The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest untrusted natural language input (e.g., strategy descriptions, social media content) to generate executable Python scripts. This creates a surface for indirect prompt injection where malicious input could attempt to manipulate the generated code's logic.
Ingestion points: SKILL.md (Step 1: Understand the Strategy).
Boundary markers: Not explicitly defined in the generation workflow.
Capability inventory: Generated scripts can perform network requests using urllib and execute financial transactions via the Simmer SDK's trade methods.
Sanitization: No explicit sanitization or pre-validation of the strategy description text is required of the agent before code generation, though a post-generation validation script is provided.
[EXTERNAL_DOWNLOADS]: The skill workflow involves using npx clawhub@latest to publish the generated skills. This command downloads and executes the latest version of the publishing tool from the npm registry, which is a standard part of the platform's developer workflow.
[DATA_EXFILTRATION]: The skill and its generated templates are designed to access the SIMMER_API_KEY from environment variables for platform authentication. The provided scripts/status.py utility correctly transmits this key to the official platform endpoint (api.simmer.markets) and encourages secure handling practices over hardcoding.
[COMMAND_EXECUTION]: The skill provides utility scripts for account management and skill validation, and it instructs the agent to execute shell commands for publishing to the platform's registry.

simmer-skill-builder