The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install the parallel-cli tool using curl -fsSL https://parallel.ai/install.sh | bash in the SKILL.md file. This pattern allows a remote server to execute arbitrary shell commands on the local system during installation, which is a high-risk security practice.
[COMMAND_EXECUTION]: The file scripts/generate_schematic.py utilizes the subprocess.run function to execute another Python script (generate_schematic_ai.py) with user-supplied arguments. While it uses the list-based argument passing method, spawning subprocesses based on user-controlled inputs creates a potential surface for command manipulation.
[DATA_EXFILTRATION]: User research queries are transmitted to external endpoints at api.parallel.ai and openrouter.ai. While necessary for the skill's primary function, this constitutes the transmission of potentially sensitive research data to third-party AI providers.
[EXTERNAL_DOWNLOADS]: The skill dynamically fetches data from remote APIs (parallel.ai and openrouter.ai) during research lookups and schematic generation, introducing content from external sources into the agent's environment.
[PROMPT_INJECTION]: The skill has a risk of indirect prompt injection as it ingests untrusted data from external research results and presents it to the agent.
Ingestion points: API responses containing research snippets from api.parallel.ai and openrouter.ai are processed in research_lookup.py.
Boundary markers: The instructions do not define delimiters or specific 'ignore' directives for the research content before displaying it or saving it to the sources/ directory.
Capability inventory: The skill has access to the Bash tool and performs subprocess calls in scripts/generate_schematic.py.
Sanitization: No explicit sanitization or filtering of the research output was observed before it is integrated into the session context.

research-lookup