The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted content from Twitter/X (tweets, profiles, and trends), presenting a significant surface for Indirect Prompt Injection attacks.\n
Ingestion points: Methods like api.search(), api.user_tweets(), and api.trends() described in SKILL.md and references/examples.md fetch external data into the agent's context.\n
Boundary markers: The instructions do not define delimiters or "ignore embedded instructions" markers for the scraped data.\n
Capability inventory: The skill utilizes tool capabilities for network access and local file system writes (SQLite database management).\n
Sanitization: No sanitization or filtering protocols are specified for external content before processing by the agent.\n- [EXTERNAL_DOWNLOADS]: The skill documentation encourages installing the twscrape library from a non-trusted third-party GitHub repository.\n
Evidence: pip install git+https://github.com/vladkens/twscrape.git in references/installation.md.\n- [CREDENTIALS_UNSAFE]: The skill's account management workflow involves handling sensitive Twitter logins, email passwords, and session cookies in plain text.\n
Evidence: Examples in SKILL.md and references/examples.md demonstrate adding accounts via api.pool.add_account with credentials and cookies.\n- [PROMPT_INJECTION]: The skill-report.json file contains a self-referential security audit claiming the skill is safe. This metadata poisoning is intended to override the verdict of automated security scanners.

twscrape