Enterprise cryptographic services for agentic platforms. Provides secure key management, signing, encryption, and credential vending with mTLS authentication and comprehensive audit logging.
AgentKMS defines 9 tools with mixed quality. Tools 7-9 (crypto_sign, crypto_encrypt, crypto_decrypt) have excellent descriptions (150-200 chars) with clear parameter guidance and security notes. Tools 1-6 have minimal descriptions (20-50 chars) that lack context for LLM selection. All tools have proper JSON Schema with typed parameters and descriptions. No tool exposes secrets as parameters (good). Tool composition is sound, each does one thing, but discovery tools (list_providers) lack guidance on when to call them.
Decrypt ciphertext using an AgentKMS-managed key. Returns base64-encoded plaintext. If a context (AAD) was provided during encryption, it must be provided here too.
Encrypt data using an AgentKMS-managed key. The plaintext must be base64-encoded. Returns base64-encoded ciphertext and the key version used. The encryption key never leaves AgentKMS.
Sign a payload hash using an AgentKMS-managed asymmetric key. Returns the base64-encoded signature and the key version used. The private key never leaves AgentKMS. IMPORTANT: payload_hash must be the hex-encoded SHA-256 hash of the payload — do NOT send the payload itself.
Decrypt data using a configured key
Encrypt data using a configured key
Retrieve a credential from the vault
Tools 1-6 have descriptions under 50 characters, well below the 194-char baseline for A+ tools. 'Retrieve a credential from the vault' and 'Retrieve a secret from a provider' lack context for LLM selection and do not explain when to use each tool instead of similar ones.
No distinction between get_credential and get_secret in descriptions. Both retrieve secrets but from different sources. LLMs cannot infer which to call without explicit guidance on the difference.
Tools 1-6 lack parameter descriptions explaining format, constraints, or examples. 'name' and 'key' parameters have minimal guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Retrieve a secret from a provider
List available secret providers
Sign data using a configured key
No error handling guidance in any tool description. Tools do not explain what happens on failure (e.g., credential not found, decryption failure) or what the LLM should do next.
Tools 1-6 do not document output schemas. LLMs cannot plan downstream calls or extract required fields without knowing what each tool returns.