The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses git clone within a bash script to download third-party repositories for analysis. This is a core feature of the skill, and the instructions explicitly forbid executing or installing any of the downloaded content.- [COMMAND_EXECUTION]: The skill executes a helper script, survey.sh, which performs repository inspection using standard system tools like git, bash, awk, and sed. The script follows safe practices such as variable quoting to mitigate command injection risks from user-supplied URLs.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted third-party source code. Maliciously crafted code within a repository could attempt to influence the sub-agents' analysis or the final synthesis report.
Ingestion points: Third-party code repositories cloned via git based on user-provided URLs in SKILL.md and survey.sh.
Boundary markers: The sub-agent prompts use distinct sections for context and file content, though they do not include specific 'ignore instructions within files' directives.
Capability inventory: The skill allows for file system reads and report generation but strictly prohibits code execution or installation.
Sanitization: The skill relies on the 'read-only' constraint and the underlying agent safety filters to prevent malicious content from triggering actions.

evaluate