cat security.md

We find the weak points before someone else does

We review code, applications, infrastructure and AI systems for vulnerabilities, explain each finding in terms your engineers can act on, and check the fixes afterwards. Every engagement starts with written authorization from the system owner.

Services

Five kinds of security work

Each can be ordered separately. They are often combined: a code audit before a release, a penetration test after it, and checks in CI so that the same issues do not return.

01

Code security audit

We read the code by hand and run static analysis alongside it. Tools find the common patterns; manual review finds flaws in logic and access control that tools miss. Findings are ranked by risk, and each comes with a concrete fix.

  • Manual review. Authentication, authorization, input handling, cryptography use and business logic.
  • Static analysis. Tool results are triaged by hand, so the report contains confirmed issues and not raw scanner output.
  • Dependencies. Third-party libraries with known vulnerabilities and how they are reachable from your code.
  • Ranked findings. Each issue has a severity, the affected code location and a specific recommended fix.
02

Penetration testing

We test web applications, APIs and infrastructure the way an attacker would, within a scope agreed in writing. The mode is chosen with you: black box with no prior knowledge, grey box with user accounts and documentation, or white box with access to the source.

  • Web applications. Injection, broken access control, session handling, client-side issues and logic flaws.
  • APIs. Authorization at object and function level, input validation, rate limiting and exposure of excess data.
  • Infrastructure. Exposed services, misconfiguration, outdated software and paths for lateral movement.
  • Agreed mode. Black, grey or white box, with testing windows and limits fixed before the start.
03

Security checks in CI

We wire automated checks into your pipeline so that issues are caught before release, not after. The checks are tuned to your codebase so that they report real problems and the team does not learn to ignore them.

  • Static analysis. Rules matched to your languages and frameworks, run on every merge request.
  • Dependency scanning. Known vulnerabilities in direct and transitive dependencies.
  • Secret scanning. Keys, tokens and passwords caught before they reach the repository history.
  • Container image scanning. Base images and installed packages checked before an image is published.
04

Cloud and configuration review

Many incidents start with configuration, not code: a role with too many rights, a storage bucket open to the internet, a secret in an environment variable. We review how your cloud accounts and clusters are set up and show what to tighten.

  • IAM and access rights. Roles, policies and service accounts checked against least privilege.
  • Network exposure. What is reachable from the internet, and what is reachable between internal segments.
  • Secrets storage. Where secrets live, who can read them and how they are rotated.
  • Kubernetes configuration. RBAC, network policies, pod security settings and access to the cluster API.
05

Security of AI systems

LLM features add risks that ordinary application testing does not cover. A model follows instructions hidden in the data it reads, and an agent with tools can be steered into actions its user never asked for. We test these paths in your assistants, agents and RAG systems.

  • Prompt injection. Direct and indirect: through user input, documents, web pages and tool results.
  • Data leakage through LLMs. System prompts, other users' data and internal documents appearing in answers.
  • Unsafe tool use by agents. Which actions an agent can be made to take, and with whose permissions.
  • Access control in RAG. Whether retrieval respects the document permissions of the person asking.
Process

How an engagement runs

Four steps. Nothing is tested until the first one is signed.

  1. Scope

    Written authorization from the system owner, the list of targets, and the rules of engagement: what is allowed, when, and who to call if something goes wrong.

  2. Testing

    Manual and automated work within the agreed scope. A critical finding is reported to you immediately, not held back for the final report.

  3. Report

    A written report with findings ranked by severity, followed by a call with your engineers to go through each one.

  4. Retest of fixes

    After your team fixes the issues, we check each fix again and record the result in a retest report.

Deliverables

What you receive

The report is written for two readers: the person who decides what to fix first, and the engineer who does the fixing.

01

Executive summary

One or two pages in plain language: what was tested, the overall state, and the main risks for the business.

02

Technical findings with severity

Each finding described separately, with a severity rating and an explanation of what an attacker could do with it.

03

Reproduction steps

Exact requests, parameters and conditions, so that your engineers can reproduce the issue and confirm it themselves.

04

Fix recommendations

A specific fix for each finding, tied to your stack and code, with a suggested order of work.

05

Retest report

After the fixes: the status of every finding, marked as fixed, partly fixed or still open.

Principles

Rules we do not bend

Security testing means being trusted with access. These are the terms we work under on every engagement.

  • Written permission only. We test a system only with the written permission of its owner. Without it, we do not start.
  • Confidentiality. We sign an NDA before receiving any access or documentation.
  • Findings go to the client only. We do not disclose or publish what we find to anyone else.
  • Minimal access and data handling. We ask for the least access the work needs, touch as little real data as possible, and delete what we hold when the engagement ends.
  • No destructive testing in production. Anything that could affect availability or data in production is done only with your explicit agreement.
Contact

Tell us what needs checking

Write to security@kv.solutions or use the contact form. A few lines are enough: what the system is, what worries you, and when you need the result.