The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automates the download of testing and security tools from well-known sources. It clones the official Semgrep rules repository from GitHub and installs the Playwright-CLI skill from Microsoft's official repository.
[COMMAND_EXECUTION]: The provided scripts (setup-tools.sh and check-tools.sh) execute system commands to install global Node.js packages and Python modules. It also uses git worktree and pkill -9 node to manage and clean up the environment during code evaluation cycles.
[DYNAMIC_EXECUTION]: To evaluate code implementations, the skill dynamically generates JavaScript test harnesses from templates (api_test.template.mjs, ui_test.template.mjs) by populating them with project-specific routes and logic, then executes them using Node.js.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a broad attack surface for indirect prompt injection because it ingests and processes untrusted project code and prompts for evaluation.
Ingestion points: The skill recursively discovers and reads project files and git diffs in the working directory (referenced in EVAL_STANDARDS.md).
Boundary markers: No explicit markers or "ignore instructions" warnings are applied to the code being evaluated.
Capability inventory: The skill uses fetch for network operations, fs for file writing (evidence and results), and executes generated scripts via Node.js.
Sanitization: The skill analyzes and executes tests against source code as-is without sanitization, assuming the code itself may be the subject of security findings.

code-arena-cn