@weapp-agent/cli 0.0.0 → 0.1.0

package/package.json

@@ -1,8 +1,10 @@
 {
   "name": "@weapp-agent/cli",
-  "version": "0.0.0",
-  "description": "Bootstrap placeholder for Weapp Agent CLI trusted publishing",
   "type": "module",
+  "version": "0.1.0",
+  "private": false,
+  "description": "AI coding agent for WeChat mini-programs",
+  "author": "ice breaker <hi@sonofmagic.top>",
   "license": "MIT",
   "homepage": "https://vite.weapp.dev/guide/agent/getting-started",
   "repository": {
@@ -10,13 +12,61 @@
     "url": "git+https://github.com/weapp-vite/weapp-vite.git",
     "directory": "packages/agent-cli"
   },
+  "bugs": {
+    "url": "https://github.com/weapp-vite/weapp-vite/issues"
+  },
+  "keywords": [
+    "weapp-vite",
+    "weapp",
+    "agent"
+  ],
+  "bin": {
+    "weapp-agent": "dist/index.mjs"
+  },
   "files": [
     "README.md",
-    "LICENSE"
+    "dist",
+    "skills"
   ],
+  "engines": {
+    "node": ">=24.15.0"
+  },
+  "dependencies": {
+    "@ai-sdk/anthropic": "4.0.76",
+    "@ai-sdk/openai": "4.0.89",
+    "@ai-sdk/openai-compatible": "3.0.66",
+    "@modelcontextprotocol/client": "^2.3.1",
+    "@modelcontextprotocol/server": "^2.3.1",
+    "@weapp-vite/acceptance": "0.1.0",
+    "@weapp-vite/mcp": "2.0.0",
+    "ai": "7.0.133",
+    "commander": "15.0.0",
+    "diff": "9.0.0",
+    "execa": "10.1.0",
+    "ink": "8.0.0",
+    "ink-text-input": "6.0.0",
+    "react": "19.3.0",
+    "weapp-ide-cli": "6.2.0",
+    "zod": "4.6.5"
+  },
+  "devDependencies": {
+    "@types/react": "^19.3.0",
+    "@weapp-agent/core": "0.0.0",
+    "@weapp-agent/mini-program": "0.0.1",
+    "@weapp-agent/providers": "0.0.0",
+    "tsdown": "0.23.0"
+  },
   "publishConfig": {
     "access": "public",
-    "registry": "https://registry.npmjs.org",
-    "tag": "bootstrap"
+    "registry": "https://registry.npmjs.org"
+  },
+  "scripts": {
+    "build": "tsdown",
+    "dev": "tsx src/index.ts",
+    "typecheck": "tsc --noEmit",
+    "test": "vitest run",
+    "lint": "eslint .",
+    "prebuild": "node ../../scripts/weapp-agent/sync-skill.mjs",
+    "test:pack": "node ../../scripts/weapp-agent/smoke-pack.mjs"
   }
-}
+}

