The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data such as market questions, outcome names, and news feeds to generate trading advice, creating a vulnerability surface for adversarial influence on the agent's reasoning.
Ingestion points: Market data is fetched via Gamma and CLOB APIs in scripts/advisor.py and scripts/backtest.py. The strategy methodology in SKILL.md also involves analysis of external news sources for news-driven edges.
Boundary markers: The instructions do not define explicit delimiters or use "ignore embedded instructions" warnings for the external market data or news content being processed.
Capability inventory: The skill utilizes requests for network access, interacts with a local SQLite database, and executes Python scripts that generate natural language trade recommendations.
Sanitization: While the scripts perform type casting (e.g., converting market prices to floats), there is no evidence of text-based sanitization or validation for descriptions and news content ingested into the agent context.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands and local Python scripts to perform its primary functions.
Evidence: SKILL.md contains multiple bash code blocks directing the agent to activate virtual environments and run scripts such as scan_markets.py, advisor.py, and backtest.py located within the skill directory.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to retrieve market and orderbook data from external API endpoints.
Evidence: scripts/advisor.py and scripts/backtest.py communicate with gamma-api.polymarket.com and clob-api.polymarket.com. These are recognized as well-known service endpoints associated with the Polymarket platform and are documented neutrally as necessary for the skill's functionality.

polymarket-strategy-advisor