The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow utilizes standard engineering tools including git, curl, ripgrep (rg), and jq to manage version control, interact with infrastructure APIs, and verify file contents. These activities are transparently documented as core functions of the various agent roles.\n- [PROMPT_INJECTION]: The multi-agent loop relies on reading markdown handoff files, creating a surface for indirect prompt injection. However, the system is designed to mitigate this risk through a human-led discovery phase to establish goals and a 'Reviewer' role that is mandated to be 'ruthless to claims' and perform independent evidence-based verification.\n- [DATA_EXFILTRATION]: While the skill uses network-capable tools like curl and POST for API interactions (e.g., updating Grafana dashboards), it includes explicit instructions for the Reviewer to redact secrets discovered in handoff files, demonstrating a commitment to secure data handling practices.

handoff-iteration-loop