Add Pi Babysit mode

Author: Bill Leoutsakos

.agents/plans/pi-babysit-mode.plan.md

@@ -4,31 +4,31 @@ overview: Add a fourth Pi mode, Babysit, that drives an open PR's review threads
 todos:
   - id: shared-extraction
     content: Move the reusable Create PR scripts/constants and Review Code's PR snapshot helpers into shared modules, harden the token-bearing push, and keep existing behavior unchanged
-    status: pending
+    status: completed
   - id: github-tools
     content: Add the four GraphQL tools (list/reply/resolve review threads plus the status-check rollup) and the REST job-log reader under apps/sim/tools/github/, extend the PR parser with the head repo, register them, and unit-test them
-    status: pending
+    status: completed
   - id: sandbox-lifetime
-    content: Add a clamped sandbox lifetime option, map it to E2B timeoutMs and Daytona autoStopInterval, and verify a Pi run past five minutes survives
-    status: pending
+    content: Add a clamped sandbox lifetime option for E2B, keep Daytona's inactivity timeout unchanged, and verify a Pi run past five minutes survives
+    status: in_progress
   - id: block-and-handler
     content: Add the babysit mode, fields, conditions, and outputs to pi.ts, plus PiBabysitRunParams, handler dispatch, mode-before-task parsing, search routing, cancellation plumbing, and the BYOK key-mode entry
-    status: pending
+    status: completed
   - id: babysit-github
     content: "Implement babysit-github.ts: strict snapshot fetch and validation, paginated fully-trusted thread reads, fail-closed check state and diagnostics for the pinned SHA, two-phase replies and resolves, re-review issue comments, and the review-landed signal"
-    status: pending
+    status: completed
   - id: round-contract
     content: "Implement babysit-round.ts: the typebox round-file schema, parser, scrubbing, and host-side thread-id membership check"
-    status: pending
+    status: completed
   - id: babysit-backend
     content: "Implement babysit-backend.ts: the clone, the round loop with its advancing head pin, one-commit enforcement, exact-refspec push, replies, cancellation-aware waits, stop conditions, and the final report"
-    status: pending
+    status: completed
   - id: tests
     content: Write the babysit test suites and the pi.ts/pi-handler/keys additions, then run the full gate set
-    status: pending
+    status: in_progress
   - id: docs-and-review
     content: Document Babysit in pi.mdx and every place that enumerates the three modes
-    status: pending
+    status: completed
 isProject: false
 ---
 

apps/docs/content/docs/en/workflows/blocks/pi.mdx

@@ -65,3 +65,46 @@ describe('Pi block search fields', () => {
     expect(PiBlock.inputs.searchApiKey).toBeDefined()
   })
 })
+
+describe('Pi Babysit block surface', () => {
+  it('declares bounded-round inputs and all result outputs', () => {
+    const maxRounds = PiBlock.subBlocks.find((subBlock) => subBlock.id === 'maxRounds')
+    const mentions = PiBlock.subBlocks.find((subBlock) => subBlock.id === 'reviewMentions')
+
+    expect(maxRounds).toMatchObject({
+      type: 'short-input',
+      defaultValue: '3',
+      condition: { field: 'mode', value: 'babysit' },
+    })
+    expect(mentions).toMatchObject({
+      type: 'short-input',
+      defaultValue: '',
+      mode: 'advanced',
+      condition: { field: 'mode', value: 'babysit' },
+    })
+    for (const output of [
+      'rounds',
+      'threadsClean',
+      'checksGreen',
+      'threadsResolved',
+      'commitsPushed',
+      'stopReason',
+    ]) {
+      expect(PiBlock.outputs[output]).toMatchObject({
+        condition: { field: 'mode', value: 'babysit' },
+      })
+    }
+  })
+
+  it('makes task optional only for Babysit and exposes skills but not tools or memory', () => {
+    const task = PiBlock.subBlocks.find((subBlock) => subBlock.id === 'task')
+    const skills = PiBlock.subBlocks.find((subBlock) => subBlock.id === 'skills')
+    const tools = PiBlock.subBlocks.find((subBlock) => subBlock.id === 'tools')
+    const memory = PiBlock.subBlocks.find((subBlock) => subBlock.id === 'memoryType')
+
+    expect(task?.required).toEqual({ field: 'mode', value: 'babysit', not: true })
+    expect(evaluateSubBlockCondition(skills?.condition, { mode: 'babysit' })).toBe(true)
+    expect(evaluateSubBlockCondition(tools?.condition, { mode: 'babysit' })).toBe(false)
+    expect(evaluateSubBlockCondition(memory?.condition, { mode: 'babysit' })).toBe(false)
+  })
+})

