What problem does it solve? Complex multi-session agent work often fails due to lost context, unverified sub-agent claims, self-review blind spots, and unsafe operations. This Skill provides the operational discipline—slash commands, evidence standards, approval boundaries, and delegation rules—needed to keep long-running Hermes Agent tasks correct and verifiable. ## Core Features & Use Cases - Slash Command Reference: Defines when to use /new, /compress, /snapshot, /rollback, /goal, /background, /steer, and /queue for context and session management. - Evidence Ladder & Verification: Ranks evidence strength from passing tests down to unverified user reports, and requires independent verification of all sub-agent claims. - Sub-Agent Isolation Rules: Enforces reviewer-not-implementer and fixer-not-reviewer separation, with a maximum of two automated fix rounds before escalating to the user. - Use Case: Before starting a large refactoring across multiple sessions, load this Skill to snapshot state, delegate independent reviews via delegate_task, and verify every sub-agent's "tests passed" claim by running the commands yourself. ## Quick Start Load this Skill before starting a complex multi-session task and ask the agent to follow its evidence standards, context reset rules, and delegation boundaries throughout the work.
