Agent skills supply chain · Part of the Elixir Platform

Skills are your agents’ newest dependency. Govern them like it.

ElixirHub scans, tests, signs, and version-pins agent skills before they reach a workspace. Follow each version to the decisions it shaped, so auditors can trace it end to end.

See how a skill reaches your workspace
Example release manifestfinance/three-way-match
Official
Version
2.4.1 · pinned, immutable
Publisher
Verified · key attested
Security scan
No findings
Tests
Golden set passed · no regression
Signature
SHA-256 content hash
Decisions traced
Linked in ElixirData
An illustrative release for a three-way invoice match.

The problem

What comes with a skill install?

Skills bring new instructions and sometimes executable code. Before a skill reaches production, teams need to know what it can access, which version will run, and who approved it.

Why skills need a supply chain
Skills run with your agent’s access
A bundled script can reach the repositories, internal APIs, and data available to the agent running it.
Copies lose versions and origin
A skill pulled from Git can change under the same name. Pinning preserves the version you reviewed.
Public catalogs can carry malicious skills
A listing may contain untrusted instructions or scripts. Review the package before granting it access.

How it works

Five gates from draft to install.

  1. Draft

    The publisher builds the skill and declares its tools and inputs in the manifest.

  2. Scanned & tested

    Review the package and run sandbox, functional, adversarial, and regression tests.

  3. Signed

    Bind the publisher key and attestations to the release’s content hash.

  4. Published

    Make the immutable version discoverable by trust tier and domain.

  5. Installed

    Check that the release’s tier meets workspace policy.

Skill testing

Scan for risk. Test for the work the skill has to do.

Evaluate each version before signing. Repeat the tests when models, tools, or test cases change to catch regressions before rollout.

Sandbox run

Run scripts in isolation and flag access the skill did not declare.

Golden-set tests

Score each test case against expected results and declared failure modes.

Adversarial tests

Check whether misleading inputs cause a skill to expose data or exceed its permissions.

Your own test cases

Evaluate private workspace cases and catch regressions before rollout.

Why ElixirHub

What you can show an auditor.

See who approved the release, what it was tested against, and where it influenced a decision.

Provenance
See the publisher, the release signature, and the immutable version that was approved.
Proof
Review scan findings and test results together to see how a skill meets your controls.
Policy at install
Set the trust tiers each workspace can accept. Community skills stay off by default.
Traceability
Follow the skill version to the agent decisions it shaped in ElixirData.

Part of the Elixir Platform

ElixirHub governs the skill. The platform governs what happens next.

Three products cover the registry, the agent’s actions, and the record of what was decided.

  1. ElixirHub

    What the skill is

    Publish, scan, evaluate, sign, and version skills before teams install them.

  2. ElixirClaw

    What the agent may do

    Apply permissions, decision boundaries, and guardrails at the Agent Gateway.

  3. ElixirData

    What was decided, and why

    Use the Context Graph to trace decisions back to the skill version that informed them.

Trust tiers

Three tiers. Your workspace decides.

Set which skill releases a workspace can install based on who published them and what evidence they carry.

Official

Regulated default

  • Published and signed by ElixirHub
  • Architecture review
  • Full evaluator suite
  • 12-month deprecation notice

Verified Partner

Opt in

  • Signed partner agreement
  • Partner key attestation
  • Evaluator suite
  • Continuous vulnerability scanning

Community

Off by default

  • Open publishing
  • Publisher-owned signature
  • Scanned before listing
  • Enabled per workspace

Compare

From storing artifacts to governing agent skills.

Scroll to see the full comparison.

CapabilityTypical artifact registryElixirHub
Version, signing, and access controlCore registry functionsPinned releases and workspace policies
Security scan before installPackage scanningScan evidence alongside release metadata
Functional, adversarial, and regression testsExternal test toolingEvaluation evidence by release
Domain golden setsExternal evaluation toolingExpected outcomes for regulated workflows
Decision traceabilityArtifact metadataLinked to the decision record in ElixirData

Built for regulated work

Explore skills for the work auditors check.

Banking & financial services

Reconciliation, KYC review, SOX and IFRS controls, and materiality checks.

Explore skills

Government services

Bilingual notices, case evidence, and licence checks.

Explore skills

Healthcare & life sciences

Knowledge search, documentation quality, and operations workflows.

Explore skills

Security & SRE

Log correlation, alert triage, and dependency analysis.

Explore skills

For security and AI leaders

Govern the skills your agents already run.

Map where skills come from, what runs in production, and which controls to put in place first.

Book an assessment

For skill publishers

Reach workspaces that install reviewed skills.

Prepare your skill and submit a version for review in the ElixirHub catalog.

Publish to ElixirHub