feat(skill): orchestrate workspace business goals

.claude-plugin/marketplace.json

@@ -5,14 +5,14 @@
     "name": "multica-ai"
   },
   "metadata": {
-    "description": "Teach a local coding agent to operate Multica through the authenticated multica CLI",
+    "description": "Operate Multica and orchestrate workspace-aware business workflows through the authenticated multica CLI",
     "version": "1.1.0"
   },
   "plugins": [
     {
       "name": "multica-cli",
       "source": "./",
-      "description": "Operate Multica from a local coding agent via the multica CLI: issues, comments, metadata, mentions, and status side effects.",
+      "description": "Operate Multica and turn business goals into workspace-aware execution plans through the authenticated multica CLI.",
       "version": "1.1.0",
       "author": {
         "name": "multica-ai"

.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "multica-cli",
-  "description": "Teach a local coding agent to operate Multica through the authenticated multica CLI.",
+  "description": "Operate Multica and orchestrate workspace-aware business workflows through the authenticated multica CLI.",
   "version": "1.1.0",
   "author": {
     "name": "multica-ai"

.cursor/rules/multica-cli.mdc

@@ -43,6 +43,9 @@ around the missing flags.
 - **Never store or expose tokens**, and never bypass workspace permissions via private HTTP APIs.
 - **Say when the CLI can't do it.** Name the step and point at Multica Web rather than
   letting a partial run read as a completed one.
+- **Orchestrate open-ended business goals.** Search only relevant workspace resources,
+  synthesize existing context and capabilities, prefer reuse, then present one complete
+  execution plan for confirmation. Follow `references/orchestration.md` for the full flow.
 
 ## Full reference
 

.github/workflows/lint.yml

@@ -78,6 +78,9 @@ jobs:
       - name: Lint regression tests
         run: scripts/test-lint.py
 
+      - name: Business orchestration contract
+        run: python3 scripts/test-orchestration-contract.py
+
       - name: Lint documented commands
         run: |
           if [ "${{ matrix.cli }}" = "min" ]; then

EXAMPLES.md

@@ -54,6 +54,31 @@ multica issue assign <issue-id> --to-id <user-id>
 `issue create` refuses when an active duplicate already exists. Read that issue
 before reaching for `--allow-duplicate`.
 
+## Turn a business goal into an execution plan
+
+> Find the team's existing data-labeling workflow and capabilities, then plan
+> weekly rainy-day dataset labeling. Execute it after I confirm.
+
+The agent should inspect relevant existing resources, including Issue
+discussions where prior decisions may live, and synthesize what can be reused.
+It then presents the complete resource choices, capability gaps, mutations,
+dependencies, and acceptance criteria, and can execute it after one
+confirmation:
+
+```bash
+multica issue search "rainy data labeling" --include-closed --output json
+multica project list --output json
+multica agent list --output json
+multica skill list --output json
+multica autopilot list --output json
+```
+
+This is targeted discovery, not a full workspace dump. The agent gets details
+only for plausible matches, cites their identifiers in the proposal, prefers
+reuse when behavior need not change, and treats unknown sharing as shared. A
+material difference discovered during execution pauses the affected step for a
+revised confirmation.
+
 ## Record progress without starting another run
 
 > Mark MUL-123 in progress — I'm already working on it.

README.md

@@ -6,6 +6,8 @@ A portable skill that teaches any local coding agent — Claude Code, Codex, Cur
 and others — how to operate [Multica](https://github.com/multica-ai/multica)
 through the authenticated `multica` CLI: read and triage issues, reply to
 comments safely, manage metadata, and handle mention/status side effects.
+For open-ended goals, it can discover relevant workspace context and
+capabilities so teams reuse existing work instead of rebuilding it.
 
