The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The script scripts/kol_discovery.py ingests untrusted data from external LinkedIn posts via the Apify API. This content, including post text and previews, is stored in a CSV and presented to the AI agent for review. If a LinkedIn post contains malicious instructions, it could potentially influence the agent's behavior during Phase 3 (Review & Refine).
Ingestion points: LinkedIn post content (text and content fields) retrieved from the harvestapi/linkedin-post-search actor in scripts/kol_discovery.py.
Boundary markers: The results are written to a CSV file and presented to the agent without specific delimiters or instructions to ignore embedded commands.
Capability inventory: The script performs file system writes (CSV) and makes network requests to the Apify API.
Sanitization: Untrusted post content is truncated but not escaped or sanitized for prompt safety.
[DATA_EXFILTRATION]: The scripts/kol_discovery.py script is vulnerable to a path traversal attack in its output logic. The client_name parameter, read from the kol-discovery.json configuration file, is used to construct the output CSV path: f"{config['client_name']}-kols-{timestamp}.csv". If the client_name contains directory traversal sequences like ../../, the script could write files to locations outside the intended leads directory.
[DATA_EXFILTRATION]: The load_env function in scripts/kol_discovery.py recursively searches for .env files in the current directory and up to five levels of parent directories. This behavior could lead to the unintentional loading and potential exposure of sensitive environment variables or credentials from outside the skill's immediate workspace if executed in a hierarchical directory structure.
[EXTERNAL_DOWNLOADS]: The skill fetches data from Apify's official API (api.apify.com) to retrieve LinkedIn post data using the harvestapi/linkedin-post-search actor. This is standard functionality for interacting with this well-known service.

kol-discovery