The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a shell script scripts/check-upstream.sh and instructs the agent to execute it using bash to verify the state of upstream dependencies.
[EXTERNAL_DOWNLOADS]: The check-upstream.sh script and SKILL.md instructions utilize curl to fetch content from external sources including raw.githubusercontent.com and auth.estifanos.com.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface by downloading external documentation and README files from GitHub repositories which the agent is directed to read and follow.
Ingestion points: The check-upstream.sh script downloads files from raw.githubusercontent.com into the /tmp directory (e.g., /tmp/robel-auth-main-readme.md).
Boundary markers: None. The instructions do not specify delimiters or warnings to ignore instructions embedded within the fetched content.
Capability inventory: The agent has the ability to execute shell commands, perform network requests, and modify the project codebase.
Sanitization: There is no evidence of sanitization or filtering applied to the external content before it is processed by the agent.

robel-auth