 English | [简体中文](./README.zh.md)
 
@@ -31,6 +33,14 @@ drive Multica safely; it never bypasses workspace permissions or stores secrets.
 - Inspecting run history and token usage; cancelling a run when asked
 - Linking pull requests back to Multica issues
 - Saying so plainly when the CLI cannot do something, instead of faking it
+- Turning open-ended goals into executable plans through targeted workspace
+  discovery
+- Summarizing existing team context and matching available Agent and Skill
+  capabilities
+- Reusing existing work instead of rebuilding it, with sharing-risk checks
+- Presenting the complete business orchestration plan for one confirmation,
+  then executing it in dependency order (Agent and Skill mutations are
+  confirmed separately)
 
 ## Install
 
@@ -85,6 +95,13 @@ Ask your agent to work with Multica once the skill is installed, for example:
 Read MUL-123 with the multica CLI and draft a reply for me to review.
 ```
 
+Or give it an outcome and let it find the relevant workspace context first:
+
+```text
+Find the team's existing data-labeling workflow and capabilities, then propose
+a plan for weekly rainy-day dataset labeling. Execute it after I confirm.
+```
+
 For write operations (comments, status changes, mentions, new issues), the agent
 should ask before making state changes unless you have already clearly
 authorized that exact action. See [EXAMPLES.md](./EXAMPLES.md) for more.
@@ -96,6 +113,7 @@ linted against a real `multica` binary:
 
 ```bash
 scripts/lint-skill-commands.py            # or --verbose to see every check
+python3 scripts/test-orchestration-contract.py
 ```
 
 It fails when a documented command or flag no longer exists, and when a CLI

README.zh.md

@@ -28,6 +28,11 @@ profile、当前 workspace，以及对每条命令的显式授权。这个 skill
 - 查看 run 历史与 token 用量；在被要求时取消 run
 - 把 pull request 关联回 Multica issue
 - CLI 做不到的事直说，并指向 Multica Web，而不是假装已完成
+- 针对开放式业务目标，定向检索 workspace 内相关资源并形成可执行方案
+- 归纳团队已有业务信息，匹配 Agent、Skill 等现有能力
+- 优先复用已有成果、避免重复建设，并根据共享风险决定是否新建隔离资源
+- 在聊天中展示完整业务编排方案，用户一次确认后按依赖顺序执行（Agent / Skill
+  变更仍需单独确认）
 
 ## 安装
 
@@ -80,6 +85,12 @@ cp -R skills/multica-cli/* ~/.cursor/skills/multica-cli/
 用 multica CLI 读一下 MUL-123，帮我起草一条回复让我 review。
 ```
 
+也可以直接描述业务目标，让 agent 先查找 workspace 内已有信息和能力：
+
+```text
+查找团队已有的数据送标流程和能力，设计每周雨天数据送标方案，等我确认后执行。
+```
+
 对于写操作（评论、状态变更、mention、新建 issue），除非用户已经明确授权这个具体动作，
 否则 agent 应在改动状态前先确认。更多示例见 [EXAMPLES.md](./EXAMPLES.md)。
 
@@ -90,6 +101,7 @@ cp -R skills/multica-cli/* ~/.cursor/skills/multica-cli/
 
 ```bash
 scripts/lint-skill-commands.py            # 加 --verbose 可看到每一条检查
+python3 scripts/test-orchestration-contract.py
 ```
 
 当文档里写的命令或 flag 已经不存在、或者 CLI 里有 `SKILL.md` 从未提及的命令时，

scripts/lint-skill-commands.py

@@ -54,6 +54,7 @@
 # docs that get copy-pasted just as often, so they rot the same way.
 DOC_FILES = [
     "skills/multica-cli/SKILL.md",
+    "skills/multica-cli/references/orchestration.md",
     "EXAMPLES.md",
     "README.md",
     "README.zh.md",

