debugging-workflows

от github

Отладка рабочих процессов GitHub Actions путем загрузки логов, анализа сводок и понимания того, как агентные рабочие процессы и брандмауэр AWF работают вместе.

npx skills add https://github.com/github/gh-aw-firewall --skill debugging-workflows

Debugging Workflows Skill

Use this skill when you need to debug GitHub Actions workflows, download workflow logs or summaries, or understand how agentic workflows and the AWF firewall work together.

Quick Start

Download Workflow Logs

Use the download-workflow-logs.ts script to download logs from a workflow run:

# Download logs from the latest workflow run
npx tsx .github/skills/debugging-workflows/download-workflow-logs.ts

# Download logs from a specific run ID
npx tsx .github/skills/debugging-workflows/download-workflow-logs.ts --run-id 1234567890

# Download logs from a specific workflow
npx tsx .github/skills/debugging-workflows/download-workflow-logs.ts --workflow test-integration.yml

# Save logs to a specific directory
npx tsx .github/skills/debugging-workflows/download-workflow-logs.ts --output ./my-logs

Download Workflow Summary

Use the download-workflow-summary.ts script to get a summary of workflow runs:

# Get summary of latest workflow runs
npx tsx .github/skills/debugging-workflows/download-workflow-summary.ts

# Get summary for a specific workflow run
npx tsx .github/skills/debugging-workflows/download-workflow-summary.ts --run-id 1234567890

# Get summary for a specific workflow file
npx tsx .github/skills/debugging-workflows/download-workflow-summary.ts --workflow test-integration.yml

# Get summary as JSON
npx tsx .github/skills/debugging-workflows/download-workflow-summary.ts --format json

GitHub CLI Commands

The gh CLI is essential for debugging workflows. Here are the most useful commands:

List Workflow Runs

# List recent workflow runs
gh run list --limit 10

# List runs for a specific workflow
gh run list --workflow test-integration.yml --limit 10

# List only failed runs
gh run list --status failure --limit 10

# List runs in JSON format for parsing
gh run list --json databaseId,name,status,conclusion,createdAt --limit 10

View Workflow Run Details

# View a specific run
gh run view <run-id>

# View run with job details
gh run view <run-id> --verbose

# View run as JSON
gh run view <run-id> --json jobs,conclusion,status

Download Run Logs

# Download all logs for a run
gh run download <run-id>

# Download specific artifact
gh run download <run-id> --name <artifact-name>

# Download to specific directory
gh run download <run-id> --dir ./logs

Watch a Running Workflow

# Watch a workflow run in real-time
gh run watch <run-id>

# Watch with exit code (useful for CI)
gh run watch <run-id> --exit-status

Re-run Failed Jobs

# Re-run failed jobs only
gh run rerun <run-id> --failed

# Re-run all jobs
gh run rerun <run-id>

Understanding Agentic Workflows

What are Agentic Workflows?

Agentic workflows are GitHub Actions workflows that use AI agents (like GitHub Copilot or Claude) to perform tasks. They are defined using markdown + YAML frontmatter format in .github/workflows/*.md files and compiled to GitHub Actions YAML (.lock.yml files).

Key Components

  1. Workflow File Format: .github/workflows/<name>.md

    • YAML frontmatter for configuration
    • Markdown body for AI instructions
    • Compiles to .github/workflows/<name>.lock.yml
  2. Triggers (on: field):

    • Standard GitHub events: issues, pull_request, push, schedule
    • Command triggers: /mention in issues/comments
    • workflow_dispatch for manual triggers
  3. Safe Outputs: Controlled way for AI to create GitHub entities

    • create-issue: - Create GitHub issues
    • create-pull-request: - Create PRs with git patches
    • add-comment: - Add comments to issues/PRs
    • add-labels: - Add labels to issues/PRs
    • create-discussion: - Create GitHub discussions
  4. Tools Configuration (tools: field):

    • github: - GitHub API tools
    • agentic-workflows: - Workflow introspection tools
    • edit: - File editing tools
    • web-fetch: / web-search: - Web access tools
    • bash: - Shell command tools

