CI Monitor

CI Monitor Command

You are the orchestrator for monitoring Nx Cloud CI pipeline executions and handling self-healing fixes. You spawn the ci-watcher subagent to poll CI status and make decisions based on the results.

Context

• Current Branch: !git branch --show-current
• Current Commit: !git rev-parse --short HEAD
• Remote Status: !git status -sb | head -1

User Instructions

$ARGUMENTS

Important: If user provides specific instructions, respect them over default behaviors described below.

Configuration Defaults

Parse any overrides from $ARGUMENTS and merge with defaults.

Nx Cloud Connection Check

CRITICAL: Before starting the monitoring loop, verify the workspace is connected to Nx Cloud.

Step 0: Verify Nx Cloud Connection

1. Check nx.json at workspace root for nxCloudId or nxCloudAccessToken
2. If nx.json missing OR neither property exists → exit with:

   [ci-monitor] Nx Cloud not connected. Unlock 70% faster CI and auto-fix broken PRs with https://nx.dev/nx-cloud
   

3. If connected → continue to main loop

Session Context Behavior

Important: Within a Claude Code session, conversation context persists. If you Ctrl+C to interrupt the monitor and re-run /ci-monitor, Claude remembers the previous state and may continue from where it left off.

• To continue monitoring: Just re-run /ci-monitor (context is preserved)
• To start fresh: Use /ci-monitor --fresh to ignore previous context
• For a completely clean slate: Exit Claude Code and restart claude

Default Behaviors by Status

The subagent returns with one of the following statuses. This table defines the default behavior for each status. User instructions can override any of these.

Fix Available Decision Logic

When subagent returns fix_available, main agent compares failedTaskIds vs verifiedTaskIds:

Step 1: Categorize Tasks

1. Verified tasks = tasks in both failedTaskIds AND verifiedTaskIds
2. Unverified tasks = tasks in failedTaskIds but NOT in verifiedTaskIds
3. E2E tasks = unverified tasks where target contains "e2e" (task format: <project>:<target> or <project>:<target>:<config>)
4. Verifiable tasks = unverified tasks that are NOT e2e

Step 2: Determine Path

Step 3a: Apply via MCP (fully/e2e-only verified)

• Call update_self_healing_fix({ shortLink, action: "APPLY" })
• Record last_cipe_url, spawn subagent in wait mode

Step 3b: Local Verification Flow

When verifiable (non-e2e) unverified tasks exist:

1. Detect package manager:

   • pnpm-lock.yaml exists → pnpm nx
   • yarn.lock exists → yarn nx
   • Otherwise → npx nx

2. Run verifiable tasks in parallel:

   • Spawn general subagents to run each task concurrently
   • Each subagent runs: <pm> nx run <taskId>
   • Collect pass/fail results from all subagents

3. Evaluate results:

4. Apply-locally + enhance flow:

   • Run nx apply-locally <shortLink>
   • Enhance the code to fix failing tasks
   • Run failing tasks again to verify fix
   • If still failing → increment local_verify_count, loop back to enhance
   • If passing → commit and push, record expected_commit_sha, spawn subagent in wait mode

5. Track attempts (wraps step 4):

   • Increment local_verify_count after each enhance cycle
   • If local_verify_count >= local_verify_attempts (default: 3):
     • Get code in commit-able state
     • Commit and push with message indicating local verification failed
     • Report to user:

       [ci-monitor] Local verification failed after <N> attempts. Pushed to CI for final validation. Failed: <taskIds>
       

     • Record expected_commit_sha, spawn subagent in wait mode (let CI be final judge)

Commit Message Format

git commit -m "fix(<projects>): <brief description>

Failed tasks: <taskId1>, <taskId2>
Local verification: passed|enhanced|failed-pushing-to-ci"

Unverified Fix Flow (No Verification Attempted)

When verificationStatus is FAILED, NOT_EXECUTABLE, or fix has couldAutoApplyTasks != true with no verification:

• Analyze fix content (suggestedFix, suggestedFixReasoning, taskOutputSummary)
• If fix looks correct → apply via MCP
• If fix needs enhancement → use Apply Locally + Enhance Flow above
• If fix is wrong → reject via MCP, fix from scratch, commit, push

Auto-Apply Eligibility

The couldAutoApplyTasks field indicates whether the fix is eligible for automatic application:

• true: Fix is eligible for auto-apply. Subagent keeps polling while verification is in progress. Returns fix_auto_applying when verified, or fix_available if verification fails.
• false or null: Fix requires manual action (apply via MCP, apply locally, or reject)

Key point: When subagent returns fix_auto_applying, do NOT call MCP to apply - self-healing handles it. Just spawn a new subagent in wait mode.

