跳转至

AI Agent Integration

Designed from the ground up for autonomous AI agents and developer workflows. Integrate via any of the three patterns below:

1. Instruction-driven agents via AGENTS.md (ZCode, DSH, Codex, Claude Code, Cursor, etc.)

Clone and run directly. The repository root AGENTS.md specifies behavioral boundaries, entry points, and verification procedures. Agents read this file automatically on session initialization and invoke the westlake CLI directly once credentials are configured.

Direct in-process function calls provide the lowest latency and strongly typed data models:

import westlake_survival
from westlake_survival.sis import get_schedule

for session in get_schedule():
    print(session.weekday, session.start_time, session.title)

Runnable examples are provided in examples/. When installed via pip install westlake-survival, the library runs without requiring a local repository clone.

3. Skill Mechanism

The standalone skill definition is located at skills/westlake-survival/SKILL.md. Copy it into your agent's skills directory:

cp -r westlake_survival/skills/westlake-survival ~/.zcode/skills/   # or ~/.agents/skills/, ~/.claude/skills/

The westlake-mcp server serves as a compatibility layer for clients restricted to tool-calling interfaces (such as Claude Desktop or chat sidebars). Agents with shell or Python execution access should prefer the CLI or Python API.

Write Safety & Consequence Gate

Exactly five write operations exist across the entire toolkit: canvas submit, canvas reply, canvas send, booking book, and booking cancel. All write paths are protected by consequence.py:

  • Mandatory --yes flag: Without --yes, execution terminates immediately with exit code 1, outputting a structured risk descriptor (action target, risk category, severity rating, and reversibility).
  • Pre-flight simulation with --dry-run: Supported on submit and book to preview request payloads and constraints without altering remote state.
  • Non-interactive by design: Interactive [y/N] prompts are intentionally omitted to prevent agent pipelines from hanging on blocked stdin buffers.

Initialization & Health Check

westlake doctor --json    # inspects environment, dependencies, and network health

Ensure westlake doctor passes before executing multi-step tasks. AI agents must never handle raw credentials directly; users configure authentication once via the local web Settings UI.