Endor Labs: AURI Agents

What it is

AURI Agents are Endor Labs' pre-built security agents for coding-agent hosts (Claude Code, Codex, Cursor and similar). They are free for every Endor Labs customer, and the page says they are available now. The companion Endor Labs Agent Kit is an open-source GitHub repo with the agent definitions and tooling for building your own.

How it works

• Execution: agents run on your side, using your LLM keys and your source-control credentials (gh for GitHub, glab for GitLab). The page says code and secrets never cross into Endor's boundary and token spend stays yours. Telemetry does flow back to the Hub for central visibility.
• Authentication: it goes through your local endorctl configuration, so existing users have nothing new to set up.
• Remaining catalog: besides the four agents named in the summary, there are six more:
• Remediation Planner, which previews fixes without opening PRs
• Upgrade Impact Analysis
• Dependency Decision Helper
• Package Risk Summary
• Repository Dependency Reviewer
• Vulnerability Explainer
• Chaining: agents can be chained together and combined with your own.
• Exception flow (AI SAST Triage): a developer requests an exception in a PR comment, and an AppSec reviewer approves it straight into policy.
• Authoring: a "Create Endor Labs Agent" skill drafts the recipe, prompt, eval cases and architecture diagram from a plain-English description. It then validates the result against the same guardrail checks Endor's own agents pass.
• Contributing: you can keep a custom agent private or open a PR. Endor reviews submissions, and accepted ones are pushed to all supported hosts by the build pipeline.

What to do

The page's Claude Code example, installing from the agent-kit repo:

mkdir -p .claude/agents
cp /path/to/endor-labs-agent-kit/claude-code/sca-remediation/sca-remediation.md \
  .claude/agents/sca-remediation.md

Then invoke it:

@agent-sca-remediation check this repository for P0 SCA findings I can start remediating

Alternatively, install as a plugin or extension by following the instructions in the ai-plugins repo.

Caveats

• Vendor claim: Endor says expert agents do the same tasks as an unaugmented agent 2.8x faster with 92% fewer tokens. This comes from its own published study, which the page does not detail, so treat it as unverified.
• Count mismatch: the page says "ten of them at launch" in one place and "twelve agents" in another.
• Attribution not shipped: scoped-credential agent attribution (to tell a person from their agent) is promised as a fast follow-up, not available now. The same goes for validated, checksummed artifacts with provenance. Until then, mutating actions run under your own credentials.
• Permissions: most agents are read-only by default. Mutating ones declare their permissions and require confirmation. Check which agents you enable, since SCA Remediation opens PRs and the exception workflow writes policy.
• Customers only: the agents draw on Endor's reachability, upgrade-impact and policy data, so they presumably need an existing Endor Labs deployment to be useful.
• Exploitation timing: Endor says it has seen CVEs go from disclosure to active exploitation in under 10 hours. The page gives no data behind this.