Assessments
Every engagement starts with written authority and ends with evidence, limitations and a clear human decision.
A one-time automated snapshot for a team shipping an agent or MCP server. Start with one agreed workflow: its tools, retrieval, memory, network access and action permissions. Supply an authorized repository, code bundle or architecture description; redacted test traces can add context.
Deterministic comparison of AI-agent or MCP architectures under declared compromise, outage and recovery disturbances.
Controlled investigation of an owned or explicitly authorized web application. Scope, rate, exclusions and state-changing actions are agreed before execution.
Repository-focused analysis that cites exact source locations and treats patterns as hypotheses until impact is demonstrated in an authorized environment.
Review of agent tools, MCP interfaces, retrieval, memory and authorization boundaries for instruction injection and unsafe authority transfer.
Independent assessment of an existing security claim, including evidence quality, reproducibility, realistic impact and scope validity.
Clear technical reporting for engineering teams, leadership or an authorized disclosure route. Reports never submit themselves.
Fit and authorization
From assessment to a practical control
Inspect the open-source proof-of-work gate and its deployment assumptions. Our SVG illustration explains single-use acceptance; an assessment can examine whether the control fits your endpoint.