The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content fetched from external GitHub repositories via the read_file() function. This ingested content could contain malicious instructions designed to influence the agent's subsequent actions.
Ingestion points: Data enters the agent's context through file content retrieved from GitHub repositories via the REST API in scripts/github_client.py.
Boundary markers: The skill does not implement explicit delimiters or warnings to ignore embedded instructions when reading file content.
Capability inventory: The skill includes network write capabilities through functions such as commit_file(), commit_files(), and create_pull_request() in scripts/github_client.py.
Sanitization: The skill performs Base64 encoding for data transmission as required by the GitHub API, but it does not sanitize or validate the logical content of the files retrieved.
[DATA_EXFILTRATION]: The skill provides tools that enable an agent to write local data to external repositories. While this is the primary intended functionality for logging and version control, it creates a vector for data exfiltration if the agent is instructed to transmit sensitive local files to an attacker-controlled repository.
[EXTERNAL_DOWNLOADS]: The skill fetches repository information and file contents from the official GitHub REST API (api.github.com).

invoking-github