The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and synthesize untrusted data from X (Twitter).
Ingestion points: External tweet content enters the agent context through the search and read functions in lib/api.ts and ct-search.ts.
Boundary markers: The skill uses trust labels like [WATCHLIST] or [SUSPICIOUS] and separators in its output formatting (lib/format.ts), but these provide limited protection against adversarial input.
Capability inventory: The skill can execute local shell commands and perform network operations to the X API and its own hosted API endpoints.
Sanitization: While it applies noise filters to remove retweets and common spam phrases (lib/filters.ts), it does not sanitize the core tweet text for potential prompt injection instructions.
[COMMAND_EXECUTION]: The installation and setup scripts (install.sh, install.ts, setup.ts) execute several system commands, including Bun.spawn, mkdir, and cp, to configure the environment and install the skill file.
[EXTERNAL_DOWNLOADS]: The skill downloads resources during setup:
Fetches the Bun runtime installer from bun.sh.
Fetches the install.sh script from the author's official GitHub repository (github.com/sendaifun/ct-alpha).
[REMOTE_CODE_EXECUTION]: The installer uses the curl | bash pattern to install Bun and its own setup script. While these are from established or vendor-owned sources, this pattern is a standard high-risk execution vector.

ct-alpha