The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill automates the setup of its execution environment by installing the Bun runtime if it is missing.
Evidence: Executes curl -fsSL https://bun.sh/install | bash within the SKILL.md preflight steps and onboard.ts script. This is a standard installation method for the runtime from its official domain.
[EXTERNAL_DOWNLOADS]: Fetches required tools and metadata from well-known services.
Evidence: Downloads dependencies from bun.sh and fetches data from established platforms including YouTube and X (Twitter) to perform extraction.
[DATA_EXFILTRATION]: Transmits extracted content and trade data to the service's primary domain.
Evidence: Sends source transcripts, metadata, and trade parameters to the paste.trade API. This behavior is documented and necessary for the skill's primary function of tracking market claims.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface due to its handling of untrusted web data.
Ingestion points: untrusted data enters via URL extraction in extract.ts and diarize.ts.
Boundary markers: markers are generally absent when the agent reads the extracted source text.
Capability inventory: the skill can execute shell commands via the Bash tool, perform network operations using fetch(), and write to the local filesystem using Bun.write().
Sanitization: extraction results are subjected to strict JSON schema validation in validate.ts before being saved or posted to the backend.

trade