The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses uvx to download and execute code directly from the wshobson/agents GitHub repository at runtime. This provides a mechanism for executing unverified remote code from a source outside the verified vendor list.
[COMMAND_EXECUTION]: The skill performs several local command executions, including running ./scripts/eval-skills.py and various make targets such as make eval-skill and make eval-certify. These operations involve shell execution and the invocation of sub-processes.
[PROMPT_INJECTION]: The skill implements an automated file-editing workflow (--fix mode) that represents an indirect prompt injection surface. It reads and processes evaluation data from EVALS.md—which is generated by an external, unverified tool—and uses this data to drive automated edits to SKILL.md files via the Edit tool.
Ingestion points: The skill reads SKILL.md (for evaluation) and tool-generated EVALS.md (to perform fixes).
Boundary markers: No specific delimiters or "ignore instructions" warnings are utilized when processing the content of these files.
Capability inventory: The skill possesses the Edit tool for writing to files and Bash for command execution.
Sanitization: There is no evidence of sanitization or validation performed on the evaluation data before it is used to influence the agent's code-writing actions.

skill-evals