Files
marqdo/examples/agent-okf-flywheel/index.mq.md
T
cflmy 9331af78e2 release(2.8.0): 合入 CHANGELOG/README/发版说明/skills prove 任务;CI 链接指向 doc/releases
- CHANGELOG.md:v2.8.0 条目(中英双语);
- README.md:版本行 2.7.1→2.8.0 + 愿景第 6 条(知识可证明);
- doc/releases/v2.8.0.md:双语发版说明(亮点/用法/已知限制);
- skills/marqdo-dev/index.mq.md:新增 prove 任务(路径/目录模式/
  require_lean/json;参数用 + 行声明升参,check 与 narrative-ratio 通过);
- 设计文档 §12.5 标记已合入;发版链接统一指向 doc/releases/v2.8.0.md。
2026-11-18 10:29:26 +08:00

3.6 KiB
Executable File

markup, title, description, import
markup title description import
occlusion agent-okf-flywheel (Wave B0) Promote solidified workbook; second plan is cache=hit — document as knowledge base (OKF).
llm agent fs json plugin sys
ext/ai/llm.mq.md ext/ai/agent.mq.md lib/fs.mq.md lib/json.mq.md lib/plugin.mq.md lib/sys.mq.md

main

返回 类型 说明
result any None; execute this page's documented steps in order and observe their output or stated effects

This is an offline knowledge-reuse workflow, not a request to an external model. It loads the agent plugin, promotes the visible workbook for one exact goal and asks the same agent goal again. The expected hit comes from the promoted knowledge base. The dummy key only satisfies model setup; it is not a credential or evidence that a model call succeeded.

The example uses its own named files under .marqdo and replaces that example knowledge base on rerun. Grant the actual plugin and filesystem effects in a disposable example root. Promotion and execution are separate outcomes: inspect promoted first, then cache, match and the returned result. A hit marker alone does not prove that an unrelated new goal is safe to reuse.

OKF flywheel offline: promote → plan same goal → cache=hit (no live LLM required for the hit path).

We first p = Compute p for the next step.

  1. p We now [Call plugin.load](> plugin.load path=p).

We now [Display this step's result](> print text="no-agent-plugin"). We now [Call sys.exit](> sys.exit code=1).

We now [Call sys.env set](> sys.env_set name="OPENAI_API_KEY" value="offline-okf-dummy"). We first model = Compute model for the next step. We first tools = Compute tools for the next step. We first 助手 = [Compute 助手](agent.agent(model=model, tools=tools, standing="offline okf flywheel")) for the next step.

We first kb = ".marqdo/agent-kb-example-flywheel" for the next step. We first wb = ".marqdo/agent-runs/example-pong-solid.mq.md" for the next step. We first goal = "Reply with exactly the word pong and nothing else." for the next step. We first body = "---\ntitle: okf pong\n---\n\n# main\n\nmsg = "pong"\n\nmsg**\n"** for the next step.

  1. fs.exists path=kb We now [Call fs.remove](> fs.remove path=kb).


We now [Call fs.write text](> fs.write_text path=wb text=body).

We first prom = [Compute prom](agent_kb_promote(kb_dir=kb, goal=goal, workbook=wb)) for the next step. We first okp = [Compute okp](json.get(value=prom, key="promoted")) for the next step.

  1. okp We now [Display this step's result](> print text="promoted").

We now [Display this step's result](> print text="promote-fail"). We now [Call sys.exit](> sys.exit code=1).

We first out = [Compute out](> 助手.plan goal=goal max_rounds=2 writeback=False kb_dir=kb explore_n=0 promote=False) for the next step. We first cache = [Compute cache](json.get(value=out, key="cache")) for the next step. We now [Display this step's result](> print text=cache). We first match = [Compute match](json.get(value=out, key="match")) for the next step. We now [Display this step's result](> print text=match). We first val = [Compute val](json.get(value=out, key="result")) for the next step. We now [Display this step's result](> print text=val).

  1. cache == "hit" We now [Display this step's result](> print text="flywheel-ok").

We now [Display this step's result](> print text="flywheel-miss"). We now [Call sys.exit](> sys.exit code=1).