Run reproducible OpenCorvus harness experiments against AutomationBench, preserve paper-grade evidence for every attempt, invalidate bug-affected runs, and maintain a self-owned leaderboard. Use for AutomationBench runs, reruns, score cataloging,…
Run reproducible OpenCorvus Mission/Base experiments against Tencent WorkBuddy Bench, require the runtime Code Skill, preserve official Harbor grading and paper evidence, and exclude bug-affected attempts. Use for WorkBuddy Bench setup, chain proofs, runs,…
Execute Advanced software delivery through requirements-to-architecture traceability, safe parallel evidence frontiers, exact implementation ownership, and independent acceptance.
Install, configure, verify, operate, and troubleshoot OpenCorvus from an assistant such as Hermes Agent, OpenClaw, Codex, or another Agent Skills-compatible host. Use when the user asks to install OpenCorvus, connect a model provider, run OpenCorvus locally,…
Deliver a concrete outcome with capability-matched execution and independent verification at proportional cost.
Coordinate an operator-defined Mission through native Mission state and fixed-profile Tasks without imposing a domain-specific workflow.
Author one portable, project-owned and traceable OpenCorvus Expert Squad through the Generate Expert Squads workflow and canonical SDK writer.
Agent-led heterogeneous Squad discovery, evidence review, blocker repair, lossless Skill-directory and portable remote MCP preservation, and canonical import into OpenCorvus.