← 返回 Skills 市场
339
总下载
0
收藏
0
当前安装
3
版本数
在 OpenClaw 中安装
/install truth-frist
功能描述
Evidence-first verification for status, config, file contents, actions, connectivity, mounts, and model selection. Use before answering any such claim.
安全使用建议
This skill is coherent and appears to do what it says: run local checks and cite evidence before answering. Before installing, confirm you trust the skill source (source/homepage unknown), and be aware it expects command-line tools (ripgrep, shell utilities, and optionally the OpenClaw CLI) to perform checks even though those binaries are not declared in metadata. If you have sensitive files, run the skill in a restricted environment or review the SKILL.md and references/patterns.md to ensure the verification commands and paths match what you want it to access. If you need stricter limits, restrict the agent's filesystem/tool permissions or test in an isolated container first.
功能分析
Type: OpenClaw Skill
Name: truth-frist
Version: 1.5.0
The 'truth-first' skill bundle is a utility designed to enforce evidence-based verification for AI agent responses. It provides a structured workflow (SKILL.md) and command templates (references/patterns.md) that instruct the agent to use standard diagnostic tools like 'ripgrep', 'ls', and 'stat' to verify system states and configurations before answering. No evidence of malicious intent, data exfiltration, or harmful prompt injection was found; the logic is consistent with its stated purpose of improving accuracy and reducing hallucinations.
能力评估
Purpose & Capability
The name/description match the SKILL.md: the skill is an evidence-first verification workflow. It expects tools like ripgrep and the OpenClaw CLI (used in examples), but the registry metadata does not list required binaries; this is a minor mismatch between documentation and declared requirements.
Instruction Scope
SKILL.md explicitly directs the agent to read files, run local commands (rg, ls, stat, openclaw gateway status, view logs) and cite evidence. Those actions are appropriate for verification tasks, but they do grant broad read/command capability over local files and system state — which is necessary for the skill's purpose but worth noting before allowing it to run with wide filesystem/tool access.
Install Mechanism
Instruction-only skill with no install spec or archive downloads. Lowest-risk install footprint.
Credentials
The skill does not request environment variables, credentials, or config paths beyond referencing common OpenClaw config locations in examples. No disproportionate credential access requested.
Persistence & Privilege
No always:true, no install scripts, and no modifications to other skills or system-wide configuration are present. Normal, non-persistent skill.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install truth-frist - 安装完成后，直接呼叫该 Skill 的名称或使用
/truth-frist触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.5.0
updated to 1.5 version, 100% fit for clawhub skill format
v1.0.1
- Added SECURITY.md to document security policies and best practices.
- Updated README.md with improvements (details not specified).
v1.0.0
# Truth-First
Evidence-first verification for status, config, file contents, actions, connectivity, mounts, and model selection.
## What It Does
**Truth-First** forces the AI to verify claims before answering. Instead of relying on memory or assumptions, it:
1. Lists all claims that need verification
2. Gathers evidence using tools (read, status, grep, logs)
3. Classifies each claim as `Verified`, `Inferred`, or `Unknown`
4. Provides next-step commands for any `Unknown`
5. Cites evidence with paths, line numbers, or command outputs
## Installation
### Option 1: Manual Install
```bash
cd ~/.openclaw/skills
git clone <repo-url> truth-first
```
### Option 2: Via OpenClaw CLI
```bash
openclaw skill install <repo-url>
```
## Usage
The skill is automatically triggered for verification tasks. No manual activation needed.
### Example Prompts
- "Check if the gateway is running properly"
- "Verify my config has the correct model settings"
- "Is the backup file intact?"
- "Can you confirm the service restarted?"
## Response Framework
When Truth-First is active, responses follow this structure:
```
## Claims to Verify
- [ ] Claim 1: ...
- [ ] Claim 2: ...
## Evidence
### Claim 1: Verified ✅
- Source: /path/to/file
- Line 42: `model: kimi-k2.5`
### Claim 2: Unknown ❓
- No direct evidence found
- Next step: `rg -n "setting" ~/.openclaw/openclaw.json`
```
## Files
- `SKILL.md` - Core skill definition
- `agents/openai.yaml` - Agent profile for OpenAI-compatible models
- `references/patterns.md` - Reusable templates and evidence commands
## Requirements
- OpenClaw 2026.2.21 or later
- `ripgrep` (rg) for fast searches
- Standard shell tools (ls, stat, cat)
## License MIT
元数据
常见问题
Truth first 是什么？
Evidence-first verification for status, config, file contents, actions, connectivity, mounts, and model selection. Use before answering any such claim. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 339 次。
如何安装 Truth first？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install truth-frist」即可一键安装，无需额外配置。
Truth first 是免费的吗？
是的，Truth first 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Truth first 支持哪些平台？
Truth first 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Truth first？
由 royhk920（@royhk920）开发并维护，当前版本 v1.5.0。
推荐 Skills