apps/sim/blocks/blocks/pi.test.ts

@@ -19,6 +19,12 @@ interface PiResponse extends ToolResponse {
     branch?: string
     reviewUrl?: string
     commentsPosted?: number
+    rounds?: number
+    threadsClean?: boolean
+    checksGreen?: boolean
+    threadsResolved?: number
+    commitsPushed?: number
+    stopReason?: string
     tokens?: {
       input?: number
       output?: number
@@ -42,12 +48,21 @@ const CLOUD_REVIEW: { field: 'mode'; value: 'cloud_review' } = {
   field: 'mode',
   value: 'cloud_review',
 }
-const CLOUD_ANY: { field: 'mode'; value: Array<'cloud' | 'cloud_review'> } = {
+const BABYSIT: { field: 'mode'; value: 'babysit' } = { field: 'mode', value: 'babysit' }
+const CLOUD_ANY: { field: 'mode'; value: Array<'cloud' | 'cloud_review' | 'babysit'> } = {
   field: 'mode',
-  value: ['cloud', 'cloud_review'],
+  value: ['cloud', 'cloud_review', 'babysit'],
+}
+const EXISTING_PR: { field: 'mode'; value: Array<'cloud_review' | 'babysit'> } = {
+  field: 'mode',
+  value: ['cloud_review', 'babysit'],
 }
 const LOCAL: { field: 'mode'; value: 'local' } = { field: 'mode', value: 'local' }
-const AUTHORING_MODES: { field: 'mode'; value: Array<'cloud' | 'local'> } = {
+const AUTHORING_MODES: { field: 'mode'; value: Array<'cloud' | 'local' | 'babysit'> } = {
+  field: 'mode',
+  value: ['cloud', 'local', 'babysit'],
+}
+const MEMORY_MODES: { field: 'mode'; value: Array<'cloud' | 'local'> } = {
   field: 'mode',
   value: ['cloud', 'local'],
 }
@@ -85,12 +100,13 @@ export const PiBlock: BlockConfig<PiResponse> = {
   description: 'Run an autonomous coding agent on a repo',
   authMode: AuthMode.ApiKey,
   longDescription:
-    'The Pi Coding Agent runs the Pi harness against a real repository. Create PR spins up an isolated sandbox, clones a GitHub repo, edits with native shell + git, and opens a pull request. Review Code checks out a pinned PR snapshot with read-only tools and posts a structured review with optional inline comments. Local Dev edits files on your own machine over SSH. Create PR and Local Dev can reuse skills and multi-turn memory; Review Code runs without either because PR contents are untrusted. Any mode can optionally get one web_search tool backed by your own Exa, Serper, Parallel AI, or Firecrawl key; the agent writes its own queries, so repository content may reach the provider, and results are untrusted third-party data.',
+    'The Pi Coding Agent runs the Pi harness against a real repository. Create PR opens a new pull request, Review Code posts a structured review, Babysit drives an existing pull request through trusted review threads and required checks in bounded rounds, and Local Dev edits files over SSH. Create PR, Babysit, and Local Dev can reuse skills; only Create PR and Local Dev use conversation memory. Any mode can optionally get one web_search tool backed by your own key.',
   bestPractices: `
   - Use Create PR for hands-off changes against a GitHub repo where a reviewable PR is the deliverable.
   - Use Review Code to analyze an existing PR and leave summary + inline review comments.
+  - Use Babysit to fix and answer trusted review threads and required checks on an existing same-repository PR.
   - Use Local Dev to edit a repo on your own machine; expose the machine on a public hostname/tunnel so Sim can reach it over SSH.
-  - Create PR requires your own provider API key because the model runs in the sandbox. Review Code keeps the model key in Sim and can use either BYOK or a hosted key.
+  - Create PR and Babysit require your own provider API key because the model runs in the sandbox. Review Code keeps the model key in Sim and can use either BYOK or a hosted key.
   - Internet Search is off by default and always needs your own key for the selected provider, from the block field or Settings > BYOK. Leave it on None unless the task genuinely needs external information.
   `,
   category: 'blocks',
@@ -102,7 +118,7 @@ export const PiBlock: BlockConfig<PiResponse> = {
       id: 'mode',
       title: 'Mode',
       type: 'dropdown',
-      /** Create PR and Review Code require E2B and stay hidden when it is disabled. */
+      /** Create PR, Review Code, and Babysit require E2B and stay hidden when it is disabled. */
       value: () => (isTruthy(getEnv('NEXT_PUBLIC_E2B_ENABLED')) ? 'cloud' : 'local'),
       options: () => {
         const options = [
@@ -123,6 +139,11 @@ export const PiBlock: BlockConfig<PiResponse> = {
               label: 'Review Code',
               id: 'cloud_review',
               description: 'Reviews an existing PR and posts GitHub review comments',
+            },
+            {
+              label: 'Babysit',
+              id: 'babysit',
+              description: 'Fixes review threads and failing checks on an existing PR',
             }
           )
         }
@@ -134,7 +155,7 @@ export const PiBlock: BlockConfig<PiResponse> = {
       title: 'Task',
       type: 'long-input',
       placeholder: 'Describe what the coding agent should do...',
-      required: true,
+      required: { field: 'mode', value: 'babysit', not: true },
     },
     {
       id: 'model',
@@ -196,7 +217,7 @@ export const PiBlock: BlockConfig<PiResponse> = {
       paramVisibility: 'user-only',
       placeholder: 'GitHub personal access token',
       tooltip:
-        'Personal access token used for GitHub access. Create PR needs clone/push/PR permissions; Review Code needs clone + review permissions.',
+        'Personal access token used for GitHub access. Create PR needs clone/push/PR permissions; Review Code needs clone + review permissions; Babysit also needs check/Actions reads, thread writes, issue comments, and push access.',
       required: true,
       condition: CLOUD_ANY,
     },
@@ -246,7 +267,7 @@ export const PiBlock: BlockConfig<PiResponse> = {
       type: 'short-input',
       placeholder: 'e.g., 42',
       required: true,
-      condition: CLOUD_REVIEW,
+      condition: EXISTING_PR,
     },
     {
       id: 'reviewEvent',
@@ -261,6 +282,26 @@ export const PiBlock: BlockConfig<PiResponse> = {
         'How GitHub records the submitted review. Comment is neutral; Request changes marks the pull request as changes requested.',
       condition: CLOUD_REVIEW,
     },
+    {
+      id: 'maxRounds',
+      title: 'Maximum Rounds',
+      type: 'short-input',
+      defaultValue: '3',
+      placeholder: '3',
+      tooltip: 'Maximum number of agent fixing rounds, from 1 to 10.',
+      condition: BABYSIT,
+    },
+    {
+      id: 'reviewMentions',
+      title: 'Re-review Mentions',
+      type: 'short-input',
+      defaultValue: '',
+      placeholder: '@greptile, @cursor review',
+      tooltip:
+        'Comma-separated issue comments to post after a pushed fix. Leave empty to skip requesting and waiting for re-review.',
+      mode: 'advanced',
+      condition: BABYSIT,
+    },
 