Compiling Workflows

# Compile all workflows
gh aw compile

# Compile a specific workflow
gh aw compile <workflow-name>

# Compile with strict security checks
gh aw compile --strict

Debugging Agentic Workflows

# View status of all agentic workflows
gh aw status

# Download and analyze logs from previous runs
gh aw logs <workflow-name> --json

# Audit a specific run for issues
gh aw audit <run-id> --json

Common Issues

  1. Missing Tool Calls: Check missing_tools in audit output
  2. Safe Output Failures: Review safe_outputs.jsonl artifact
  3. Permission Issues: Verify permissions: block in frontmatter
  4. Network Blocked: Check network: configuration for allowed domains

Understanding the AWF Firewall

What is AWF?

AWF (Agent Workflow Firewall) is a tool that provides L7 (HTTP/HTTPS) egress control for GitHub Copilot CLI and other agents. It restricts network access to a whitelist of approved domains using Squid proxy and Docker containers.

Architecture Overview

┌─────────────────────────────────────────┐
│  Host (GitHub Actions Runner / Local)   │
│                                          │
│  ┌────────────────────────────────────┐ │
│  │   Firewall CLI (awf)              │ │
│  │   - Parse arguments                │ │
│  │   - Generate Squid config          │ │
│  │   - Start Docker Compose           │ │
│  └────────────────────────────────────┘ │
│           │                              │
│           ▼                              │
│  ┌──────────────────────────────────┐   │
│  │   Docker Compose                 │   │
│  │  ┌────────────────────────────┐  │   │
│  │  │  Squid Proxy Container     │  │   │
│  │  │  - Domain ACL filtering    │  │   │
│  │  │  - HTTP/HTTPS proxy        │  │   │
│  │  └────────────────────────────┘  │   │
│  │           ▲                       │   │
│  │  ┌────────┼───────────────────┐  │   │
│  │  │ Agent Container            │  │   │
│  │  │ - Full filesystem access   │  │   │
│  │  │ - iptables redirect        │  │   │
│  │  │ - All traffic → Squid      │  │   │
│  │  └────────────────────────────┘  │   │
│  └──────────────────────────────────┘   │
└─────────────────────────────────────────┘

Key Containers

  • awf-squid - Squid proxy container (IP: 172.30.0.10)

    • Filters HTTP/HTTPS traffic based on domain allowlist
    • Logs all traffic decisions
  • awf-agent - Agent execution container (IP: 172.30.0.20)

    • Runs the actual command/agent
    • Has iptables rules to redirect traffic to Squid
    • Full filesystem access via /host mount

Traffic Flow

  1. Command runs in agent container
  2. All HTTP/HTTPS traffic → iptables DNAT → Squid proxy
  3. Squid checks domain against allowlist
  4. Allowed → forward to destination
  5. Blocked → return 403 Forbidden

Squid Log Analysis

# View Squid access log (shows traffic decisions)
docker exec awf-squid cat /var/log/squid/access.log

# Find blocked domains
docker exec awf-squid grep "TCP_DENIED" /var/log/squid/access.log | awk '{print $3}' | sort -u

# Count blocked by domain
docker exec awf-squid grep "TCP_DENIED" /var/log/squid/access.log | awk '{print $3}' | sort | uniq -c | sort -rn

# Real-time blocked traffic
docker exec awf-squid tail -f /var/log/squid/access.log | grep --line-buffered TCP_DENIED

Squid Decision Codes

  • TCP_TUNNEL:HIER_DIRECT = ALLOWED (HTTPS)
  • TCP_MISS:HIER_DIRECT = ALLOWED (HTTP)
  • TCP_DENIED:HIER_NONE = BLOCKED

Running Commands Through Firewall

# Basic usage
sudo awf --allow-domains github.com 'curl https://api.github.com'

# With debug logging
sudo awf --allow-domains github.com --log-level debug 'your-command'

# Keep containers for inspection
sudo awf --allow-domains github.com --keep-containers 'your-command'

Preserved Logs Locations

With --keep-containers:

  • Squid: /tmp/awf-<timestamp>/squid-logs/access.log
  • Agent: /tmp/awf-<timestamp>/agent-logs/

