The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to download and execute a shell script from a remote URL (https://astral.sh/uv/install.sh) by piping the output of curl directly to sh. While the source is the official domain for the uv tool, this execution pattern is a security risk as it bypasses manual inspection of the script content.\n- [COMMAND_EXECUTION]: The documentation suggests executing the agent with the --dangerously-skip-permissions flag. This configuration explicitly disables the platform's security prompts that normally require user approval for file modifications and shell command execution, granting the agent full autonomy to alter the local environment.\n- [PROMPT_INJECTION]: The skill's architecture creates an indirect prompt injection surface. The agent is instructed to read directives from program.md and modify Python scripts (train.py) based on that content.\n
Ingestion points: Directive text from program.md.\n
Boundary markers: Absent; the agent is not instructed to isolate or ignore embedded commands within the directive.\n
Capability inventory: Full file-write access to train.py and shell execution via uv run.\n
Sanitization: Absent; external instructions are interpolated directly into the agent's coding task loop.

autoresearch