The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data sources like application logs, error messages, and network traces to diagnose technical issues. This represents a potential surface for indirect prompt injection, though the skill includes instructions to minimize risk.
Ingestion points: The agent is instructed in SKILL.md to read logs, HAR files, and core dumps provided by the environment or user.
Boundary markers: The 'Redact' section explicitly requires the agent to replace secrets with <REDACTED> and to selectively quote artifacts, which acts as a manual boundary.
Capability inventory: The skill utilizes standard developer tools such as curl, test runners, and headless browsers (Playwright/Puppeteer).
Sanitization: Instructions mandate the redaction of authentication headers and other sensitive data before processing.
[COMMAND_EXECUTION]: The skill involves the execution of CLI tools and a human-in-the-loop bash script to facilitate the debugging process.
Evidence: scripts/hitl-loop.template.sh provides a structured template for gathering manual observations. The skill also suggests using curl and CLI snapshots to build feedback loops.

diagnosing-bugs