A security fleet that proposes fixes and waits for you

Run agents to review your code, dependencies and configuration under one spending cap. Get findings and proposed patches, with your approval before fixes merge.

Get early access Read the docs

Updated 2 November 2026

An agent pod at the center of six concentric containment rings, its credential key kept outside, the innermost ring marked with an orange notch. — FIG. 1: Reviews on a schedule

Trigger: Daily 03:00 (example)
Runtime: Codex × 3
Budget: $6 shared cap (example)
Needs approval: pulls.merge

Reviews code, dependencies and config. You approve the fixes.

Security reviews are easy to postpone when the repository keeps changing. Give separate agents the code, dependencies and configuration to review, then bring their candidate findings together for a person to assess.

01Setup

How is the security review set up?

Run the agent you already use against a repository you own or are authorized to review. In this example, three Codex agents share one budget and work in separate environments. One reviews code paths, one reviews dependencies and one reviews configuration. Bring the review tools each task needs.

The example runs daily at 03:00 with a shared $6 cap. These are illustrative settings, not plan limits or a quoted review price. The agents can read the repository and open proposed fixes. Merging requires a person.

fleet.yamlExample

fleet: security-review
runtime: codex
goal: "Review our repository's code, dependencies and configuration"
agents: 31
triggers:
  - schedule: "0 3 * * *"
tools:
  github: { allow: [contents.read, pulls.create], require_approval: [pulls.merge] }2
budget: { fleet_usd: 6 }3

Line 4 Three isolated agents review code, dependencies and configuration under one shared budget.
Line 8 Read the repository and propose changes. Merging a fix requires your approval.
Line 9 An illustrative $6 cap for the whole review, not a separate allowance for each agent.

FIG. 2: The config file, line by line

02Compared with cron

What changes compared with a cron job?

Security fleet: a cron job on a VPS compared with Heighliner
Dimension	Cron on a VPS	Heighliner
Scope	Whatever the script's credentials allow	Only the repositories and tools you grant
Isolation	Scanners share the host	Each agent has an isolated environment
Spend	Whatever the model key allows	One shared dollar cap
Changes	Whatever the script can push	Proposed patches; merges wait for you
Record	Whatever the script logs	Findings, proposed changes, approvals and cost in one receipt

03Where it stops

Where does the review stop?

The review stays inside the repositories and tool permissions you grant. It does not include exploit attempts, scanning unrelated systems or access to production infrastructure.

Agents may be wrong. Review each candidate finding and its evidence before accepting a proposed patch. Changes to dependencies or configuration can affect how an application behaves, so merging a fix waits for your approval. This example grants no deployment access.

If the shared budget is reached, the agents pause with their work preserved. Raise the cap when you want to continue.

security-review · shared budget (example)$0.31 / $6.00

CAP $6.00

50% warn · $3.00
75% warn · $4.50
90% warn · $5.40

Warnings at 50, 75 and 90 percent. At the cap, agents pause; nothing is deleted.

PROPOSED FIX · NEEDS YOUR APPROVALsecurity-review · example

Merge a proposed dependency or configuration fix into the reviewed repository.

github · pulls.merge · requires approval

FIG. 3: Warnings, the cap and the held action (example)

04Receipt

What does the report include?

Ask the agents for affected files, supporting evidence, suggested fixes and anything they could not verify. Review the report alongside the proposed patch rather than treating an agent's severity label as a confirmed vulnerability.

The run receipt records what ran, what it cost and which actions waited for approval. The receipt shown here is a sample. Heighliner provides the runtime boundaries and record, not a certification or a promise of complete security coverage.

Run receiptrun_security_example

03:00:00Started 3 agents on the approved repository$0.00
03:01:00Reviewed code paths for potential security issues$0.08
03:02:00Reviewed dependencies and recorded candidate findings$0.11
03:03:00Reviewed configuration within the granted scope$0.10
03:04:00Prepared findings and a proposed patch for human review$0.02
03:04:10Merge held for approval; no fix deployed$0.00

Total$0.31

sha256 3b9e…c04d · signed

FIG. 4: One run, itemised (sample)

§ 05Questions

Questions

Does the security fleet change production on its own?

Not in this example. It reads the repositories you grant and can open proposed fixes. It has no deployment access, and merges require your approval.

Does Heighliner include a security scanner?

Heighliner runs the agents and tools you bring. This is an example workflow, not a built-in scanner or a guarantee that every vulnerability will be found.

Can I limit what the agents review?

Yes. Grant only the repositories and tools needed for the review. The example does not include exploit attempts, external target scanning or production access.

06More use cases