The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads its runtime configuration (runtime.yaml), strategy logic (condor-scanner.py), and utility scripts (condor_config.py) from the vendor's GitHub repository at github.com/Senpi-ai/senpi-skills during the installation process.
[REMOTE_CODE_EXECUTION]: The installation process involves downloading and executing Python scripts and a YAML configuration, which is the intended operational model for this trading skill.
[COMMAND_EXECUTION]: The script uses subprocess.run to call the mcporter CLI tool, which handles communications with the Hyperliquid exchange for fetching data and placing trades.
[DATA_EXPOSURE]: In scripts/condor_config.py, the SKILL_DIR variable is hardcoded to mantis-strategy, which contradicts the skill's name (condor-strategy) and the installation path provided in SKILL.md. This results in the skill reading from and writing to the directory of a different strategy, potentially leading to state collision.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external market data.
Ingestion points: Data enters the agent context through market_list_instruments and leaderboard_get_markets calls in condor-scanner.py.
Boundary markers: None (data is processed as structured JSON, not natural language prompts).
Capability inventory: Subprocess execution (subprocess.run) and file system writes (atomic_write) are available to the script.
Sanitization: The script uses json.loads for parsing and enforces numeric types via safe_float, which prevents the data from being interpreted as executable instructions.

condor-strategy