Route agent tasks through confidence-based arbitration to select execution modes.
npx skills add https://github.com/songlairui/agent-monad --skill sancai-zhen
Please help me install this Agent Skill. Skill: sancai-zhen Source: https://github.com/songlairui/agent-monad/tree/main/skills/sancai-zhen Command: npx skills add https://github.com/songlairui/agent-monad --skill sancai-zhen
This Skill removes rigid step-by-step prompting from agent work by routing tasks internally based on confidence, so clear requests can move straight into execution while vague requests naturally get the extra questions they need.
Use sancai-zhen to interpret the request, build the consensus page if needed, and start the task with the right balance of clarification and autonomous execution.
High-intent search queries and answers about installing and using this skill.
Confidence-based agent routing evaluates task clarity against thresholds to automatically choose between seed, panorama, execution, and exception handling modes without exposing the route. It moves clear requests straight to execution while guiding vague requests through clarification.
The best way to manage ambiguous agent context is building a consensus page that models intent, outcome, current state, boundaries, and risks before acting. This shared understanding allows the agent to confirm only uncertain parts before autonomous execution.
You reduce unnecessary back-and-forth by applying confidence thresholds that let the agent continue working with progress updates when confidence is high, pausing only for missing information, contradictions, or irreversible actions during autonomous execution.
Yes, you can use internal arbitration to handle blocked work and feedback-driven iteration. The routing mechanism applies exception handling to blocked tasks and iterates based on feedback while maintaining graph and session state updates.
You do not need a consensus page for every agent task. It is primarily built for ambiguous ideas and vague requests to establish shared understanding, while clear implementation requests can bypass extensive clarification and move straight into execution.
A limitation of using confidence thresholds for state updates is that the agent may pause execution if it detects contradictions or missing information, requiring human intervention before it can safely proceed with irreversible actions.
songlairuiSeed ideas, render panoramas, and compress graph state in persistent JSON.
tcytanCoordinates plan-and-execute issue workflows with session tracking and verification.
masharrattCoordinate container-based CFN Loop orchestration across implementers, validators, and product owner flows.
pazdedavDefine DAG workflow graphs with JSON schema for agent orchestration.
