Soul Guard
Detects and alerts on prompt injection backdoors in agent identity files (SOUL.md) — runs as a startup check or background watcher. Protects agent identity from silent manipulation before any instruction is loaded.
Install
npx skills add Parad0x-Labs/openclaw-skills --skill soul-guardgit clone --depth 1 https://github.com/Parad0x-Labs/openclaw-skills.git /tmp/openclaw-skills
mkdir -p ~/.claude/skills
cp -r /tmp/openclaw-skills/plugins/soul-guard ~/.claude/skills/soul-guardPersonal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.
From the SKILL.md
# Soul Guard
Detects prompt injection backdoors in agent identity files. SOUL.md files tell
agents who they are — if one gets tampered with ("ignore previous instructions",
"you are now..."), Soul Guard catches it before it affects the agent.
Runs as a startup check (once before the agent loads its identity) or a
background watcher (re-checks on file change). Alerts on any injection pattern
or unauthorized modification.
## When to use
- Any agent that loads its identity from a SOUL.md or similar identity file
- Shared agent deployments where the identity file could be written by untrusted parties
- Any agent that needs to assert its own identity hasn't been silently overwritten
## Source
github.com/Parad0x-Labs/openclaw-skills/tree/main/plugins/soul-guardDetails
- Skill name
soul-guard- Repository
- Parad0x-Labs/openclaw-skills
- Path
plugins/soul-guard/SKILL.md- License
- MIT
- Repo updated
- 13 November 2026
FAQ
What does the Soul Guard skill do?
Detects and alerts on prompt injection backdoors in agent identity files (SOUL.md) — runs as a startup check or background watcher. Protects agent identity from silent manipulation before any instruction is loaded.
How do I install the Soul Guard skill?
skills CLI (Claude Code, Cursor, Codex and more): npx skills add Parad0x-Labs/openclaw-skills --skill soul-guard. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.
Is Soul Guard popular?
It lives in Parad0x-Labs/openclaw-skills, which has 23 GitHub stars and 3 forks.
Related skills
Agent Passport
23Parad0x-Labs/openclaw-skills
On-chain identity for OpenClaw agents — .null name, ETH↔Solana binding, verifiable agent identity without touching private keys
Context Capsule
23Parad0x-Labs/openclaw-skills
Compresses older OpenClaw agent session history into a bounded, lane-change-aware context capsule — keeps recent messages verbatim, flags abandoned directions, quarantines injected instructions, and redacts secrets. Local, deterministic, any model.
Correction Intent Tagger
23Parad0x-Labs/openclaw-skills
Detects when a user message corrects a prior one, builds an active state where corrections override originals, and injects only the live instruction set into agent context. Prevents model drift when users change their minds mid-session.
Web0 Onboard
23Parad0x-Labs/openclaw-skills
web0 setup for an OpenClaw agent — plan your identity + paid x402 storefront, get the receipt-anchoring status and your .null name status (mainnet registrar retired 2026-08-29; existing names resolve read-only, mainnet registration frozen). Sell services for USDC on Solana. Non-custodial (your wallet signs).
Parad0x MCP Server
23Parad0x-Labs/openclaw-skills
MCP server exposing the Parad0x Labs stack — x402 payment quotes, receipt hashing (no on-chain anchoring in this server; a devnet receipt_anchor is available for explicit use), read-only .null resolution and Dark Passport lookup, outcome receipts, receipt compression, and program status discovery. Runs over stdio, works with Claude Desktop and any MCP client.
Stop Slop Zh
11leeguooooo/stop-slop-zh
写、起草、生成、润色任何面向人读的中文时,都要用本 Skill——公众号、博客、知乎、小红书、技术文档、技术方案、设计文档、README、公文、学术摘要、产品文案、复盘、提案、新闻稿、邮件、对外通知,以及回别人的评论 / IM / PR 意见。无论是从零写新稿,还是改稿、审稿,只要在产出中文长文,写完、定稿、交付、发布前都必须调用它过一遍:它识别并消除中文 AI 写作的典型痕迹——套话、排比三件套、名词化、抽象主语、金句收尾、总分总八股、标点指纹、标题套路、评级悬念开头、转述腔、提示语冒号、翻译腔、相邻句同构、同义词轮换——让文本读起来像有人在说话,而不是 AI 在生成。技术与对外文档另有专项(第三节:不写术语表、不解释读者本来就会的概念、结论文档不写过程、对外文档不写内部日记);回复类另有专项(第四节:不写开场白、不写客套、陈述事实而不是承诺)。简言之:只要这次任务的产物里有一段是要给人读的中文,就触发本 Skill。
Browse the most popular Claude skills, all ai, llms & agents skills, or write your own with the skills generator.