AI Agent Skill

Teach your AI coding assistant to diagnose Python applications using Peeka.

Table of Contents

  1. What is peeka-diagnostics
  2. Skill Contents Overview
    1. Diagnostic Decision Tree
    2. 4 Diagnostic Playbooks
    3. JSONL Parsing
  3. Installation
    1. Prerequisites
    2. Method 1: Project-Level Installation (Recommended)
    3. Method 2: Global Installation
  4. Usage
    1. OpenCode
    2. Other AI Tools
    3. Trigger Keywords
  5. Example Scenarios
    1. Scenario 1: Slow API Endpoint
    2. Scenario 2: Memory Leak
    3. Scenario 3: Thread Deadlock
  6. Maintaining the Skill
    1. Updating
    2. Custom Extensions
  7. FAQ
    1. AI doesn’t use peeka for diagnostics?
    2. AI diagnostic commands fail?
    3. Where can I find the skill file?

What is peeka-diagnostics

peeka-diagnostics is an AI Agent skill file (SKILL.md). Once installed, your AI coding assistant (OpenCode, Cursor, Cline, etc.) can:

  • Auto-diagnose: Choose the right peeka commands based on observed symptoms (slow requests, memory leaks, deadlocks, etc.)
  • Execute diagnostics: Attach to running Python processes via peeka-cli and collect data in real time
  • Parse results: Analyze structured JSONL output using jq pipelines
  • Follow playbooks: Apply built-in diagnostic workflows (performance, exceptions, memory, threads)

The skill focuses on Peeka’s core diagnostic workflow: watch, trace, stack, monitor, top, sc, sm, memory, inspect, logger, thread, and run, with flag guidance, jq recipes, condition expression syntax, and safety protocols.

Skill Contents Overview

Diagnostic Decision Tree

The skill includes a symptom-to-command mapping table. The AI automatically selects the right diagnostic path:

Symptom Recommended Commands Goal
Slow response / high latency watch → trace Find the slow sub-call
Wrong return value / logic bug watch to observe I/O Correlate inputs to outputs
Exception / unexpected error watch -e → stack Find where and why the exception occurs
High memory / memory leak memory command suite Find what allocates and holds memory
High CPU top → trace Find CPU-intensive code paths
Deadlock / hang thread → stack Find lock contention points
gevent/eventlet or monkey-patch anomalies patch-status → attach / watch Confirm runtime patch state and RPL integrity

4 Diagnostic Playbooks

Playbook Scenario Core Flow
A: Performance Slow APIs, high latency watch for slow calls → trace call tree → pinpoint bottleneck
B: Exceptions Production errors watch -e to capture → stack for context → root cause analysis
C: Memory Growing memory usage memory start tracking → snapshots → diff → referrers trace
D: Threads Deadlocks, stuck threads thread list states → filter WAITING → stack inspection

JSONL Parsing

All CLI output is JSONL. The AI uses jq for structured analysis:

# Filter slow calls
peeka-cli watch "module.func" -n 10 | jq 'select(.type == "observation" and .cost > 100)'

# Extract exception info
peeka-cli watch "module.func" -e -n 5 | jq 'select(.success == false) | {func: .func_name, error: .exception}'

# Memory top analysis
peeka-cli memory --action top | jq '.data.top_allocations[:5]'

Installation

Prerequisites

  • Peeka installed (see Installation Guide)
  • An AI coding assistant that supports Skill/SKILL.md files

Copy the skill file into your project’s .agents/skills/ directory:

# From your project root
mkdir -p .agents/skills/peeka-diagnostics

# Download the skill file from the Peeka repository
curl -o .agents/skills/peeka-diagnostics/SKILL.md \
  https://raw.githubusercontent.com/peeka-project/peeka/master/.agents/skills/peeka-diagnostics/SKILL.md

Resulting directory structure:

your-project/
├── .agents/
│   └── skills/
│       └── peeka-diagnostics/
│           └── SKILL.md
├── src/
│   └── ...
└── ...

