技能详情（站内镜像，无评论）
许可证：MIT-0
MIT-0 ·免费使用、修改和重新分发。无需归因。
版本：v1.0.0
统计：⭐ 4 · 2.3k · 8 current installs · 8 all-time installs
⭐ 4
安装量（当前） 8
🛡 VirusTotal ：良性 · OpenClaw ：良性
Package：agent-docs
安全扫描（ClawHub）
- VirusTotal ：良性
- OpenClaw ：良性
OpenClaw 评估
The skill is an instruction-only guide for writing agent-optimized docs and its requirements, scope, and lack of installs/credentials are coherent with that purpose.
目的
Name/description match the actual content: the SKILL.md and reference file are documentation guidance for agent-consumable docs. The skill requests no binaries, env vars, or installs — which is appropriate for an authored-guidance skill.
说明范围
The runtime instructions stay within the stated purpose (how to structure AGENTS.md / llms.txt / RAG-friendly docs). They recommend reading local docs (e.g., using read_file) which is expected. A prompt-injection pattern was detected by the scanner, but it appears inside examples and an OWASP-style discussion demonstrating attacks and mitigations — so its presence is likely educational rather than malicious. Still: any agent implementation usi…
安装机制
No install spec and no code files — lowest-risk model. Nothing will be written to disk or fetched on install by the skill itself.
证书
The skill declares no env vars or credentials. Example snippets showing process.env or GET requests are illustrative of attack vectors and mitigation, not actual requirements. There is no unexplained credential access.
持久
always is false, and the skill does not request persistent privileges or modify system/agent-wide configs. Autonomous invocation is allowed (platform default); this is reasonable for a guidance skill but operators should be mindful of agent tooling the skill is used with (see guidance).
综合结论
This skill is an authored guide for writing LLM-friendly documentation and is internally consistent with that purpose. It doesn't request credentials or install anything. Before you install or let agents use it autonomously, verify that: (1) your agent enforces domain allow-lists and sanitizes any external content the agent might fetch (the skill itself emphasizes this); (2) AGENTS.md/llms.txt files do not contain secrets or credentials; and (…
安装（复制给龙虾 AI）
将下方整段复制到龙虾中文库对话中，由龙虾按 SKILL.md 完成安装。
请把本段交给龙虾中文库（龙虾 AI）执行：为本机安装 OpenClaw 技能「Agent Docs」。简介：Create documentation optimized for AI agent consumption. Use when writing SKILL…。
请 fetch 以下地址读取 SKILL.md 并按文档完成安装：https://raw.githubusercontent.com/openclaw/skills/refs/heads/main/skills/tylervovan/agent-docs/SKILL.md
（来源：yingzhi8.cn 技能库）SKILL.md
---
name: agent-docs
description: Create documentation optimized for AI agent consumption. Use when writing SKILL.md files, README files, API docs, or any documentation that will be read by LLMs in context windows. Helps structure content for RAG retrieval, token efficiency, and the Hybrid Context Hierarchy.
---
# Agent Docs
Write documentation that AI agents can efficiently consume. Based on Vercel benchmarks and industry standards (AGENTS.md, llms.txt, CLAUDE.md).
## The Hybrid Context Hierarchy
Three-layer architecture for optimal agent performance:
### Layer 1: Constitution (Inline)
**Always in context.** 2,000–4,000 tokens max.
```markdown
# AGENTS.md
> Context: Next.js 16 | Tailwind | Supabase
## 🚨 CRITICAL
- NO SECRETS in output
- Use `app/` directory ONLY
## 📚 DOCS INDEX (use read_file)
- Auth: `docs/auth/llms.txt`
- DB: `docs/db/schema.md`
```
**Include:**
- Security rules, architecture constraints
- Build/test/lint commands (top for primacy bias)
- Documentation map (where to find more)
### Layer 2: Reference Library (Local Retrieval)
**Fetched on demand.** 1K–5K token chunks.
- Framework-specific guides
- Detailed style guides
- API schemas
### Layer 3: Research Assistant (External)
**Gated by allow-lists.** Edge cases only.
- Latest library updates
- Stack Overflow for obscure errors
- Third-party llms.txt
## Why This Works
**Vercel Benchmark (2026):**
| Approach | Pass Rate |
|----------|-----------|
| Tool-based retrieval | 53% |
| Retrieval + prompting | 79% |
| **Inline AGENTS.md** | **100%** |
**Root cause:** Meta-cognitive failure. Agents don't know what they don't know—they assume training data is sufficient. Inline docs bypass this entirely.
## Core Principles
### 1. Compressed Index > Full Docs
An 8KB compressed index outperforms a 40KB full dump.
**Compress to:**
- File paths (where code lives)
- Function signatures (names + types only)
- Negative constraints ("Do NOT use X")
### 2. Structure for Chunking
RAG systems split at headers. Each section must be self-contained:
```markdown
## Database Setup ← Chunk boundary
Prerequisites: PostgreSQL 14+
1. Create database...
```
**Rules:**
- Front-load key info (chunkers truncate)
- Descriptive headers (agents search by header text)
### 3. Inline Over Links
Agents can't autonomously browse. Each link = tool call + latency + potential failure.
| Approach | Token Load | Agent Success |
|----------|------------|---------------|
| Full inline | ~12K | ✅ High |
| Links only | ~2K | ❌ Requires fetching |
| Hybrid | ~4K base | ✅ Best of both |
### 4. The "Lost in the Middle" Problem
LLMs have U-shaped attention:
- **Strong:** Start of context (primacy)
- **Strong:** End of context (recency)
- **Weak:** Middle of context
**Solution:** Put critical rules at TOP of AGENTS.md. Governance first, details later.
### 5. Signal-to-Noise Ratio
Strip everything that isn't essential:
- No "Welcome to..." preambles
- No marketing text
- No changelogs in core docs
Formats like llms.txt and AGENTS.md mechanically increase SNR.
## llms.txt Standard
Machine-readable doc index for agents:
```markdown
# Project Name
> One-line project description.
## Authentication
- [Setup](docs/auth/setup.md): Environment vars and init
- [Server](docs/auth/server.md): Cookie handling
## Database
- [Schema](docs/db/schema.md): Full Prisma schema
```
**Location:** `/llms.txt` at domain root
**Companion:** `/llms-full.txt` — full concatenated docs, HTML stripped
## Security Considerations
### Inline = Trusted
AGENTS.md is part of your codebase. Controlled, version-pinned.
### External = Attack Surface
- Indirect prompt injection via hidden text
- SSRF risks if agents can browse freely
- Dependency on external uptime
**Mitigation:** Domain allow-lists, human-in-the-loop for external retrieval.
## Anti-Patterns
1. **Pasting 50 pages** — triggers "Lost in the Middle"
2. **"See external docs"** — agents can't browse autonomously
3. **Generic advice** — "Write clean code" (use specific constraints)
4. **TOC-only docs** — indexes without content
5. **Trusting retrieval alone** — 53% vs 100% pass rate
## Advanced Patterns
For detailed guidance on RAG optimization, multi-framework docs, and API templates, see [references/advanced-patterns.md](references/advanced-patterns.md).
## Validation Checklist
- [ ] Critical governance at TOP of doc
- [ ] Total inline context under 4K tokens
- [ ] Each H2 section self-contained
- [ ] No external links without inline summary
- [ ] Negative constraints explicit ("Do NOT...")
- [ ] File paths and signatures, not full code