The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/add_skill.py downloads Python scripts and other files from GitHub repositories (raw.githubusercontent.com) and saves them to a local directory. These scripts are then registered with the local agent platform for execution through the import-local API call.
[EXTERNAL_DOWNLOADS]: The skill fetches data from skills.sh for search results and downloads content from GitHub to facilitate skill installation.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute local Python scripts (find_skills.py and add_skill.py) to search for and install external content.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from an external registry and third-party GitHub repositories.
Ingestion points: Data is ingested from the skills.sh search API and file contents fetched from GitHub repositories (specifically SKILL.md frontmatter).
Boundary markers: The SKILL.md instructions include mandatory human-in-the-loop checkpoints, requiring the agent to wait for user confirmation before proceeding with installation turns.
Capability inventory: The skill has the capability to write files to the local disk and make POST requests to a local API (localhost:62610) to register new executable code.
Sanitization: The installation script filters for specific binary file extensions (e.g., .exe, .so) during the download process, but does not sanitize the logical content of the downloaded scripts.

skill-finder