The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external content retrieved from the Perplexity API (grounded web answers and search results). This constitutes an attack surface where malicious instructions embedded in web pages could influence the agent's behavior.
Ingestion points: Data enters the context via createAgent.ts, search.ts, and getAgentResponse.ts through calls to api.perplexity.ai.
Boundary markers: The skill separates the answer and sources fields, providing structure, but does not implement explicit LLM-level delimiters or 'ignore' instructions for the retrieved text.
Capability inventory: The skill includes network operations (fetch) and local script execution via spawnSync in cli.js.
Sanitization: Output is validated against Zod schemas, but the textual content of the answer is not sanitized for prompt injection patterns.
[COMMAND_EXECUTION]: The cli.js entry point uses node:child_process.spawnSync to execute the connector's own TypeScript or compiled JavaScript files. This is used strictly for internal delegation to manage runtime environment compatibility and does not execute untrusted user input or remote code.
[CREDENTIALS_SAFE]: The skill uses a resolver-based authentication model, recommending the use of environment variables (env:PERPLEXITY_API_KEY) or Zapier-managed connections. The documentation explicitly warns against dumping environment variables to prevent accidental credential leakage in transcripts.

perplexity