The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted user data which is subsequently forwarded to a high-privilege remote environment.
Ingestion points: User-provided "Prompt text" and "attachments" defined in SKILL.md.
Boundary markers: Use of Markdown headers (## Task, ## Context) is encouraged to structure the generated payload.
Capability inventory: The downstream T3 environment executes in full-access mode, which includes file system modification and code execution capabilities.
Sanitization: No explicit sanitization or filtering of user-supplied prompt content is specified before transmission.
[COMMAND_EXECUTION]: The skill relies on local script execution and system tools to manage its workflow.
Evidence: Execution of python3 ~/.agents/skills/t3code/scripts/launch_thread.py and git worktree list --porcelain for environment setup and verification.
[DATA_EXFILTRATION]: The skill transmits project data and instructions to a remote endpoint using authentication tokens.
Evidence: WebSocket RPC commands (thread.turn.start) are sent to a configurable httpBaseUrl with bearer authorization.

t3code-thread