The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external GitHub forks, specifically reading SKILL.md files and commit messages to generate summaries and synthesis.
Ingestion points: In step A4, the skill fetches the content of SKILL.md from external forks using the GitHub API.
Boundary markers: The instructions do not specify any delimiters or warnings to the agent when processing this untrusted external content.
Capability inventory: The skill uses gh api for network interaction, writes files to the output/ and memory/ directories, and sends notifications via a ./notify script.
Sanitization: There is no mention of sanitizing, escaping, or filtering the external fork content before the agent synthesizes it into descriptions.
[SAFE]: The skill uses gh api to interact with GitHub, which is a well-known service. The use of base64 -d in steps A4 and B2 is a standard procedure for decoding file contents returned by the GitHub API and does not constitute malicious obfuscation.

fork-fleet