The Agent Skills Directory

[CREDENTIALS_UNSAFE]: A hardcoded RapidAPI key was discovered in the configuration for the Amazon data API. This exposes credentials to any user of the skill and violates secure secret management practices.
Evidence: 'x-rapidapi-key': '4e842e5834msh0fff10888eda799p121fefjsn8712ea690aee' found in 'apis/realtime-amazon-data/config.json'.
[DATA_EXFILTRATION]: The skill includes a robust suite of utility functions designed to transmit retrieved data to external third-party services.
The 'lib/utils.js' file contains dedicated logic for pushing data to arbitrary Webhook URLs, AWS S3 buckets, FTP servers, Airtable, and Google Sheets.
While these are functional features, they provide a pre-built mechanism for automated data exfiltration if the agent is compromised or misled.
[PROMPT_INJECTION]: The skill possesses a high vulnerability surface for indirect prompt injection due to its core architecture.
Ingestion points: The skill ingests raw, untrusted data from dozens of public sources including Reddit, Quora, and Google Search results (referenced in 'SKILL.md').
Boundary markers: No delimiters or protective instructions are used when interpolating this untrusted data into the agent's context.
Capability inventory: The agent is granted powerful local capabilities including 'Bash' for shell execution and 'Write/Edit' for file modification.
Sanitization: There is no evidence of content sanitization or instruction filtering before the data is processed by the AI.
[COMMAND_EXECUTION]: The skill is designed to execute local Node.js scraper scripts via the shell to perform its primary tasks.
Instructions in 'SKILL.md' direct the agent to use the 'Bash' tool to run commands such as 'node --env-file=.env apis/{api_id}/scrape.js'.
[EXTERNAL_DOWNLOADS]: The skill performs numerous network operations to fetch data from various external API domains.
Primary connections include 'api.openwebninja.com' and 'rapidapi.com' (documented in 'meta.json' files).
The skill also supports connections to generic webhooks, AWS S3, and Slack endpoints for data delivery.

openwebninja_universal_scraper