Pipeline Monitor Skill

Spawns a lightweight Haiku agent to monitor a running pipeline in the background. Alerts only when something goes wrong.

When to Use

Invoke /pipeline-monitor [run_id] when:

•A pipeline run is in progress and you want hands-off monitoring
•You want to continue other work while the pipeline runs
•You need to be alerted if the pipeline fails or gets stuck

Usage

code

/pipeline-monitor 86

Or without a run ID (auto-detects active run):

code

/pipeline-monitor

What the Monitor Does

•
Checks every 60 seconds for:
- •Fetch progress (conversations fetched count)
- •Phase transitions (classification → embedding → themes → stories)
- •Error messages or failure status
- •Stuck detection (no progress for 5+ minutes)
•
Output behavior:
- •Normal: One-line status every 2 minutes (e.g., [09:05] OK - 1700 fetched, classification)
- •Alert: ALERT: [details] with stack trace from /tmp/feedforward-app.log
- •Complete: Final stats summary
- •Failed: Error details and log excerpt
•
Runs until pipeline status is completed or failed

Implementation

When invoked, Claude spawns a Haiku agent with this prompt:

code

You are monitoring FeedForward pipeline Run {run_id}.

## Check Commands (run every 60s):

1. Fetch progress:
   grep "Fetched" {output_file} | tail -1

2. Current status:
   grep "status=" {output_file} | tail -1

3. Check for errors:
   grep -i "error\|failed\|broken pipe" {output_file} | tail -3

4. DB status (if needed):
   psql "postgresql://localhost:5432/feedforward" -c \
     "SELECT status, current_phase, conversations_classified, themes_extracted, stories_created FROM pipeline_runs WHERE id = {run_id};"

## Output Rules:
- Normal: "[HH:MM] OK - {count} fetched, {phase} phase"
- Error: "ALERT: {error}" then check /tmp/feedforward-app.log
- Complete: "COMPLETE: {convos} classified, {themes} themes, {stories} stories"
- Failed: "FAILED: {error}" with log excerpt

## Loop:
sleep 60 between checks. Stop when status = completed or failed.

Files Monitored

File	Purpose
`/private/tmp/claude/.../tasks/{task_id}.output`	Pipeline script stdout
`/tmp/feedforward-app.log`	Server logs with stack traces

Example Output

code

[09:00] OK - 500 fetched, classification phase
[09:02] OK - 1200 fetched, classification phase
[09:04] OK - 1763 fetched, embedding_generation phase
[09:06] OK - embedding complete, theme_extraction phase
[09:08] OK - 145 themes, story_creation phase
[09:10] COMPLETE: 1763 classified, 145 themes, 12 stories, 89 orphans

Or on failure:

code

[09:05] ALERT: Pipeline failed with [Errno 32] Broken pipe
Stack trace from /tmp/feedforward-app.log:
  File "src/classification_pipeline.py", line 464, in run_pipeline_async
    print(f"\n{'='*60}")
BrokenPipeError: [Errno 32] Broken pipe

Integration with dev-pipeline-run.sh

The monitor complements dev-pipeline-run.sh:

•Start pipeline: ./scripts/dev-pipeline-run.sh --days 45
•When it backgrounds, invoke: /pipeline-monitor
•Continue with other work
•Get alerted only if something breaks