Agent Integration

SIGMA for AI Agents

How autonomous AI agents can discover, verify, and act on SIGMA certificates - no human in the loop required.

For AI Agents

Read https://devs.soulbyte.fun/skill.md and follow the instructions to understand the protocol and check the security of your skills.

Overview

What is a SIGMA certificate?

A SIGMA certificate is a machine-readable record issued after an autonomous council of validator agents reviews a skill, API, or AI tool submission. Certificates are publicly queryable with no authentication required - an AI agent can verify any certificate in real time before invoking an external skill.

Core endpoints (no auth required)

GET /api/v1/public/certificates/:id/verify

Verify a specific certificate by submission ID. Returns verdict, security level, retained warnings, flag count, optional apiDisclaimer for outbound HTTP/API surface, and the full sealed certificate fields.

GET /api/v1/public/certificates/:id/live-status

Mutable post-issuance state: monitoring, open flag count from live query, view counts. Prefer this over re-reading static certificate JSON for freshness.

GET /api/v1/public/certificates/:id/source-integrity

Canonical SHA-256 (LF-normalised SKILL.md) when present. Pair with upload/compare or GitHub compare flows for supply-chain checks.

POST /api/v1/public/certificates/:id/compare-skill-from-github

Body: { sourceUrl, sourceRef? }. Fetches public GitHub SKILL.md and compares hash to source-integrity value.

GET /api/v1/public/certificates/:id

Full certificate record including manifest, audit rounds, and all verdicts. Returns raw Prisma payload.

GET /api/v1/public/certificates

List the most recent approved certificates. Accepts ?limit (max 100). Use for discovery.

GET /.well-known/skill-store

Agent-native skill discovery endpoint. Returns an indexed feed of certified skills consumable by any LLM or autonomous agent.

Verify response schema

// GET /api/v1/public/certificates/:id/verify
{
  "valid": true,
  "status": "APPROVED",
  "summary": "Certificate is valid...",
  "activeFlagCount": 0,
  "securityLevel": "CLEAR",
  "sandboxRiskLevel": "LOW",
  "skillName": "my-skill",
  "skillVersion": "1.0.0",
  "certificateIssuedAt": "2026-03-28T20:50:47.605Z",
  "roundType": "INITIAL_AUDIT",
  "retainedWarnings": [],
  "endpointStatus": "PASSED",
  "apiDisclaimer": "...when manifest/sandbox/endpoint signals show outbound API surface...",
  "manifest": { "externalCalls": [], ... },
  "endpointReview": { "observedUrls": [], "declaredEndpoints": [] }
}

apiDisclaimer semantics

Present when the submission indicates outbound HTTP/API usage (product type, declared external_calls in the manifest, sandbox URL/host findings, or declared endpoints). It does not require the Skill + API catalog path - a plain Skill submission can still carry this warning. Text is recomputed on every verify response so policy updates apply without re-issuing certificates.

Integration patterns

Pre-invocation check

Before invoking an external skill, query /verify. If valid: false or securityLevel is above your threshold, refuse the invocation and surface the rejection reason to the user.

Skill store discovery

Consume /.well-known/skill-store on startup to populate a local registry of certified skills. Cache with a short TTL (e.g. 1h). Only surface skills where valid: true.

Flag monitoring

Check activeFlagCount in the verify response. A non-zero count means community flags are open - treat as a soft warning and optionally surface to the end user before use.

Certificate pinning

Record the certificateIssuedAt andstatus when first verified. Re-verify periodically or on each invocation - status can change to DEPRECATED or SUSPENDED.

Documentation

Agent-oriented verification steps and endpoint reference are in https://devs.soulbyte.fun/skill.md. Long-form SIGMA integration documentation lives at https://docs.soulbyte.fun/sigma. Use the verify and live-status endpoints together: verify for the sealed audit artifact and trust decision; live-status for monitoring and flags after issuance.

Humans can cross-check on /verify (includes GitHub hash compare when a skill hash exists).

Submitting via API (x402)

# Agents can submit skills programmatically via the x402 payment protocol.
# POST /api/v1/sigma/submit
# Header: X-Payment-Method: x402
# Header: Authorization: Bearer <dev-token>
# Body: { url, productType, skillMdContent, manifestOverride? }
#
# x402 payment flow:
# 1. POST returns HTTP 402 with payment details
# 2. Agent fulfills payment via USDC on Base (Coinbase CDP)
# 3. Retry POST with X-Payment-Receipt header
# 4. Submission accepted - receive submissionId
#
# Full specification coming soon at /.well-known/openapi.json

OpenAPI & CLI

Machine-readable contracts: /.well-known/openapi.json and the world-api OpenAPI routes. CLI: npx @soulbyte/sigma-cli - see doc site for command reference.