{"id":"708df97c-4e88-43cb-8719-8039af0f6db2","slug":"dcl-skills","kind":"skill","name":"dcl-skills","summary":"Security & compliance skills for AI agents (jailbreak defense, PII/secret redaction, hallucination checks) with a DeFi-specialized suite — free instruction-only mode or paid on-chain-verified mode via","owner":"Fronesis-Labs","repoUrl":"https://github.com/Fronesis-Labs/dcl-skills","license":"Apache-2.0","licenseTier":"permissive","trust":"community","stars":2,"match":null,"fewStars":true,"format":"skill","createdAt":"2026-08-01T05:11:25.000Z","addedAt":"2026-10-21T04:54:58.804Z","logoUrl":"https://avatars.githubusercontent.com/u/264291404?v=4","linkOnly":false,"collection":null,"skillPath":null,"categories":["cloud-infra"],"tags":["agent-safety","ai-safety","compliance","defi","llm-security","pii-detection-ai","prompt-injection","x402"],"runtime":"binary","local":true,"cloudCapable":false,"cloudCapability":"unknown","cloudReadyOnPlatform":false,"needsApiKey":false,"reviewed":false,"syncedAt":"2026-10-28T00:19:21.012Z","sample":false,"url":"https://conduits.dev/tools/dcl-skills","language":"","transports":["stdio"],"install":{},"requirements":{"env":[],"upstreamMayCharge":false},"tools":null,"readme":"# DCL Skills\n\n**Don't trust the agent. Trust the proof.**\n\nCompliance and security skills for AI agents — hallucination detection,\nPII redaction, prompt-injection defense, secret-leak detection, and a\nDeFi-specialized variant set for agents that touch wallets and on-chain\nactions.\n\nPublished on [ClawHub](https://clawhub.ai/daririnch) under\n[@daririnch](https://clawhub.ai/daririnch).\n\n## Why This Exists\n\nAn agent that can act autonomously is also an agent that can be tricked,\ndrift off its instructions, or leak something it shouldn't — silently,\nwith no record. These skills give an agent (or the pipeline running it) a\nway to check itself *before* acting and to leave a verifiable trail of what\nit checked. Most skills work two ways: a free, instruction-only checklist\nyou can run yourself right now, and an optional live audit backed by the\nDCL Trust Oracle for a cryptographically verifiable verdict.\n\n## Structure\n\n```\ndcl-skills/\n├── skills/\n│   ├── core/      # general-purpose agent-safety skills\n│   └── crypto/     # DeFi-specialized variants (wallets, trades, MEV, on-chain provenance)\n├── docs/          # pipeline design, security model, whitepaper\n├── tests/golden/  # golden-set eval cases per skill\n└── .github/workflows/lint-skills.yml\n```\n\n## Skills\n\n### Core\n\n| Skill | Purpose |\n|---|---|\n| `dcl-policy-enforcer` | Enforce compliance policy on agent actions before execution |\n| `dcl-sentinel-trace` | Detect & redact PII in agent outputs |\n| `dcl-prompt-firewall` | Detect & block prompt injection |\n| `dcl-secret-leak-detector` | Detect leaked secrets/credentials |\n| `dcl-provenance-tracker` | Verify skill version integrity / detect supply-chain drift after updates |\n| `dcl-skill-auditor` | Pre-install static security scanner for ClawHub skills |\n| `dcl-semantic-drift-guard` | Detect hallucination / context drift vs grounding sources |\n| `dcl-pipeline` | Meta-skill orchestrating the above as one pipeline |\n\n### Crypto\n\n| Skill | Purpose |\n|---|---|\n| `dcl-prompt-firewall-crypto` | Injection defense specialized for DeFi contexts |\n| `dcl-wallet-guardian` | Validate wallet-affecting actions before execution |\n| `dcl-trade-verifier` | Verify trade/swap parameters before execution |\n| `dcl-mev-compliance` | Check MEV exposure / compliance before broadcast |\n| `dcl-provenance-tracker-crypto` | Verify provenance of on-chain data/contracts referenced |\n| `dcl-semantic-drift-crypto` | Crypto-specific hallucination detection (token/contract claims) |\n| `dcl-output-sanitizer` | Sanitize agent output before display/execution |\n| `dcl-crypto-commit` | Final compliance checkpoint before irreversible on-chain action |\n\n## Free by Default, Verifiable if You Need It\n\nEvery skill runs standalone as a free, instruction-only checklist — no\naccount, no network call, no dependency on Fronesis Labs infrastructure.\nMost also offer an optional live check against the DCL Trust Oracle for a\ncryptographically sealed, tamper-evident verdict instead of a self-reported\none, metered per call (settled automatically via x402, USDC on Base) so\nthere's no subscription to manage for occasional or high-volume use.\n`dcl-crypto-commit` is the one skill with no free equivalent, since its\nwhole purpose is writing to the on-chain audit chain. `dcl-provenance-tracker`\nand `dcl-provenance-tracker-crypto` are the opposite case — 100%\ninstruction-only by design, with only an *optional* on-chain cross-check of\na prior record. See each skill's `SKILL.md` for details.\n\n## Status\n\nAll 16 `SKILL.md` files under `skills/` (8 core + 8 crypto) have been\nmigrated from ClawHub and are complete — see\n[docs/pipeline.md](docs/pipeline.md) for the per-skill migration log,\nincluding notes on which MCP tool names/prices were inferred by analogy\nrather than confirmed from a source doc (mainly the crypto suite, which\nstarted from thin ClawHub listings).\n\n## License\n\n- **This repo**: Apache License 2.0 — see [LICENSE](LICENSE).\n- **ClawHub listings**: platform-enforced MIT-0, cannot be overridden per\n  skill.\n\nSee [docs/licensing.md](docs/licensing.md) for the full breakdown,\nincluding an open item on x402/paid-audit wording vs ClawHub's\nno-pricing-metadata rule.\n"}