The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents several high-impact shell commands intended for emergency intervention. These include forced process termination (kill -9), deletion of database keys (redis-cli del), and recursive file removal (rm agents/memory/working/*). While these align with the Site Reliability Engineer role defined in the skill, they represent a significant capability for an autonomous agent.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from external logs and history, creating a potential surface for indirect prompt injection.
Ingestion points: The agent is instructed to read AUDIT_LOG.md and the last 10 steps of its history to determine system state.
Boundary markers: The skill does not define specific delimiters for the audit log content, although it recommends hardcoded thresholds for circuit breakers to prevent dynamic logic manipulation.
Capability inventory: The skill possesses capabilities for process management (kill), file system modification (rm), and database interaction (redis-cli).
Sanitization: The skill includes instructions to natively reject override prompts found in memory (e.g., [SYSTEM OVERRIDE]), though it lacks specific sanitization for the content of the AUDIT_LOG.md file.

production-health