The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The resources/setup.sh script executes a self-update command using npx skills update from a remote repository (github/awesome-copilot). While this repository belongs to the author's namespace, it represents a mechanism for executing remote code updates.
[EXTERNAL_DOWNLOADS]: The setup.sh script performs automated installations of the pixie-qa package from public registries (pip, uv, poetry). The metadata identifies the source as a non-trusted third-party repository (github.com/yiouli/pixie-qa).
[COMMAND_EXECUTION]: The skill requires the agent to execute a shell script (resources/setup.sh) that manages environment initialization, package management, and background service operations.
[INDIRECT_PROMPT_INJECTION]: The skill instructions in SKILL.md (Step 1) direct the agent to ingest and prioritize external, user-specified files such as EVAL_SPEC.md or REQUIREMENTS.md to define its workflow.
Ingestion points: The agent reads user-provided documentation, specification files, and the application's own source code to determine capabilities and criteria.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when reading these external files.
Capability inventory: The skill has significant capabilities including script execution, package installation, and modifying project files.
Sanitization: There is no mention of sanitizing or validating the content of the ingested specification files before the agent adopts their instructions.
[PERSISTENCE]: The setup.sh script starts a background web server process via pixie start. The skill instructions explicitly note that this process remains active across sessions and must be manually terminated by the user, representing a functional persistence mechanism.

eval-driven-dev