The Agent Skills Directory

[COMMAND_EXECUTION]: The runbook templates provided in the examples/runbooks/ directory (e.g., cache-invalidation.md, database-failover.md, ddos-mitigation.md) contain standard shell and CLI commands for infrastructure management. These include redis-cli, pg_ctl, aws route53, and curl calls to well-known service APIs (Cloudflare, AWS). These commands are consistent with the skill's primary purpose of incident response and represent industry-standard operational procedures.
[CREDENTIALS_UNSAFE]: The integration scripts (pagerduty-slack.py, postmortem-generator.py, statuspage-auto-update.py) are designed to interact with external APIs. They correctly utilize environment variables (e.g., SLACK_BOT_TOKEN, STATUSPAGE_API_KEY) to manage secrets rather than hardcoding them in the source code. This aligns with secure coding practices for secret management.
[INDIRECT_PROMPT_INJECTION]: The scripts postmortem-generator.py and statuspage-auto-update.py ingest data from external chat platforms (Slack) to automate documentation tasks. This creates a surface where untrusted chat content could be interpolated into generated files. However, the risk is inherent to the functionality provided, and the scripts include specific regex-based parsing to identify and validate relevant content before processing.
[DYNAMIC_EXECUTION]: While the skill includes several Python scripts that implement webhooks or API clients using standard frameworks like Flask, they do not utilize dangerous dynamic execution patterns such as eval() or exec() on untrusted input. The logic is deterministic and focused on structured data transformation.

managing-incidents