The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from arbitrary GitHub README files, creating a surface for indirect prompt injection attacks.
Ingestion points: The scripts/sponsor_miner.py script fetches README text from various external repositories via the GitHub API and stores it locally for analysis.
Boundary markers: The collected data is presented to the agent in shortlist.csv and run-summary.md as "context" or "fit_notes". There are no strong delimiters or instructions to ignore potential commands embedded within this untrusted external text.
Capability inventory: The skill has the ability to write files to the local workspace, perform network operations to the GitHub API, and execute local commands via subprocess.
Sanitization: The script performs basic text cleaning (removing extra whitespace) but does not sanitize the input for natural language instructions or prompt injection patterns before the agent reviews it.
[COMMAND_EXECUTION]: The skill relies on the execution of local system commands to function.
The scripts/sponsor_miner.py script executes the gh auth token command via subprocess.run to programmatically retrieve the user's GitHub authentication credentials.
The SKILL.md instructions require the agent to execute a bundled Python script, which takes a user-supplied URL as a command-line argument.

sponsor-miner