The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation describes using the pnpm action CLI for testing actions locally and pnpm actions:audit for static analysis of action definitions to identify redundant or dead code.
[EXTERNAL_DOWNLOADS]: Documents tools like web-search and web-request that allow the agent to fetch and process external content from the public internet for analysis.
[REMOTE_CODE_EXECUTION]: Describes a run-code utility that provides a sandboxed environment for executing code, including support for background tasks and long-running analysis jobs.
[DATA_EXFILTRATION]: Documents the provider-api-request trio, which facilitates authenticated HTTP requests to external service providers while incorporating security boundaries like host allow-listing and secret redaction.
[PROMPT_INJECTION]: The skill identifies an attack surface for indirect prompt injection via the ingestion of untrusted data from the web.
Ingestion points: External data enters the context via web-search, web-request, and provider-api-docs (SKILL.md).
Boundary markers: The framework suggests using specific responseMode settings (e.g., markdown) and search filters to limit the ingestion of untrusted raw HTML (SKILL.md).
Capability inventory: The framework exposes run-code, workspaceWrite, and provider-api-request as potential targets for instruction hijacking (SKILL.md).
Sanitization: The documentation highlights framework-level protections including secret redaction, private-network blocking, and authenticated request injection (SKILL.md).

actions