The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill explicitly prompts the user to input sensitive API credentials (APIFY_TOKEN and ELEVENLABS_API_KEY) into the chat session. This practice exposes secrets in the conversation history and environment variables.
[REMOTE_CODE_EXECUTION]: The instructions command the agent to dynamically generate and run a local script (temp_scraper.mjs). This script uses the apify-client to perform network operations and execute remote 'actors' (code) on the Apify platform.
[EXTERNAL_DOWNLOADS]: The skill requires the agent to install external third-party software at runtime using npm install apify-client. This introduces dependencies from a public registry without version pinning or integrity checks.
[COMMAND_EXECUTION]: The agent is directed to use shell commands to manage files and dependencies, specifically writing a temporary file to disk and then deleting it after execution.
[DATA_EXFILTRATION]: While intended for legitimate functionality, the script is designed to extract data from an external platform (Meta Ads Library) and ingest it into the agent's context. Combined with the requirement for API tokens, this pattern mimics data staging techniques.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data scraped from the Meta Ads Library.
Ingestion points: Data is fetched via an Apify actor and parsed into the agent's context (SKILL.md).
Boundary markers: None identified; the agent is told to use the scraped 'insights' directly for strategy generation.
Capability inventory: The skill has file-write capabilities, network access, and the ability to execute JavaScript code.
Sanitization: There is no evidence of sanitization or filtering of the scraped content before it is processed by the LLM.

performance-marketing-analysis