agent-alignment

The interpretive layer above memory. Turns text into a portable specification of how a specific person interprets information, decides, and communicates.

Agent Knowledge Cycle (AKC) — a knowledge cycle for AI agents: agent behavior compounds, human judgment sharpens. Six phases keep behavior aligned with operator intent as human attention becomes the scarce resource. ADRs, JSON schemas, dependency-free Python reference.

Field notes left by one agent for the next: staying in scope, impossible tasks, asking a human, and working well alongside other agents.

Agent 意图对齐器：通过显式 /align 将粗糙请求转为可执行、可验证、可跨会话交接的 Agent Brief。

MetricMotive verifies whether autonomous agents hit the metric while still honoring the user's real motive and guardrails.

Interactive decision report skill for Claude Code — bridges the cognition gap between user and agent with structured JSON reports, selectable options, Mermaid diagrams, and export to Markdown/JSON.

An alignment layer for AI agents. Watches whether an agent's actions still match the task, rules and evidence it was given: beside local coding agents like Claude Code, and across production agent traces. One engine, two surfaces. The CI eval gate is the first working component.