scripts/test-orchestration-contract.py

@@ -0,0 +1,96 @@
+#!/usr/bin/env python3
+"""Pin the portable Skill's business-orchestration behavior."""
+
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parent.parent
+
+
+def require(path: str, anchors: list[str], failures: list[str]) -> None:
+    target = ROOT / path
+    if not target.is_file():
+        failures.append(f"{path}: file is missing")
+        return
+    content = " ".join(target.read_text(encoding="utf-8").split())
+    for anchor in anchors:
+        if anchor not in content:
+            failures.append(f"{path}: missing {anchor!r}")
+
+
+def main() -> int:
+    failures: list[str] = []
+
+    require(
+        "skills/multica-cli/SKILL.md",
+        [
+            "open-ended business goal",
+            "references/orchestration.md",
+            "Route by intent, not resource count",
+        ],
+        failures,
+    )
+    require(
+        "skills/multica-cli/references/orchestration.md",
+        [
+            "targeted read-only discovery",
+            "Issues, Projects, Agents, Squads, Skills, and Autopilots",
+            "search or list first",
+            "business context and available capabilities",
+            "dedicated",
+            "shared",
+            "unknown",
+            "Treat unknown as shared",
+            "Prefer reuse over creating a duplicate",
+            "The in-chat orchestration design is the execution plan",
+            "one user confirmation",
+            "execute the plan directly",
+            "separate confirmation immediately before changing an Agent or Skill",
+            "material deviation",
+            "Multica Web",
+        ],
+        failures,
+    )
+    require(
+        "README.md",
+        [
+            "targeted workspace discovery",
+            "reuse existing work instead of rebuilding it",
+            "one confirmation",
+        ],
+        failures,
+    )
+    require(
+        "README.zh.md",
+        [
+            "定向检索",
+            "避免重复建设",
+            "一次确认",
+        ],
+        failures,
+    )
+    require(
+        "EXAMPLES.md",
+        [
+            "Turn a business goal into an execution plan",
+            "inspect relevant existing resources",
+            "execute it after one confirmation",
+        ],
+        failures,
+    )
+
+    if failures:
+        print(f"{len(failures)} orchestration contract check(s) failed:", file=sys.stderr)
+        for failure in failures:
+            print(f"  - {failure}", file=sys.stderr)
+        return 1
+
+    print("Business orchestration contract passed.")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

skills/multica-cli/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: multica-cli
-description: "Use when a local coding agent (Codex, Claude Code, Cursor, or similar) needs to operate Multica through the authenticated `multica` CLI: reading or updating issues, comments, metadata, labels, custom properties, subscribers, projects, agents, squads, runtimes, repos, skills, autopilots, attachments, or workspace state; searching issues; inspecting or cancelling agent runs; replying to a Multica issue from an external agent; creating or triaging issues; checking linked pull requests; or safely handling Multica mention/status side effects without relying on the Multica hosted agent runtime."
+description: "Use when a local coding agent (Codex, Claude Code, Cursor, or similar) needs to operate Multica through the authenticated `multica` CLI: reading or updating issues, comments, metadata, labels, custom properties, subscribers, projects, agents, squads, runtimes, repos, skills, autopilots, attachments, or workspace state; turning an open-ended business goal into a workflow; searching issues; inspecting or cancelling agent runs; replying to a Multica issue from an external agent; creating or triaging issues; checking linked pull requests; or safely handling Multica mention/status side effects without relying on the Multica hosted agent runtime."
 ---
 
 # Multica CLI
@@ -173,6 +173,14 @@ with that number, and the prefix is not validated — pasting an identifier from
 another tracker can put an unrelated local issue at the top. The `match_source`
 field is a display hint, not a filter.
 
+## Business Goal Orchestration
+
+For an open-ended business goal, resource-selection question, automation, or
+set of dependent writes, read `references/orchestration.md` before proposing a
+solution or changing state. Route by intent, not resource count. A concrete
+operation on a known target stays in the direct read/write workflow in this
+file.
+
 ## Write Workflow
 
 Treat writes as side-effecting. If the user did not clearly ask for the write,

skills/multica-cli/agents/openai.yaml

@@ -1,4 +1,4 @@
 interface:
   display_name: "Multica CLI"
-  short_description: "Operate Multica through the local CLI"
-  default_prompt: "Use $multica-cli to work with a Multica issue through the authenticated local CLI."
+  short_description: "Operate and orchestrate Multica through the local CLI"
+  default_prompt: "Use $multica-cli to operate Multica or turn a business goal into a workspace-aware execution plan through the authenticated local CLI."