Merge pull request #8 from AElfProject/codex/ironclaw-trusted-mcp

Codex/ironclaw trusted mcp

Author: hzz780

README.md

@@ -5,7 +5,7 @@
 [![Unit Tests](https://github.com/AElfProject/aelf-node-skill/actions/workflows/test.yml/badge.svg)](https://github.com/AElfProject/aelf-node-skill/actions/workflows/test.yml)
 [![Coverage](https://img.shields.io/endpoint?url=https://AElfProject.github.io/aelf-node-skill/coverage.json)](https://AElfProject.github.io/aelf-node-skill/coverage.json)
 
-AElf Node Skill provides MCP, CLI, and SDK interfaces for AElf public nodes with `REST for reads, SDK for contract execution, and selective fallback for fee estimate`.
+AElf Node Skill provides MCP, CLI, and SDK interfaces for AElf public nodes with `REST for reads, SDK for contract execution, and selective fallback for fee estimate`, plus OpenClaw and IronClaw integration.
 
 ## Features
 
@@ -62,20 +62,50 @@ bun run test:unit
 
 ## Setup CLI
 
-This repository includes a one-command setup CLI for Claude, Cursor, and OpenClaw.
+This repository includes a one-command setup CLI for Claude, Cursor, OpenClaw, and IronClaw.
 
 ```bash
 bun run setup claude
 bun run setup cursor
 bun run setup cursor --global
+bun run setup ironclaw
 bun run setup openclaw
 bun run setup openclaw --config-path /path/to/openclaw-config.json
 bun run setup list
 bun run setup uninstall claude
 bun run setup uninstall cursor --global
+bun run setup uninstall ironclaw
 bun run setup uninstall openclaw --config-path /path/to/openclaw-config.json
 ```
 
+## IronClaw
+
+```bash
+bun run setup ironclaw
+bun run setup uninstall ironclaw
+```
+
+The IronClaw setup writes a stdio MCP entry to `~/.ironclaw/mcp-servers.json` and installs this repo's `SKILL.md` to `~/.ironclaw/skills/aelf-node-skill/SKILL.md`.
+
+Important trust model note:
+
+- Use the trusted skill path above for write-capable flows such as `aelf_send_contract_transaction`.
+- Do not rely on `~/.ironclaw/installed_skills/` for the primary install path when you need write approval behavior.
+- This MCP server emits both standard MCP camelCase annotations and IronClaw-compatible snake_case annotations so the current IronClaw source can honor read/write hints.
+
+Remote activation contract:
+
+- GitHub repo/tree URLs are discovery sources only, not the final IronClaw install payload.
+- Preferred IronClaw activation from npm: `bunx -p @blockchain-forever/aelf-node-skill aelf-node-setup ironclaw`
+- Prefer ClawHub / managed install for OpenClaw when available; otherwise use `bunx -p @blockchain-forever/aelf-node-skill aelf-node-setup openclaw`
+- Local repo checkout remains a development smoke-test path only.
+
+Minimal smoke test:
+
+1. `bun run setup ironclaw`
+2. Ask IronClaw for `latest block height on AELF`
+3. Ask it to `send a contract transaction` and confirm approval appears before execution
+
 Installed package users can also run:
 
 ```bash

README.zh-CN.md

@@ -5,7 +5,7 @@
 [![Unit Tests](https://github.com/AElfProject/aelf-node-skill/actions/workflows/test.yml/badge.svg)](https://github.com/AElfProject/aelf-node-skill/actions/workflows/test.yml)
 [![Coverage](https://img.shields.io/endpoint?url=https://AElfProject.github.io/aelf-node-skill/coverage.json)](https://AElfProject.github.io/aelf-node-skill/coverage.json)
 
-AElf Node Skill 提供 MCP、CLI、SDK 三种接口，采用“读走 REST、合约执行走 SDK、手续费估算选择性 fallback”的架构访问 AElf 公共节点。
+AElf Node Skill 提供 MCP、CLI、SDK 三种接口，采用“读走 REST、合约执行走 SDK、手续费估算选择性 fallback”的架构访问 AElf 公共节点，并支持 OpenClaw 与 IronClaw 集成。
 
 ## 功能
 
@@ -62,20 +62,50 @@ bun run test:unit
 
 ## Setup CLI
 
-仓库内置一键 setup，支持 Claude、Cursor、OpenClaw。
+仓库内置一键 setup，支持 Claude、Cursor、OpenClaw、IronClaw。
 
