The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs the agent to use the execute_code tool to fetch positions and watchlists from a connected brokerage. This involves generating and running code at runtime to interact with external APIs.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant ingestion surface for untrusted data that is later persisted into long-term memory files.
Ingestion points: Data is fetched from external brokerage accounts (positions, watchlists) and user responses via AskUserQuestion (markets, focus, style, risk, preferences).
Capability inventory: The skill uses execute_code, writes to .agents/user/profile/*.json, and manages workspaces via manage_workspaces.
Boundary markers: The skill refers the agent to .agents/user/profile/README.md for schema validation and refusal reasons before writing, which acts as a security boundary.
Sanitization: The agent is instructed to map positions to a specific instrument_type schema and carry specific numeric/currency fields, providing some structural sanitization of external data.
[COMMAND_EXECUTION]: The skill utilizes execute_code and manage_workspaces, which are high-privilege tools capable of influencing the agent's operating environment and persistent state.

onboarding