The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external artifacts, creating a potential surface for indirect prompt injection.
Ingestion points: Instructions in SKILL.md involve reading codebase documentation (GLOSSARY.md, CONTEXT.md, ADRs), user-provided bug reports, and diagnostic artifacts such as HAR files, event logs, and network traces.
Boundary markers: The instructions do not specify the use of delimiters or clear separation between the ingested data and the agent's instructional context.
Capability inventory: The skill has significant capabilities listed in SKILL.md (Phase 1), including executing curl commands, CLI tools, headless browser scripts (Playwright/Puppeteer), and custom bash reproduction scripts.
Sanitization: The skill includes a dedicated 'Redact' section in SKILL.md that explicitly instructs the agent to replace secrets with <REDACTED> and keep credentials in environment variables, which mitigates data exposure risk.
[DYNAMIC_EXECUTION]: The skill utilizes a bash template (scripts/hitl-loop.template.sh) that the agent is expected to edit and execute at runtime to facilitate human-in-the-loop debugging. While this is a standard operational pattern for this skill's purpose, it represents the dynamic assembly and execution of script content.

diagnosing-bugs