The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches external skill definitions and reference materials from GitHub repositories using the git clone command within the bundled skill-dl script.
[COMMAND_EXECUTION]: The skill executes various system commands, including bash, git, curl, and npx, to perform its discovery, download, and file-processing workflows.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill reads untrusted SKILL.md and reference files downloaded from the internet during the research phase (Step 4a in SKILL.md).
Boundary markers: There are no explicit boundary markers or instructions to treat external content as untrusted data during the synthesis of the new skill.
Capability inventory: The agent has the capability to write files (Step 7), execute shell commands via bundled scripts, and access the network via curl and git.
Sanitization: No explicit sanitization or filtering is performed on the content of the downloaded skills before they are processed by the agent.
[DATA_EXFILTRATION]: The skill optionally sends user-defined search keywords to the Serper Google API (google.serper.dev) if the user provides a SERPER_API_KEY environment variable.

build-skill