infisical-kms — Claude Code skill: encrypt without the key
A guide that walks the agent through Infisical KMS: encrypt, decrypt, sign, verify and HMAC with keys that never leave the platform.
MIT 238 installs ★ 41 updated 2026-10-19
npx skills add Infisical/ai-skills --skill infisical-kms -a claude-codeWhat infisical-kms does
The skill gives the agent a procedure for Infisical KMS. First it confirms the task is a key operation and not secret storage. Then it picks a keyUsage (encrypt-decrypt, sign-verify or generate-verify-mac), an algorithm, an exportability setting and a root of trust, and only after that calls the API.
Two reference files sit beside SKILL.md. references/keys-and-operations.md covers key creation, algorithm values, the encrypt, decrypt, sign, verify and MAC endpoints, and rotation. references/external-kms-and-kmip.md handles AWS KMS and GCP KMS backing, KMIP servers, Kubernetes encryption-at-rest and Sigstore cosign. Algorithm strings differ by subsystem: symmetric ones are lowercase with hyphens (aes-256-gcm), asymmetric and signing ones use UPPER_SNAKE (ECC_NIST_P256), post-quantum uses ML_DSA_65 while PKI writes ML-DSA-65. A value copied from infisical-pki will not work here.
Plaintext and ciphertext travel base64-encoded both ways. Every operation is a network call with a size limit, so big payloads get a locally generated data key and KMS only wraps that key. Rotation adds a version for future calls and leaves existing ciphertext on the old one; it does not re-protect stored data. Marking a key isExportable trades away the main guarantee, since the material can then be pulled out.
What it helps with
- encrypt a database column with a key the app never holds
- sign a payload and verify the signature on another service
- generate an HMAC to check message integrity
- wrap a large payload with envelope encryption instead of shipping it to KMS
- point the root of trust at AWS KMS, GCP KMS or a KMIP server
Example prompts for the agent
- Encrypt the
card_numbercolumn with anencrypt-decryptkey and return base64. - Sign this payload with a
sign-verifykey usingRSASSA_PSS_SHA_256and show me the request. - Set up envelope encryption: generate the data key locally and let KMS wrap it.
How to install infisical-kms
1Skills CLI (Vercel)
npx skills add Infisical/ai-skills --skill infisical-kms -a claude-code2Manually: put the files in place
~/.claude/skills/infisical-kms/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/infisical-kms/
1Skills CLI (Vercel)
npx skills add Infisical/ai-skills --skill infisical-kms -a codex2Manually: put the files in place
.agents/skills/infisical-kms/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Infisical/ai-skills --skill infisical-kms -a cursor2Manually: put the files in place
.cursor/skills/infisical-kms/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Infisical/ai-skills skills/infisical-kms2Manually: put the files in place
.github/skills/infisical-kms/SKILL.md1Skills CLI (Vercel)
npx skills add Infisical/ai-skills --skill infisical-kms -a gemini-cli2Platform CLI
gemini skills install https://github.com/Infisical/ai-skills --path skills/infisical-kmsManually: put the files in place
.devin/skills/infisical-kms/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add Infisical/ai-skills --skill infisical-kms -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about infisical-kms
How do I install infisical-kms in Claude Code?
Run npx skills add Infisical/ai-skills --skill infisical-kms -a claude-code. The skill lands at ~/.claude/skills/infisical-kms/SKILL.md.
What does the infisical-kms skill do?
It walks the agent through Infisical KMS work: key usage, algorithm choice, exportability, root of trust, then the encrypt, decrypt, sign, verify and MAC calls.
How is a KMS key different from a secret in Infisical?
A secret is stored and handed back when you ask for it. A KMS key stays on the platform: you send plaintext or a digest and get the result, never the key material.