package/skills/weapp-acceptance/SKILL.md

@@ -0,0 +1,21 @@
+---
+name: weapp-acceptance
+description: Validate changes in weapp-vite native or Wevu WeChat projects with Weapp Agent build checks, deterministic DevTools scenarios, and source-bound evidence. Use after implementing a mini-program feature or investigating a failed acceptance task.
+---
+
+# Weapp acceptance
+
+Keep planning and code edits in the host agent. Weapp Agent executes checks and collects evidence; it does not need another model or API key.
+
+1. Call `weapp_project_inspect` when available, or `wv accept <project> --inspect --json`. Limit acceptance claims to supported weapp-vite native/Wevu projects on WeChat.
+2. Maintain `acceptance.scenarios` in `weapp-acceptance.config.json` for the requested behavior. Read [scenarios.md](references/scenarios.md) when creating or changing a scenario. Preserve unrelated checks and user changes. Choose assertions from the requested outcome, not from whichever output the current code produces.
+3. Start `weapp_acceptance_start`, retain its `jobId`, and query `weapp_acceptance_status` until it is no longer `running`. Use CLI `wv accept <project> --json` if MCP is unavailable; CLI waits for completion. Do not launch duplicate jobs to poll progress.
+4. Read `weapp_acceptance_report` and relevant named artifacts. Repair the cause of failures within the user's task, then run a new acceptance job. Repeated environment failures require resolving the stated prerequisite, not repeated navigation or input.
+
+Honor existing host authorization. A tool's `action_required` result does not authorize silently adding `--trust`. Explain the concrete scripts/configuration that need review. Scenario edits invalidate the stored trust decision. Never lower required checks, delete assertions, or replace actual values with expected values to obtain a pass.
+
+Reports use `version: 2`. Claim acceptance passed only when `passed` is true and `snapshot.stale` is false; state the report's `requiredChecks`. Build-only acceptance is not runtime verification. `verify --json` is the legacy command report and is not a full acceptance verdict.
+
+Report failures, missing evidence, and environment blockers distinctly. Screenshots support inspection but do not prove visual correctness. `runtime: wechat-devtools` describes the developer-tool simulator, not a physical device. Interrupted tasks are not replayed: inspect the last recorded step and current state before starting a fresh job.
+
+When reporting completion, provide the job ID, checks performed, failed/unverified categories, and relevant screenshot/log artifacts. Do not claim a runtime action occurred unless the report records it.

package/skills/weapp-acceptance/agents/openai.yaml

@@ -0,0 +1,6 @@
+interface:
+  display_name: '小程序改动验收'
+  short_description: '通过 weapp-vite 执行确定性场景并检查证据新鲜度'
+  default_prompt: 'Use $weapp-acceptance 检查项目、执行验收并读取报告，明确区分通过、失败和未验证。'
+policy:
+  allow_implicit_invocation: true

package/skills/weapp-acceptance/references/scenarios.md

@@ -0,0 +1,40 @@
+# Scenario contract (version 1)
+
+Add model-free configuration to the project, preserving existing fields:
+
+```json
+{
+  "version": 1,
+  "verification": [
+    { "kind": "build", "command": "pnpm", "args": ["run", "build"] }
+  ],
+  "acceptance": {
+    "requiredChecks": ["build", "devtools"],
+    "scenarios": ["acceptance/counter.json"],
+    "timeoutMs": 600000
+  }
+}
+```
+
+Use the project's actual package manager and scripts. Available required categories: typecheck, build, test, devtools. The default requires build and devtools; categories omitted from requirements are still reported when configured. An arbitrary command labelled devtools is not runtime proof for version 2 acceptance.
+
+```json
+{
+  "version": 1,
+  "name": "counter increments",
+  "steps": [
+    { "action": "route", "path": "/pages/agent-proof/index" },
+    { "action": "wait", "selector": "#count", "timeoutMs": 5000 },
+    { "action": "assert", "selector": "#count", "text": "0" },
+    { "action": "tap", "selector": "#increment" },
+    { "action": "assert", "selector": "#count", "text": "1" },
+    { "action": "screenshot" }
+  ]
+}
+```
+
+Each scenario starts with route (reLaunch), contains at least one assert, and has a unique name. `find`, `tap`, and `wait` take selector; `input` takes selector and string value. `assert` compares exact element text, polling for up to timeoutMs (default 5000, maximum 60000). `wait` waits for an element. No JavaScript/eval step is supported. Use real route names and selectors from the project.
+
+Navigation, input and taps run once. Only read assertions poll. Each successful scenario gets an automatic screenshot; a failed scenario attempts a screenshot. Console evidence is collected at the end when the connection remains available. Credentials, configured scripts, acceptance files and the AppID require review before trusting a project. Use an isolated test environment for interactions with external effects.
+
+No automatic upload, publishing or submission is part of this workflow.