Normal execution (after cleanup):

  • Squid: /tmp/squid-logs-<timestamp>/access.log
  • Agent: /tmp/awf-agent-logs-<timestamp>/
# Find preserved logs
ls -ldt /tmp/awf-* /tmp/squid-logs-* 2>/dev/null | head -5

# View preserved Squid logs
sudo cat $(ls -t /tmp/squid-logs-*/access.log 2>/dev/null | head -1)

Debugging Workflow Failures

Step-by-Step Process

  1. Identify the failing workflow run

    gh run list --status failure --limit 5
    
  2. Get run details

    gh run view <run-id> --verbose
    
  3. Download logs

    gh run download <run-id> --dir ./logs
    # Or use the script:
    npx tsx .github/skills/debugging-workflows/download-workflow-logs.ts --run-id <run-id>
    
  4. Analyze the failure

    • Check job logs for error messages
    • Look for timeout issues
    • Check for permission errors
    • Review network-related errors
  5. For agentic workflows, audit the run

    gh aw audit <run-id> --json
    
  6. If firewall-related, check Squid logs

    # If containers are still running
    docker exec awf-squid cat /var/log/squid/access.log
    
    # Or check preserved logs
    sudo cat /tmp/squid-logs-*/access.log
    

Common Failure Patterns

Permission Denied

Error: Resource not accessible by integration

Fix: Check permissions: in workflow frontmatter

Domain Blocked

curl: (56) Recv failure: Connection reset by peer

Fix: Add domain to --allow-domains or network: configuration

Timeout

Error: The operation was canceled.

Fix: Increase timeout-minutes in workflow configuration

Missing Tool

Tool 'xyz' not found

Fix: Add tool to tools: configuration in workflow frontmatter

Related Documentation

Больше skills от github

console-rendering
github
Инструкции по использованию системы консольного рендеринга на основе тегов структур в Go
official
acquire-codebase-knowledge
github
Используйте этот навык, когда пользователь явно просит составить карту, задокументировать или ознакомиться с существующей кодовой базой. Активируйте по запросам вроде "составь карту этой кодовой базы", "задокументируй…
official
acreadiness-assess
github
Run the AgentRC readiness assessment on the current repository and produce a static HTML dashboard at reports/index.html. Wraps `npx github:microsoft/agentrc…
official
acreadiness-generate-instructions
github
Генерирует настраиваемые файлы инструкций для ИИ-агентов через команду инструкций AgentRC. Создаёт .github/copilot-instructions.md (по умолчанию, рекомендуется для Copilot в VS…
official
acreadiness-policy
github
Помочь пользователю выбрать, написать или применить политику AgentRC. Политики настраивают оценку готовности, отключая нерелевантные проверки, переопределяя влияние/уровень, задавая…
official
add-educational-comments
github
Добавляет учебные комментарии в файлы с кодом, превращая их в эффективные учебные ресурсы. Адаптирует глубину и тон объяснений под три настраиваемых уровня знаний: начальный, средний и продвинутый. Автоматически запрашивает файл, если он не предоставлен, с нумерованным списком для быстрого выбора. Расширяет файлы до 125% только за счёт учебных комментариев (жёсткое ограничение: 400 новых строк; 300 для файлов длиннее 1000 строк). Сохраняет кодировку файла, стиль отступов, синтаксическую корректность и...
official
adobe-illustrator-scripting
github
Пишите, отлаживайте и оптимизируйте скрипты автоматизации Adobe Illustrator с помощью ExtendScript (JavaScript/JSX). Используйте при создании или изменении скриптов, которые управляют…
official
agent-governance
github
Декларативные политики, классификация намерений и журналы аудита для контроля доступа и поведения инструментов ИИ-агентов. Компонуемые политики управления определяют разрешённые/заблокированные инструменты, фильтры контента, ограничения скорости и требования к утверждению — хранятся как конфигурация, а не код. Семантическая классификация намерений обнаруживает опасные запросы (утечка данных, повышение привилегий, инъекция подсказок) до выполнения инструмента с помощью сигналов на основе шаблонов. Декоратор управления на уровне инструмента применяет политики на функции...
official