 ```bash
 bun run setup claude
 bun run setup cursor
 bun run setup cursor --global
+bun run setup ironclaw
 bun run setup openclaw
 bun run setup openclaw --config-path /path/to/openclaw-config.json
 bun run setup list
 bun run setup uninstall claude
 bun run setup uninstall cursor --global
+bun run setup uninstall ironclaw
 bun run setup uninstall openclaw --config-path /path/to/openclaw-config.json
 ```
 
+## IronClaw
+
+```bash
+bun run setup ironclaw
+bun run setup uninstall ironclaw
+```
+
+IronClaw 安装会向 `~/.ironclaw/mcp-servers.json` 写入 stdio MCP entry，并把当前仓库的 `SKILL.md` 安装到 `~/.ironclaw/skills/aelf-node-skill/SKILL.md`。
+
+关于 trust model 的说明：
+
+- 如果要执行 `aelf_send_contract_transaction` 这类写能力，必须使用上面的 trusted skill 路径。
+- 不要把 `~/.ironclaw/installed_skills/` 当成主安装路径，否则写操作的 approval 行为会不稳定。
+- 当前 MCP server 会同时输出标准 MCP camelCase annotations 和 IronClaw 兼容 snake_case annotations，确保 IronClaw 能识别读写 hints。
+
+远程激活契约：
+
+- GitHub repo/tree URL 只用于 discovery，不是最终的 IronClaw 安装载体。
+- 推荐的 IronClaw npm 激活命令：`bunx -p @blockchain-forever/aelf-node-skill aelf-node-setup ironclaw`
+- OpenClaw 若有 ClawHub / managed install 则优先使用；否则回退到 `bunx -p @blockchain-forever/aelf-node-skill aelf-node-setup openclaw`
+- 本地 repo checkout 仅保留给开发阶段 smoke test。
+
+最短 smoke test：
+
+1. `bun run setup ironclaw`
+2. 让 IronClaw 查询 `latest block height on AELF`
+3. 再让它 `send a contract transaction`，确认执行前会出现 approval
+
 安装后的包也可直接执行：
 
 ```bash

SKILL.md

@@ -1,19 +1,43 @@
 ---
 name: "aelf-node-skill"
+version: "0.1.3"
 description: "AElf node querying and contract execution skill for agents."
+activation:
+  keywords:
+    - node
+    - rpc
+    - block
+    - transaction
+    - contract
+    - fee
+    - chain status
+    - system contract
+  exclude_keywords:
+    - wallet
+    - dex
+    - dao
+    - explorer
+    - guardian
+  tags:
+    - node
+    - rpc
+    - aelf
+    - blockchain
+  max_context_tokens: 1800
 ---
 
 # AElf Node Skill
 
 ## When to use
 - Use this skill when you need AElf chain query, contract view/send, and fee estimation tasks.
+- Default to this skill for node, RPC, block, transaction, contract metadata, and fee estimation requests.
 
 ## Capabilities
 - Chain reads: status, block, transaction result, metadata
 - Contract operations: view call and transaction sending
 - Node registry import/list with REST-first and SDK fallback strategy
 - Shared signer resolution for write operations: `explicit -> context -> env`
-- Supports SDK, CLI, MCP, and OpenClaw integration from one codebase.
+- Supports SDK, CLI, MCP, OpenClaw, and IronClaw integration from one codebase.
 
 ## Safe usage rules
 - Never print private keys, mnemonics, or tokens in channel outputs.
@@ -24,13 +48,21 @@ description: "AElf node querying and contract execution skill for agents."
 ## Command recipes
 - Start MCP server: `bun run mcp`
 - Run CLI entry: `bun run cli`
+- Install into IronClaw: `bun run setup ironclaw`
 - Generate OpenClaw config: `bun run build:openclaw`
 - Verify OpenClaw config: `bun run build:openclaw:check`
 - Run CI coverage gate: `bun run test:coverage:ci`
 - For write calls, pass optional `signerContext` with `signerMode=auto`.
 
+## Distribution / Activation
+- GitHub repo/tree URLs are discovery-only for hosts and agents.
+- Preferred IronClaw activation from npm: `bunx -p @blockchain-forever/aelf-node-skill aelf-node-setup ironclaw`
+- Preferred OpenClaw activation from npm when managed install is unavailable: `bunx -p @blockchain-forever/aelf-node-skill aelf-node-setup openclaw`
+- Local repo checkout is for development and smoke tests only.
+
 ## Limits / Non-goals
 - This skill focuses on domain operations and adapters; it is not a full wallet custody system.
 - Do not hardcode environment secrets in source code or docs.
 - Avoid bypassing validation for external service calls.
 - `signerMode=daemon` is reserved and returns `SIGNER_DAEMON_NOT_IMPLEMENTED` in this release.
+- Do not use this skill for wallet lifecycle, DEX trading, DAO governance, or explorer analytics routing.