The Agent Skills Directory

[SAFE]: The skill shows no signs of malicious behavior. The code is well-structured, modular, and performs the functions described in its documentation without attempting to hide intent or exfiltrate sensitive data to unauthorized third parties.
[CREDENTIALS_UNSAFE]: The skill handles an X API Bearer Token. It follows security best practices by prompting the user for the token during setup and storing it in a local environment file (~/.config/env/global.env). This approach avoids hardcoding secrets and is a standard way to manage API credentials for local development tools.
[COMMAND_EXECUTION]: The installation and setup scripts (install.ts, setup.ts) utilize Bun.spawn and native file system modules to configure the skill and run verification tests. These operations are limited to the user's home directory configuration and the skill's own directory, representing typical installation behavior for a CLI-based agent skill.
[INDIRECT_PROMPT_INJECTION]: The skill functions by ingesting and processing untrusted content from X/Twitter, which creates a potential surface for indirect prompt injection.
Ingestion points: Raw tweet text and metadata are ingested in lib/api.ts through the search and thread hydration features.
Boundary markers: Content is formatted into structured text blocks in lib/format.ts without specific escaping or boundary markers like XML tags.
Capability inventory: The skill can execute Twitter searches and suggest the use of other tools (DeFi Llama, Backpack, Polymarket) based on the findings.
Sanitization: The skill applies log-scaled engagement filters and author credibility checks to prioritize high-signal content and flags suspicious patterns (e.g., bot-like engagement) to help the agent distinguish between noise and quality information. It also includes explicit warnings that contract addresses are unverified.

ct-alpha