Method 2: Global Installation

If your AI tool supports a global skills directory (e.g., ~/.config/opencode/skills/):

mkdir -p ~/.config/opencode/skills/peeka-diagnostics
curl -o ~/.config/opencode/skills/peeka-diagnostics/SKILL.md \
  https://raw.githubusercontent.com/peeka-project/peeka/master/.agents/skills/peeka-diagnostics/SKILL.md

Usage

OpenCode

Load the skill via load_skills in OpenCode:

task(
  category="deep",
  load_skills=["peeka-diagnostics"],
  prompt="My API is slow, diagnose PID 12345 with peeka"
)

Or reference it directly in conversation:

@peeka-diagnostics My Python service memory keeps growing, PID 54321, help me investigate

Other AI Tools

For Cursor, Cline, or other AI tools with custom instruction support:

  1. Ensure the skill file is at .agents/skills/peeka-diagnostics/SKILL.md
  2. The AI tool will auto-discover and load the skill
  3. When you describe Python diagnostic problems, the AI will apply the skill’s knowledge

Trigger Keywords

These keywords activate the peeka-diagnostics skill:

  • debug python, diagnose python
  • slow app, memory leak, high CPU
  • trace function, watch expression
  • thread deadlock, runtime debugging
  • profile python, peeka

Example Scenarios

Scenario 1: Slow API Endpoint

“My /api/users endpoint went from 50ms to 2s response time, target process PID is 12345”

The AI will automatically:

  1. peeka-cli attach 12345
  2. peeka-cli sc "*user*" to discover relevant classes
  3. peeka-cli watch "myapp.api.users.get_users" -n 5 --condition "cost > 100" to filter slow requests
  4. peeka-cli trace "myapp.api.users.get_users" -n 3 -d 5 to break down the call tree
  5. Analyze JSONL output and pinpoint the bottleneck sub-function
  6. peeka-cli reset && peeka-cli detach to clean up

Scenario 2: Memory Leak

“My Python service memory grew from 200MB to 2GB after a few hours, PID 54321”

The AI will automatically:

  1. Attach and start tracemalloc tracking
  2. Take multiple snapshots at intervals and diff them
  3. Use memory --action top to find the largest allocation sources
  4. Use memory --action referrers to trace reference chains
  5. Output analysis report with fix recommendations

Scenario 3: Thread Deadlock

“My service is stuck, all requests are timing out, PID 33333”

The AI will automatically:

  1. peeka-cli thread to list all thread states
  2. Filter threads in WAITING state
  3. Get detailed stack traces for suspect threads
  4. Analyze lock contention and provide recommendations

Maintaining the Skill

Updating

When Peeka releases a new version, update the skill file:

curl -o .agents/skills/peeka-diagnostics/SKILL.md \
  https://raw.githubusercontent.com/peeka-project/peeka/master/.agents/skills/peeka-diagnostics/SKILL.md

Custom Extensions

The skill file is plain Markdown — extend it to fit your project:

  • Add project-specific diagnostic patterns
  • Add shortcuts for commonly-watched function patterns
  • Supplement with project-specific troubleshooting steps

FAQ

AI doesn’t use peeka for diagnostics?

  • Confirm the skill file is correctly installed at .agents/skills/peeka-diagnostics/SKILL.md
  • Use Python diagnostic keywords explicitly in your prompt
  • Verify peeka-cli is installed and accessible

AI diagnostic commands fail?

  • Check Peeka installation: peeka-cli --help
  • Confirm the target process is still running: ps -p <pid>
  • Check permissions (ptrace_scope, CAP_SYS_PTRACE)

Where can I find the skill file?

Back to top

Copyright © 2026 Peeka contributors. Distributed under the Apache License 2.0.

This site uses Just the Docs, a documentation theme for Jekyll.