Skip to content

API.md

Latest commit

 

History

History
77 lines (50 loc) · 4.35 KB

API.md

File metadata and controls

77 lines (50 loc) · 4.35 KB
Raw

App Server API

默认本地服务：

Default local server:

http://127.0.0.1:7345

核心端点 / Core Endpoints

Method Path 用途 / Purpose
GET /health 健康与运行时状态 / Health and runtime status
POST /sessions 创建 Agent 会话 / Create an Agent session
GET /sessions/{id}/events SSE 事件流 / SSE event stream
POST /sessions/{id}/cancel 取消会话 / Cancel a session
POST /config/llm 更新 LLM 配置 / Update LLM configuration
POST /llm/probe 探测已配置的 LLM Provider / Probe configured LLM provider

运行时证据 / Runtime Evidence

Method Path 用途 / Purpose
GET /runtime/readiness 无网络运行时就绪 gate / No-network runtime readiness gate
GET /runtime/portable portable 环境诊断 / Portable environment diagnostics
GET /sessions/{id}/evidence?steps=20 紧凑运行时证据包 / Compact runtime evidence bundle
GET /agent/onboarding?session_id=...&steps=20 外部 Agent onboarding bundle / External Agent onboarding bundle

Maker 集成 / Maker Integration

Method Path 用途 / Purpose
GET /maker/setup-status Maker 设置状态 / Maker setup status
GET /maker/tool-audit Maker 远端/本地工具审计 / Maker remote/local tool audit
POST /maker/repair 在支持时修复 Maker 接入 / Repair Maker access where supported
POST /maker/repair-default-project 修复默认 Maker 项目 authority / Repair default Maker project authority
POST /maker/build-smoke/run 通过受控 Executor 跑 Maker build smoke / Run guarded Maker build smoke

Provider 调用证明 / Provider Proof

使用 /llm/probe 验证已配置的 provider 是否真的被调用。MiniMax 应观察到 /text/chatcompletion_v2；OpenAI-compatible providers 应观察到 /chat/completions；Claude-style providers 应观察到 /messages。

Use /llm/probe to verify that a configured provider is actually called. MiniMax should observe /text/chatcompletion_v2; OpenAI-compatible providers should observe /chat/completions; Claude-style providers should observe /messages.

Agent 内部 API (Python) / Agent Internal API (Python)

这些是 agent 进程内可被新代码直接 import 的公共 API。文档接口稳定；具体行为以源文件 docstring 为准。

These are public Python APIs that new agent code can import directly. The contract is stable; see the per-module docstring for behavior.

Multimodal LLM / Multimodal LLM

  • from llm.content import ContentBlock, TextBlock, ImageBlock — pass text or image to a provider. to_anthropic_messages(blocks) / to_openai_messages(blocks) translate for the wire.
  • from llm.interface import LLMInterface — providers opt into multimodal via supports_multimodal = True and override think_multimodal.

Project introspection / Project introspection

  • from agent.project_introspection import ProjectIntrospector, register_introspection_tools — six read-only tools (project.manifest, project.asset_read, project.asset_search, project.code_search, project.preview_capture, project.skill_pack).

Skill packs / Skill packs

  • from agent.skill_packs import SkillPackRegistry, SkillPackStorage, SkillPack — project-side knowledge base. get_or_create_registry(project_root) seeds defaults.
  • from agent.skill_packs.bootstrap import bootstrap_default_packs, get_or_create_registry — seed and lookup.

Typed sub-goal DAG / Typed sub-goal DAG

  • from agent.goal_dag import SubGoalSpec, SubGoalType, GoalDAGScheduler, topological_layers — typed parallel sub-goal execution with per-type sub-loops in agent.typed_subloop.

Feature / ticket state machine / Feature / ticket state machine

  • from agent.feature_state import FeatureLedger, FeatureState, Feature, render_progress_md — append-only ledger at <artifacts_root>/.ttmevolve/features.jsonl.

Goal loop artifact isolation / Goal loop artifact isolation

GoalLoop.__init__ accepts an optional artifacts_root parameter. Production code passes nothing (artifacts live under the project root); tests pass artifacts_root=tmp_path or set the TTMEVOLVE_GOAL_ARTIFACTS_ROOT env var. See CONTRIBUTING.md for the dev-vs-artifact isolation rule.