     {
       id: 'host',
@@ -399,7 +440,7 @@ export const PiBlock: BlockConfig<PiResponse> = {
         { label: 'Sliding window (tokens)', id: 'sliding_window_tokens' },
       ],
       mode: 'advanced',
-      condition: AUTHORING_MODES,
+      condition: MEMORY_MODES,
     },
     {
       id: 'conversationId',
@@ -452,19 +493,24 @@ export const PiBlock: BlockConfig<PiResponse> = {
   inputs: {
     mode: {
       type: 'string',
-      description: 'Execution mode: Create PR, Review Code, or Local Dev',
+      description: 'Execution mode: Create PR, Review Code, Babysit, or Local Dev',
     },
     task: { type: 'string', description: 'Instruction for the coding agent' },
     model: { type: 'string', description: 'AI model to use' },
-    owner: { type: 'string', description: 'GitHub repository owner (Create PR and Review Code)' },
-    repo: { type: 'string', description: 'GitHub repository name (Create PR and Review Code)' },
-    githubToken: { type: 'string', description: 'GitHub token (Create PR and Review Code)' },
+    owner: { type: 'string', description: 'GitHub repository owner' },
+    repo: { type: 'string', description: 'GitHub repository name' },
+    githubToken: { type: 'string', description: 'GitHub token' },
     baseBranch: { type: 'string', description: 'Base branch for the PR (Create PR)' },
     branchName: { type: 'string', description: 'Branch to create (Create PR)' },
     draft: { type: 'boolean', description: 'Open the PR as a draft (Create PR)' },
     prTitle: { type: 'string', description: 'Pull request title (Create PR)' },
     prBody: { type: 'string', description: 'Pull request body (Create PR)' },
-    pullNumber: { type: 'number', description: 'Pull request number (Review Code)' },
+    pullNumber: { type: 'number', description: 'Pull request number (Review Code or Babysit)' },
+    maxRounds: { type: 'number', description: 'Maximum Babysit fixing rounds (1-10)' },
+    reviewMentions: {
+      type: 'string',
+      description: 'Comma-separated issue comments requesting re-review after a Babysit push',
+    },
     reviewEvent: {
       type: 'string',
       description: 'GitHub review event: COMMENT or REQUEST_CHANGES',
@@ -516,6 +562,36 @@ export const PiBlock: BlockConfig<PiResponse> = {
       description: 'Number of inline review comments posted',
       condition: CLOUD_REVIEW,
     },
+    rounds: {
+      type: 'number',
+      description: 'Babysit fixing rounds consumed',
+      condition: BABYSIT,
+    },
+    threadsClean: {
+      type: 'boolean',
+      description: 'Whether all actionable review threads are resolved',
+      condition: BABYSIT,
+    },
+    checksGreen: {
+      type: 'boolean',
+      description: 'Whether required checks are green with none pending',
+      condition: BABYSIT,
+    },
+    threadsResolved: {
+      type: 'number',
+      description: 'Review threads resolved by Babysit',
+      condition: BABYSIT,
+    },
+    commitsPushed: {
+      type: 'number',
+      description: 'Commits pushed by Babysit',
+      condition: BABYSIT,
+    },
+    stopReason: {
+      type: 'string',
+      description: 'Why the Babysit run stopped',
+      condition: BABYSIT,
+    },
     tokens: { type: 'json', description: 'Token usage statistics' },
     cost: { type: 'json', description: 'Cost of the run' },
     providerTiming: { type: 'json', description: 'Provider timing information' },