The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection via its reasoning loop integration.
Ingestion points: The background subagent polls the local hub's bridge at http://127.0.0.1:8765/api/agent/next for task kind and payload data (SKILL.md).
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are used when processing the API response. Mitigation relies on the agent's 'Iron Rule' to verify all cards against oracle text.
Capability inventory: The agent is empowered to export deck data to the filesystem (deck.json), execute other skills (/deck-strat, /lgs-search), and run shell commands.
Sanitization: The instructions do not specify any validation or sanitization of the payloads retrieved from the local bridge.
[DYNAMIC_EXECUTION]: The skill's operational instructions require the agent to execute dynamically generated Python code via python -c to communicate with the local server and parse JSON responses.
[COMMAND_EXECUTION]: The skill setup and background heartbeat loop involve executing various shell commands (uv sync, uv run, curl) within the user's terminal environment.

deck-forge