The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest untrusted user content, such as strategy descriptions, tweets, or external API documentation, to generate executable Python code. There is a risk that malicious input could influence the generated logic to perform unauthorized trading actions or exfiltrate data.
Ingestion points: Strategy descriptions, tweet contents, and external API documentation (SKILL.md).
Boundary markers: None specified for the untrusted input in the code generation instructions.
Capability inventory: Generated scripts have access to network operations via urllib and trading capabilities via the simmer-sdk (references/skill-template.md).
Sanitization: No explicit sanitization or validation of the input strategy text is described.
[EXTERNAL_DOWNLOADS]: The skill workflow involves downloading and executing the ClawHub CLI tool via npx clawhub@latest and requires the installation of the simmer-sdk Python package. These are established vendor resources used for publishing and core functionality.
[COMMAND_EXECUTION]: The instructions direct the agent to execute a local validation script (scripts/validate_skill.py) and the npx command to publish the generated skill to an external registry.
[DATA_EXFILTRATION]: The generated trading skills are configured to interact with external APIs, including api.simmer.markets and clob.polymarket.com. This requires the use of the SIMMER_API_KEY environment variable. While these operations are fundamental to the skill's purpose, they involve handling sensitive credentials and performing external network requests.

simmer-skill-builder