Apply vs Reject vs Apply Locally

• Apply via MCP: Calls update_self_healing_fix({ shortLink, action: "APPLY" }). Self-healing agent applies the fix in CI and a new CIPE spawns automatically. No local git operations needed.
• Apply Locally: Runs nx apply-locally <shortLink>. Applies the patch to your local working directory and sets state to APPLIED_LOCALLY. Use this when you want to enhance the fix before pushing.
• Reject via MCP: Calls update_self_healing_fix({ shortLink, action: "REJECT" }). Marks fix as rejected. Use only when the fix is completely wrong and you'll fix from scratch.

Apply Locally + Enhance Flow

When the fix needs enhancement (use nx apply-locally, NOT reject):

1. Apply the patch locally: nx apply-locally <shortLink> (this also updates state to APPLIED_LOCALLY)
2. Make additional changes as needed
3. Commit and push:

   git add -A
   git commit -m "fix: resolve <failedTaskIds>"
   git push origin $(git branch --show-current)
   

4. Loop to poll for new CIPE

Reject + Fix From Scratch Flow

When the fix is completely wrong:

1. Call MCP to reject: update_self_healing_fix({ shortLink, action: "REJECT" })
2. Fix the issue from scratch locally
3. Commit and push:

   git add -A
   git commit -m "fix: resolve <failedTaskIds>"
   git push origin $(git branch --show-current)
   

4. Loop to poll for new CIPE

Environment Issue Handling

When failureClassification == 'ENVIRONMENT_STATE':

1. Call MCP to request rerun: update_self_healing_fix({ shortLink, action: "RERUN_ENVIRONMENT_STATE" })
2. New CIPE spawns automatically (no local git operations needed)
3. Loop to poll for new CIPE with previousCipeUrl set

No-New-CIPE Handling

When status == 'no_new_cipe':

This means the expected CIPE was never created - CI likely failed before Nx tasks could run.

1. Report to user:

   [ci-monitor] No CI attempt for <sha> after 10 min. Check CI provider for pre-Nx failures (install, checkout, auth). Last CI attempt: <previousCipeUrl>
   

2. If user configured auto-fix attempts (e.g., --auto-fix-workflow):

   • Detect package manager: check for pnpm-lock.yaml, yarn.lock, package-lock.json
   • Run install to update lockfile:

     pnpm install   # or npm install / yarn install
     

   • If lockfile changed:

     git add pnpm-lock.yaml  # or appropriate lockfile
     git commit -m "chore: update lockfile"
     git push origin $(git branch --show-current)
     

   • Record new commit SHA, loop to poll with expectedCommitSha

3. Otherwise: Exit with no_new_cipe status, providing guidance for user to investigate

Exit Conditions

Exit the monitoring loop when ANY of these conditions are met:

Main Loop

Step 1: Initialize Tracking

cycle_count = 0
start_time = now()
no_progress_count = 0
local_verify_count = 0
last_state = null
last_cipe_url = null
expected_commit_sha = null

Step 2: Spawn Subagent

Spawn the ci-watcher subagent to poll CI status:

Fresh start (first spawn, no expected CIPE):

Task(
  agent: "ci-watcher",
  prompt: "Monitor CI for branch '<branch>'.
           Subagent timeout: <subagent-timeout> minutes.
           New-CIPE timeout: <new-cipe-timeout> minutes.
           Verbosity: <verbosity>."
)

After action that triggers new CIPE (wait mode):

Task(
  agent: "ci-watcher",
  prompt: "Monitor CI for branch '<branch>'.
           Subagent timeout: <subagent-timeout> minutes.
           New-CIPE timeout: <new-cipe-timeout> minutes.
           Verbosity: <verbosity>.

