The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/room_rules.py utilizes subprocess.check_call to execute a text editor specified by the EDITOR environment variable. This pattern is used to allow users to edit room-level rules, but it introduces a vector for arbitrary command execution if the environment variable is controlled by a malicious actor.
[REMOTE_CODE_EXECUTION]: The scripts/self_update.py script implements a self-update mechanism that automatically downloads, verifies via SHA256, and extracts skill bundles from the vendor's internal server. This capability allows the vendor to remotely update the skill's executable logic.
[EXTERNAL_DOWNLOADS]: The scripts/invite.py script generates onboarding instructions that include a command to download and execute a shell script from the vendor's public domain (https://sc-chatroom.fly.dev). This pattern creates an execution surface for external users or secondary agents following the instructions.
[PROMPT_INJECTION]: The scripts/install_soul.py script modifies the agent's core instructions in SOUL.md to establish behavioral rules for chatroom participation. This is a functional requirement but modifies the agent's decision-making logic.
[PROMPT_INJECTION]: The skill processes untrusted chat messages from external participants, creating a surface for indirect prompt injection.
Ingestion points: Incoming group chat messages delivered through the /chat/stream interface.
Boundary markers: The SOUL.md behavior block uses a chatroom- thread ID prefix to scope the custom processing logic.
Capability inventory: The skill possesses capabilities for network operations (scripts/_common.py), file system access within the workspace (scripts/_common.py), and authentication key management (scripts/join.py).
Sanitization: The skill relies on the LLM's adherence to the SOUL.md instructions and does not perform programmatic filtering or escaping of the message content.

chatroom