The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill utilizes npx to download and execute the gitoutput tool directly from the vendor's GitHub repository (AnEntrypoint/gitoutput) at runtime.
[COMMAND_EXECUTION]: The skill performs shell command execution by passing a user-supplied target (local path or remote Git URL) as an argument to the npx command string.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted codebase data by rendering it into images for analysis, creating a surface where malicious instructions embedded in the source code could influence the agent's behavior.
Ingestion points: The target argument in SKILL.md allows for both local directory paths and remote Git URLs to be ingested for analysis.
Boundary markers: The skill does not specify the use of boundary markers or instructions to ignore embedded commands during the visual analysis phase.
Capability inventory: The skill is granted Bash, Read, Write, and Glob tools, enabling it to execute commands, read rendered image files, and write the final report to disk.
Sanitization: There is no evidence of sanitization or filtering of the codebase content prior to the rendering and reporting process.

codebase-vision-report