The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied topics and external context which are interpolated into subagent prompts without explicit sanitization.
Ingestion points: User-provided debate topics and context processed in Workflows/Debate.md and Workflows/Quick.md, as well as configuration files read from ~/.claude/LIFEOS/USER/CUSTOMIZATIONS/SKILLS/Council/.
Boundary markers: The prompt templates in Workflows/Debate.md and Workflows/Quick.md lack explicit delimiters or instructions to ignore commands embedded within the user data.
Capability inventory: The skill spawns subagents using the Agent tool, performs local network requests via curl, and writes execution logs to the local file system.
Sanitization: No validation or escaping is applied to user-controlled strings before they are included in subagent prompts.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands for utility purposes.
Notifications: Shell commands using curl are used to send JSON payloads to http://localhost:31337/notify for status updates.
Logging: Execution metadata is appended to ~/.claude/LIFEOS/MEMORY/SKILLS/execution.jsonl using the echo command.
Input flow: Both curl and echo commands incorporate summaries or action descriptions derived from user input, which presents a risk of command injection if the agent does not properly escape the content.

Council