Source text
SKILL.md · MIT
--- name: infisical-kms description: "Guide for Infisical KMS — managing cryptographic keys and performing encrypt/decrypt, sign/verify, and HMAC operations without the key ever leaving the platform. Covers KMS key creation with the three key usages (encrypt-decrypt, sign-verify, generate-verify-mac), symmetric algorithms (aes-256-gcm, aes-128-gcm), asymmetric algorithms (RSA_4096, ECC_NIST_P256/P384/P521, post-quantum ML_DSA_44/65/87), signing algorithms (RSASSA_PSS, RSASSA_PKCS1_V1_5, ECDSA, ML-DSA), HMAC algorithms, key rotation, exportable keys, external KMS backing with AWS KMS or GCP KMS, KMIP server integration, Kubernetes encryption-at-rest, and Sigstore cosign signing. Use this skill when someone asks about: Infisical KMS, encrypting data with a managed key, envelope encryption, sign and verify with Infisical, HMAC, KMIP, external KMS, or 'how do I encrypt data without handling the key'. For doing cryptography with a key Infisical never releases. Not for storing a value and reading it back, which is a secret (infisical-setup), nor for X.509 certificates (infisical-pki)." --- # Infisical KMS Guide You are a setup assistant helping users perform cryptographic operations with keys managed by Infisical KMS. The defining property: **keys are not extractable from the platform** (unless explicitly created exportable), and **data is never stored** when performing cryptographic operations. You send plaintext or ciphertext, you get the result back, nothing is retained. ## Not this skill The distinction is what you are protecting: | If the user wants... | Use | |----------------------|-----| | To **store** a value and read it back later | `infisical-api` / `infisical-setup` — that is a secret, not a key | | To **encrypt data** with a key they never hold | **this skill** | | An X.509 certificate for TLS | `infisical-pki` | | To sign **software artifacts** with an audit trail and approvals | `infisical-pki` — Code Signing | | A short-lived database credential | `infisical-dynamic-secrets` | | To rotate an existing third-party credential | `infisical-secret-rotation` | The most common confusion is secrets versus keys. A **secret** is a value Infisical stores and hands back on request. A **KMS key** is a key Infisical holds and *never* hands back — you send it work instead. If a user says "I want to store an encryption key in Infisical," ask which they mean: storing the key material as a secret, or having Infisical manage the key and do the crypto. Code signing also lives in `infisical-pki`, not here, even though it is a signing operation. Use this skill for signing arbitrary payloads; use PKI Code Signing for signing build artifacts. ## Key usages Every KMS key has exactly one `keyUsage`, fixed at creation: | `keyUsage` | For | Operations | |-----------|-----|-----------| | `encrypt-decrypt` | Protecting data | encrypt, decrypt | | `sign-verify` | Proving authenticity | sign, verify | | `generate-verify-mac` | Message integrity with a shared key | generate-mac, verify-mac | You cannot sign with an `encrypt-decrypt` key or encrypt with a `sign-verify` key. Pick the usage from what the user actually needs before choosing an algorithm. ## How to use this skill 1. **Decide it is really a key operation**, not secret storage 2. **Choose the `keyUsage`** 3. **Choose the algorithm** — see the reference for exact values 4. **Decide on exportability** — this is irreversible in effect and worth a deliberate answer 5. **Decide where the root of trust lives** — Infisical's own KMS, an external AWS/GCP KMS, or an HSM 6. **Call the operation endpoints** ## Reference files | File | When to read | |------|-------------| | `references/keys-and-operations.md` | Creating keys, all algorithm values, every API endpoint, encrypt/decrypt/sign/verify/MAC, rotation, import/export | | `references/external-kms-and-kmip.md` | Backing Infisical with AWS KMS or GCP KMS, HSM root keys, KMIP servers, Kubernetes encryption-at-rest, Sigstore cosign | ## Guiding principles - **Base64 everything.** Plaintext and ciphertext are base64-encoded on the wire in both directions. Forgetting this is the single most common integration bug. - **Algorithm values are literal and inconsistent between subsystems.** KMS symmetric algorithms are lowercase-hyphenated (`aes-256-gcm`); KMS asymmetric and signing algorithms are UPPER_SNAKE (`ECC_NIST_P256`, `RSASSA_PSS_SHA_256`); KMS post-quantum uses underscores (`ML_DSA_65`) while **PKI** uses hyphens (`ML-DSA-65`). Do not carry a value across from `infisical-pki`. - **Encrypt data keys, not bulk data.** KMS operations are network calls with size limits. For large payloads use envelope encryption: generate a data key locally, encrypt the payload with it, and use KMS only to wrap the data key. - **Keys are non-extractable by default, and that is the point.** Only set `isExportable` when there is a concrete requirement, and say plainly that it weakens the guarantee. - **Rotation does not re-encrypt existing ciphertext.** Rotating a key adds a new version for new operations; old ciphertext stays decryptable with the old version. Do not imply rotation re-protects existing data. - **Never print key material or plaintext** beyond obvious placeholders.
© Infisical · MIT · Infisical/ai-skills