The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill assists the user in authoring Python scripts (scan.py, scoring.py) and configuration files (runtime.yaml, strategy.yaml). It then executes these dynamically generated scripts through local validation and deployment tools (openclaw senpi validate). This process of script generation and subsequent execution is a central feature that requires users to trust the agent's code generation capabilities.\n- [INDIRECT_PROMPT_INJECTION]: The skill relies on natural language descriptions (the 'thesis') to derive trading logic and sizing. This creates an indirect prompt injection surface where adversarial user input could attempt to influence the generated code or bypass the intended guardrails, although the multi-step interview and validation process mitigate this risk.\n
Ingestion points: User-provided strategy descriptions and choices during the 7-step design interview in SKILL.md.\n
Boundary markers: The structured 7-step interview provides natural logical boundaries for individual inputs.\n
Capability inventory: The skill can write files to the /data/workspace/strategies/ directory and execute shell commands to run Python scripts and the openclaw CLI tool.\n
Sanitization: The agent is instructed to use ast.parse for syntax checking and must achieve a 'PASS' result from senpi validate (which runs a read-only simulated tick) before deployment is permitted.

senpi-strategy-author