Day 1
01
Scope + Kickoff
We agree on protocol boundaries, agent roles, and the threat model. You hand over read-only access. We sign an NDA before touching anything.
Most AI protocols ship fast and discover their security gaps the hard way — during an incident. We run systematic active audits on your OpenClaw agent infrastructure: inter-agent channels, tool permissions, session handling, and prompt injection surfaces. You get a findings report with PoCs and explicit fix instructions before your users find the holes first.
Active audit
2-Week Engagement
Sample output only
Every audit is scoped to your protocol. Findings vary based on your agent architecture and threat model.
We don't just look at the surface. We test the full attack surface of your agent system.
How it works
01
We agree on protocol boundaries, agent roles, and the threat model. You hand over read-only access. We sign an NDA before touching anything.
02
We run systematic tests across every agent, tool, and inter-agent channel. We probe privilege boundaries, injection vectors, session handling, and failure modes.
03
A structured report with severity ratings, PoC reproduction steps, and explicit remediation guidance for each finding. Delivered as a ranked markdown doc + live walkthrough.
04
We implement the fixes directly, or work with your engineering team to validate their patches. Your choice.
Sample output
Every finding includes severity, description, proof of concept, and explicit remediation steps your team can act on immediately.
Agents configured with overbroad permissions can invoke actions outside their intended scope — including wallet moves, secret reads, and external API calls.
Remediation
Scoped IAM policies + mandatory approval gates on high-risk tool calls.
Internal agent-to-agent communication carries no message-level authentication. A compromised agent can spoof messages to other agents at will.
Remediation
Signed message envelopes + mutual TLS on the internal bus.
Agent sessions persist for days. Token theft via log exposure or a compromised sub-agent grants indefinite access.
Remediation
Short-lived JWTs (≤1hr) + automatic rotation via the OpenClaw auth layer.
Recursive agent loops have no hard cap. Under adversarial prompt conditions, agents can spin indefinitely — racking up compute cost and locking other agents out.
Remediation
Depth/gas limits + circuit breakers per agent role.
API keys and credentials appear in full in agent working memory, tool logs, and session transcripts — exposing them to any agent with read access.
Remediation
Secret masking at the tool layer + per-agent secret scoping.
Agent actions are not logged with sufficient granularity to reconstruct decision chains during an incident.
Remediation
Structured agent event log with causal chain tracing.
Downstream API rate limits hit without per-agent attribution make it impossible to identify the offending agent quickly.
Remediation
Per-agent request counters + automatic backpressure signaling.
Agents named generically make audit logs and permission traces harder to read under pressure.
Remediation
Canonical naming schema + identity hierarchy in the agent manifest.
Starter
$1,499
For protocols with a single agent or 2–3 agents sharing a simple toolchain.
Standard
$2,999
For protocols with complex inter-agent communication and multiple tool chains.
Enterprise
Custom
For protocols with custom auth, external integrations, and hardened security requirements.
Get started
Tell us about your protocol. We'll scope the engagement, confirm timelines, and send a proposal within 24 hours.