputManager } from "./clients"
+import { BACKGROUND_TASK_DESCRIPTION } from "./constants"
 import { createBackgroundOutput } from "./create-background-output"
 
 const projectDir = "/Users/yeongyu/local-workspaces/oh-my-opencode"
@@ -14,6 +16,38 @@ type ToolContextWithCallID = ToolContext & {
 }
 
 describe("createBackgroundOutput metadata", () => {
+  test("describes background task launch output as a bg id", () => {
+    // #given, #when
+    const description = BACKGROUND_TASK_DESCRIPTION
+
+    // #then
+    expect(description).toContain("background task ID")
+    expect(description).toContain("bg_")
+    expect(description).not.toContain("Returns task_id")
+  })
+
+  test("describes task_id as a background task id instead of a session id", () => {
+    // #given
+    const manager: BackgroundOutputManager = {
+      getTask: () => undefined,
+    }
+    const client: BackgroundOutputClient = {
+      session: {
+        messages: async () => ({ data: [] }),
+      },
+    }
+    const tool = createBackgroundOutput(manager, client)
+
+    // #when
+    const taskIdArg = unsafeTestValue<{ description?: string }>(tool.args.task_id)
+
+    // #then
+    expect(taskIdArg.description).toContain("background task ID")
+    expect(taskIdArg.description).toContain("bg_")
+    expect(taskIdArg.description).toContain("not a session ID")
+    expect(taskIdArg.description).toContain("ses_")
+  })
+
   test("omits sessionId metadata when task session is not yet assigned", async () => {
     // #given
     clearPendingStore()

src/tools/background-task/create-background-output.ts

@@ -93,7 +93,9 @@ export function createBackgroundOutput(manager: BackgroundOutputManager, client:
   return tool({
     description: BACKGROUND_OUTPUT_DESCRIPTION,
     args: {
-      task_id: tool.schema.string().describe("Task ID to get output from"),
+      task_id: tool.schema
+        .string()
+        .describe("background task ID (`bg_...`) from launch/completion; not a session ID (`ses_...`)."),
       block: tool.schema
         .boolean()
         .optional()

src/tools/delegate-task/task-schema.test.ts

@@ -42,11 +42,25 @@ function createDelegateTask(...args: Parameters<typeof import("./tools").createD
 
 		//#then
 		expect(description).toContain("subagent_type: Use specific agent directly")
-		expect(description).toContain("task_id: Existing task to continue")
+		expect(description).toContain("task_id: Continuation session id")
 		expect(description).not.toContain("sisyphus")
 		expect(description).not.toContain("hephaestus")
 		expect(description).not.toContain("prometheus")
 	})
+
+	test("#given task schema #when describing async mode #then it names background task ids explicitly", () => {
+		//#given
+		const toolDefinition = createDelegateTask({ manager: {} as never, client: {} as never, directory: "/tmp/test" })
+
+		//#when
+		const runInBackgroundSchema = unsafeTestValue<{ description?: string }>(toolDefinition.args.run_in_background)
+
+		//#then
+		expect(runInBackgroundSchema.description).toContain("background task ID")
+		expect(runInBackgroundSchema.description).toContain("bg_")
+		expect(runInBackgroundSchema.description).toContain("background_output")
+		expect(runInBackgroundSchema.description).not.toContain("returns task_id")
+	})
 })
 
 export {}

src/tools/delegate-task/tool-description.test.ts

@@ -15,4 +15,18 @@ describe("createDelegateTaskPresentation", () => {
     expect(description).toContain("busy/retry/running")
     expect(description).toContain("not a total wall-clock limit")
   })
+
+  test("#given continuation usage #when description is rendered #then task_id is described as a session id", () => {
+    //#given
+    const presentation = createDelegateTaskPresentation({})
+
+    //#when
+    const description = presentation.description
+
+    //#then
+    expect(description).toContain("task_id: Continuation session id")
+    expect(description).toContain("ses_")
+    expect(description).toContain("not the background task id")
+    expect(description).toContain("bg_")
+  })
 })

src/tools/delegate-task/tool-description.ts

@@ -66,15 +66,15 @@ export function createDelegateTaskPresentation(options: DelegateTaskToolOptions)
     Available categories:
   ${categoryList}
   - subagent_type: Use specific agent directly (explore, librarian, oracle, metis, momus)
-  - run_in_background: REQUIRED. true=async (returns task_id), false=sync (waits). Use background=true ONLY for parallel exploration with 5+ independent queries.
+  - run_in_background: REQUIRED. true=async (returns a background task ID like \`bg_...\` for \`background_output\`), false=sync (waits). Use background=true ONLY for parallel exploration with 5+ independent queries.
     Sync waits use a 30-minute inactivity window: OpenCode busy/retry/running status resets the window, so this is not a total wall-clock limit.
-  - task_id: Existing task to continue (from previous task output). Continues the same subagent session with FULL CONTEXT PRESERVED.
+  - task_id: Continuation session id (\`ses_...\`) from task metadata. Continues the same subagent session with FULL CONTEXT PRESERVED; not the background task id (\`bg_...\`).
   - command: The command that triggered this task (optional, for slash command tracking).
   
   **WHEN TO USE task_id:**
-  - Task failed/incomplete → task_id with "fix: [specific issue]"
+  - Task failed/incomplete → \`task(task_id="ses_...", prompt="fix: [specific issue]")\`
-  - Need follow-up on previous result → task_id with additional question
+  - Need follow-up on previous result → \`task(task_id="ses_...", prompt="Also: [question]")\`
-  - Multi-turn conversation with same agent → always task_id instead of new task
+  - Multi-turn conversation with same agent → always \`task(task_id="ses_...")\` instead of new task
   
   Prompts MUST be in English.`
 

src/tools/delegate-task/tools.ts

@@ -24,10 +24,15 @@ const delegateTaskArgsSchema = {
   load_skills: tool.schema.array(tool.schema.string()).describe("Skill names to inject. REQUIRED - pass [] if no skills needed."),
   description: tool.schema.string().optional().describe("Short task description (3-5 words). Auto-generated from prompt if omitted."),
   prompt: tool.schema.string().describe("Full detailed prompt for the agent"),
-  run_in_background: tool.schema.boolean().describe("REQUIRED. true=async (returns task_id), false=sync (waits). Use false for task delegation, true ONLY for parallel exploration."),
+  run_in_background: tool.schema
+    .boolean()
+    .describe("REQUIRED. true=async (returns background task ID `bg_...` for background_output), false=sync (waits). Use false for task delegation, true ONLY for parallel exploration."),
   category: tool.schema.string().optional().describe("REQUIRED if subagent_type not provided. Do NOT provide both category and subagent_type."),
   subagent_type: tool.schema.string().optional().describe("REQUIRED if category not provided. Do NOT provide both category and subagent_type."),
-  task_id: tool.schema.string().optional().describe("Existing task to continue. Canonical resume identifier."),
+  task_id: tool.schema
+    .string()
+    .optional()
+    .describe("Continuation session id (`ses_...`) from task metadata; not a background task id (`bg_...`)."),
   command: tool.schema.string().optional().describe("The command that triggered this task"),
 }