The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/generate_image.py allows for the transmission of local file contents to the OpenRouter API. The --input argument reads a specified file path, base64-encodes it, and includes it in a network request to https://openrouter.ai/api/v1/chat/completions. This capability could be exploited to exfiltrate sensitive files (e.g., SSH keys, credentials) if the agent is manipulated into treating them as 'image' inputs.
[COMMAND_EXECUTION]: The scripts/generate_schematic.py file utilizes subprocess.run to orchestrate execution between different scripts in the package. While the implementation uses a list of arguments to minimize shell injection risks, it represents a dynamic execution pattern that allows the skill to trigger additional processes on the host system.
[PROMPT_INJECTION]: The skill uses an indirect prompt injection surface (Category 8) by ingesting untrusted data from the research-lookup tool.
Ingestion points: External literature data enters the context via the research-lookup skill (referenced in SKILL.md).
Boundary markers: The instructions lack explicit delimiters or warnings to ignore instructions embedded within the research data.
Capability inventory: The skill has access to Bash, Write, and Edit tools, as well as subprocess calls within its scripts.
Sanitization: There is no evidence of filtering or escaping logic applied to the external research content before it is processed into manuscript prose.
[SAFE]: The skill follows security best practices for credential management by loading the OPENROUTER_API_KEY from environment variables or .env files rather than hardcoding secrets in the source code.

scientific-writing