Skill discovery and verification for coding agents. We index high-quality skills wherever they live, rank them so your agent picks the right one, and let you pin and verify what you install.
sigil + index — a spellbook of agent skills
Describe what your agent needs in natural language. Sigildex searches agent-skill SKILL.md repos and other leading sources, ranks by semantic relevance and trust signals including install counts, freshness, and safety scoring, and returns each match with a nested safety summary and a reviewable install command. Inspect the top pick, pin what you install, and verify it on every load. Index syncs daily.
{ "query": "PDF text extraction", "limit": 3 }
queryNatural-language description — semantic match, not keywords.limitHow many ranked picks to return.scoreSemantic relevance combined with trust and safety descriptors.pinInspect returns a lockfile-ready pin — verify it on every load.Sigildex is a search engine for skills, built for AI agents. It returns ranked, inspection-ready results in a single request — and keeps them trustworthy after install.
A single /discover request returns a ranked list. The DIY equivalent — paginate GitHub, parse SKILL.md frontmatter, score, dedupe, rerank in your agent — burns latency, API calls, and LLM tokens on plumbing instead of answers.
/discover returns ranked results in a slim picker shape: a nested safety object (status, recommended_action, executes_code, safety_flags), a verification_level, publisher fields, and quantitative signals (stars, installs, freshness). /inspect expands a candidate into full SKILL.md content, manifest, provenance, install options, and a lockfile-ready pin. /verify re-checks that pin by content hash or source URL — free, on every load or in CI.
Every result carries a relevance score, trust signals (install counts, freshness, safety scoring), and a reviewable install command. Your agent compares candidates, inspects the top picks, explains the choice, and installs after user confirmation.
No signups, no API keys, no contracts. 50 discovery workflows per day per IP, free. For higher-volume usage, x402 lets agents pay $0.002 per discovery workflow without account setup.
{
"mcpServers": {
"sigildex": {
"url": "https://sigildex.ai/mcp"
}
}
}
$ curl -X POST https://sigildex.ai/discover \ -H "Content-Type: application/json" \ -d '{"query": "PDF extraction", "limit": 3}'
$ npx @sigildex/cli search "PDF extraction"
--json for structured output.No accounts. No API keys. No subscriptions. Just search.
One workflow = 1 ranked skill search plus shortlist review for up to 5 candidates within an hour. Pay as you go. No signup, no API keys.
/inspect is free when using a query_id from /discover (1h TTL)/verify is always free — re-check a pinned skill by content hash or source URL, no accountHand any of these to your AI tool. Sigildex describes itself; the agent integrates.
Just point your agent at llms.txt — every other file is linked from there.
Publicly available agent-skill SKILL.md repos and other leading sources, with new sources added on a rolling basis. The index updates daily.
An open standard for HTTP-native payments. A paid endpoint returns HTTP 402 with payment requirements; the client signs a USDC payment and retries. Wallet is identity — no accounts, no API keys. Sigildex settles on Base; spec at docs.x402.org.
No. 50 discovery workflows per day per IP, no setup required. Verification is always free. Beyond the free discovery tier, give your agent an EVM wallet with USDC on Base. See the payments guide.
Semantic relevance combined with trust signals — install counts, freshness, safety scoring. Identical ranking whether free or paid.
safe safety score mean?It means Sigildex's combined score for the skill (its SKILL.md content, name, description, tags, and metadata signals) landed in the lowest-risk band. It does not mean the skill is safe to run: discovery-time review can miss novel variants or runtime-only behavior, and bundled script contents, dependencies, and install-time behavior are not analyzed. Reject any result flagged block, inspect and review the rest, and review bundled code before install — runtime safety is the agent's responsibility.
Store the pin block returned by /inspect next to the installed skill, then call /verify on load or in CI with the content hash or source URL. If upstream content drifts or the safety verdict changes, the response changes — trust is re-checked, not assumed. /verify is free, with no account or payment.