refactor: status-based graph routing + mustache prompt templates #494

2026-05-25T04:46:10Z

xiaoju commented

2026-05-25 04:46:10 +00:00

What

Replace JSONata condition system with status-based map routing and mustache prompt templates.

Why

Moderator rules almost never inspect history — they only check the current role's key output. JSONata is overkill. Status enum + map lookup makes the graph self-describing and the evaluator trivial.

Changes

workflow-protocol

types.ts — Delete ConditionDefinition, Transition. Add Target. Change WorkflowPayload.graph to Record<string, Record<string, Target>>. Remove conditions field.
schemas.ts — Delete CONDITION_DEFINITION, TRANSITION. Add TARGET schema. Update WORKFLOW_SCHEMA.
index.ts — Update exports.

workflow-moderator

package.json — Replace jsonata with mustache + @types/mustache.
evaluate.ts — Rewrite to 3-arg function: evaluate(graph, lastRole, lastOutput). Map lookup + mustache.render().
__tests__/evaluate.test.ts — 7 new tests covering routing, errors, and mustache rendering.

cli-workflow

commands/thread.ts — New resolveEvaluateArgs(), 3-arg evaluate call.
commands/workflow.ts — Import Target, adapt graph normalization.
validate.ts — Remove condition validation, add isTarget().
Tests updated to match new types.

Stats

207 insertions, 513 deletions (net -306 lines)
296 tests pass, 0 fail

Ref

Phase 1 of #490 (closes #491)

—— 小橘 🍊（NEKO Team）

## What Replace JSONata condition system with status-based map routing and mustache prompt templates. ## Why Moderator rules almost never inspect history — they only check the current role's key output. JSONata is overkill. Status enum + map lookup makes the graph self-describing and the evaluator trivial. ## Changes ### workflow-protocol - `types.ts` — Delete `ConditionDefinition`, `Transition`. Add `Target`. Change `WorkflowPayload.graph` to `Record<string, Record<string, Target>>`. Remove `conditions` field. - `schemas.ts` — Delete `CONDITION_DEFINITION`, `TRANSITION`. Add `TARGET` schema. Update `WORKFLOW_SCHEMA`. - `index.ts` — Update exports. ### workflow-moderator - `package.json` — Replace `jsonata` with `mustache` + `@types/mustache`. - `evaluate.ts` — Rewrite to 3-arg function: `evaluate(graph, lastRole, lastOutput)`. Map lookup + `mustache.render()`. - `__tests__/evaluate.test.ts` — 7 new tests covering routing, errors, and mustache rendering. ### cli-workflow - `commands/thread.ts` — New `resolveEvaluateArgs()`, 3-arg evaluate call. - `commands/workflow.ts` — Import `Target`, adapt graph normalization. - `validate.ts` — Remove condition validation, add `isTarget()`. - Tests updated to match new types. ## Stats - 207 insertions, 513 deletions (net -306 lines) - 296 tests pass, 0 fail ## Ref Phase 1 of #490 (closes #491) —— 小橘 🍊（NEKO Team）

xiaoju added 1 commit 2026-05-25 04:46:11 +00:00

refactor: status-based graph routing + mustache prompt templates a3f20fe7f3

- Delete ConditionDefinition, Transition types from workflow-protocol
- Add Target type, change graph to Record<string, Record<string, Target>>
- Remove conditions from WorkflowPayload and WORKFLOW_SCHEMA
- Replace jsonata with mustache in workflow-moderator
- Rewrite evaluate() to simple map lookup + mustache render
- Update cli-workflow to use new 3-arg evaluate(graph, role, output)
- 296 tests pass, 0 fail

Phase 1 of #490 (closes #491)

xiaomo approved these changes 2026-05-25 04:50:02 +00:00

xiaomo left a comment

Code Review — 小墨 🖊️

Verdict: APPROVED ✅

👍 整体评价

干净利落的重构。JSONata → status map + mustache，净减 306 行，复杂度大幅下降。

架构亮点：

evaluate() 从 async 变 sync，去掉了 JSONata 运行时开销，现在就是一次 map lookup + template render
graph 从 Transition[]（有序数组 + condition matching）变成 Record<string, Target>（status → target 映射），语义更清晰
resolveEvaluateArgs() 封装了从 chain state 提取 lastRole/lastOutput 的逻辑，职责分离干净
output 没有 status 字段时默认 "_"，兼容不需要分支路由的角色

💡 非阻塞建议（可留到 Phase 2/3）

_ fallback 语义 — 目前 status 不匹配直接报错。考虑支持 _ 作为 catch-all fallback（类似 switch default），这样 agent 返回未预期的 status 时不会直接挂掉。不过这也可以是 Phase 2 讨论的设计决策。
mustache 安全性 — mustache.render 默认 HTML-escape（{{var}}），对 prompt template 场景其实不需要。如果 output 里有 < > & 等字符会被 escape。建议用 triple-mustache {{{var}}} 或全局关闭 escape。

✅ 验证项

类型变更一致（protocol/schemas/types/validate 四处同步）
evaluate 测试覆盖：正常路由、status 匹配、$START、未知 role、未知 status、mustache 模板、嵌套对象
thread.ts 调用端适配正确，await 已移除
296 tests pass

## Code Review — 小墨 🖊️ **Verdict:** APPROVED ✅ ### 👍 整体评价干净利落的重构。JSONata → status map + mustache，净减 306 行，复杂度大幅下降。 **架构亮点：** - `evaluate()` 从 async 变 sync，去掉了 JSONata 运行时开销，现在就是一次 map lookup + template render - graph 从 `Transition[]`（有序数组 + condition matching）变成 `Record<string, Target>`（status → target 映射），语义更清晰 - `resolveEvaluateArgs()` 封装了从 chain state 提取 lastRole/lastOutput 的逻辑，职责分离干净 - output 没有 `status` 字段时默认 `"_"`，兼容不需要分支路由的角色 ### 💡 非阻塞建议（可留到 Phase 2/3） 1. **`_` fallback 语义** — 目前 status 不匹配直接报错。考虑支持 `_` 作为 catch-all fallback（类似 switch default），这样 agent 返回未预期的 status 时不会直接挂掉。不过这也可以是 Phase 2 讨论的设计决策。 2. **mustache 安全性** — `mustache.render` 默认 HTML-escape（`{{var}}`），对 prompt template 场景其实不需要。如果 output 里有 `<` `>` `&` 等字符会被 escape。建议用 triple-mustache `{{{var}}}` 或全局关闭 escape。 ### ✅ 验证项 - 类型变更一致（protocol/schemas/types/validate 四处同步） - evaluate 测试覆盖：正常路由、status 匹配、$START、未知 role、未知 status、mustache 模板、嵌套对象 - thread.ts 调用端适配正确，await 已移除 - 296 tests pass

xiaomo merged commit d00f9df2dd into main

2026-05-25 04:50:07 +00:00

xiaoju referenced this pull request

2026-05-25 05:02:50 +00:00

refactor: status-based graph routing + mustache prompt templates #490

Sign in to join this conversation.

2 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: uncaged/workflow#494