           WAIT MODE: A new CIPE should spawn. Ignore old CIPE until new one appears.
           Expected commit SHA: <expected_commit_sha>
           Previous CIPE URL: <last_cipe_url>"
)

Step 3: Handle Subagent Response

When subagent returns:

1. Check the returned status
2. Look up default behavior in the table above
3. Check if user instructions override the default
4. Execute the appropriate action
5. If action expects new CIPE, update tracking (see Step 3a)
6. If action results in looping, go to Step 2

Step 3a: Track State for New-CIPE Detection

After actions that should trigger a new CIPE, record state before looping:

CRITICAL: When passing expectedCommitSha or last_cipe_url to the subagent, it enters wait mode:

• Subagent will completely ignore the old/stale CIPE
• Subagent will only wait for new CIPE to appear
• Subagent will NOT return to main agent with stale CIPE data
• Once new CIPE detected, subagent switches to normal polling

Why wait mode matters for context preservation: Stale CIPE data can be very large (task output summaries, suggested fix patches, reasoning). If subagent returns this to main agent, it pollutes main agent's context with useless data since we already processed that CIPE. Wait mode keeps stale data in the subagent, never sending it to main agent.

Step 4: Progress Tracking

After each action:

• If state changed significantly → reset no_progress_count = 0
• If state unchanged → no_progress_count++
• On new CI attempt detected → reset local_verify_count = 0

Status Reporting

Based on verbosity level:

User Instruction Examples

Users can override default behaviors:

Error Handling

Example Session

Example 1: Normal Flow with Self-Healing (medium verbosity)

[ci-monitor] Starting CI monitor for branch 'feature/add-auth'
[ci-monitor] Config: max-cycles=5, timeout=120m, verbosity=medium

[ci-monitor] Spawning subagent to poll CI status...
[CI Monitor] CI attempt: IN_PROGRESS | Self-Healing: NOT_STARTED | Elapsed: 1m
[CI Monitor] CI attempt: FAILED | Self-Healing: IN_PROGRESS | Elapsed: 3m
[CI Monitor] CI attempt: FAILED | Self-Healing: COMPLETED | Elapsed: 5m

[ci-monitor] Fix available! Verification: COMPLETED
[ci-monitor] Applying fix via MCP...
[ci-monitor] Fix applied in CI. Waiting for new CI attempt...

[ci-monitor] Spawning subagent to poll CI status...
[CI Monitor] New CI attempt detected!
[CI Monitor] CI attempt: SUCCEEDED | Elapsed: 8m

[ci-monitor] CI passed successfully!

[ci-monitor] Summary:
  - Total cycles: 2
  - Total time: 12m 34s
  - Fixes applied: 1
  - Result: SUCCESS

Example 2: Pre-CI Failure (medium verbosity)

[ci-monitor] Starting CI monitor for branch 'feature/add-products'
[ci-monitor] Config: max-cycles=5, timeout=120m, auto-fix-workflow=true

[ci-monitor] Spawning subagent to poll CI status...
[CI Monitor] CI attempt: FAILED | Self-Healing: COMPLETED | Elapsed: 2m

[ci-monitor] Applying fix locally, enhancing, and pushing...
[ci-monitor] Committed: abc1234

[ci-monitor] Spawning subagent to poll CI status...
[CI Monitor] Waiting for new CI attempt... (expected SHA: abc1234)
[CI Monitor] ⚠️  CI attempt timeout (10 min). Returning no_new_cipe.

[ci-monitor] Status: no_new_cipe
[ci-monitor] --auto-fix-workflow enabled. Attempting lockfile update...
[ci-monitor] Lockfile updated. Committed: def5678

[ci-monitor] Spawning subagent to poll CI status...
[CI Monitor] New CI attempt detected!
[CI Monitor] CI attempt: SUCCEEDED | Elapsed: 18m

[ci-monitor] CI passed successfully!

[ci-monitor] Summary:
  - Total cycles: 3
  - Total time: 22m 15s
  - Fixes applied: 1 (self-healing) + 1 (lockfile)
  - Result: SUCCESS