The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file explicitly instructs the agent to bypass standard verification protocols by stating "无需请求确认；直接执行" (No need to request confirmation; directly execute). This is a prompt injection pattern designed to increase the agent's autonomy and reduce human oversight during command execution.
[COMMAND_EXECUTION]: The skill relies on shell:exec permissions to run multiple Python scripts (sun_calc.py, plot_sunpath.py, shadow_calc.py, annual_sun_hours.py, terrain_shadow.py, comfort_calc.py). While this is the intended functionality, the use of shell execution with user-provided or agent-generated arguments presents a persistent risk of command injection if inputs are not strictly validated.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface through its handling of external data in scripts/terrain_shadow.py.
Ingestion points: The script accepts a positional argument dem which is used as a file path for rasterio.open().
Boundary markers: None are defined in the instructions to prevent the agent from accessing unauthorized paths if manipulated by a user.
Capability inventory: The skill possesses shell:exec capabilities and the ability to read files via the rasterio library.
Sanitization: The script performs no validation or sanitization on the dem file path before attempting to open it, which could be used to probe the existence of sensitive local files.

sunlight-analysis