Who Dokima suits
Dokima vs prompts and skills
A prompt is one look at your code. Dokima keeps looking.
Asking Claude to review your code, or wrapping that request in a skill, gets you one pass. It might find something real, or it might miss everything that matters, and you can't easily tell which. Run it again and you'll get a different answer. Dokima uses the same kinds of models, but runs them as an ongoing review process that remembers what it has covered and what it found, then plans the next piece of work from there.
Where it works well
A prompt or skill has a useful job to do.
- Asking a quick question about one file, function or diff.
- Getting a second opinion on code you're already looking at.
- Exploring an unfamiliar codebase for a few minutes.
- One-off checks where missing something isn't costly.
Where Dokima differs
Context and review, not just another pass.
- 01It isn't a single shot. A prompt or skill looks at whatever happened to fit in its context that time. Dokima works through the repository component by component, from several security perspectives, then returns for deeper passes.
- 02It remembers. Findings, component maps, architecture notes and run history stay in the workspace. A skill starts from nothing every time and doesn't know what it reported last week.
- 03You can see what has been covered. Initial Sweep, In-depth Sweep and per-component progress show what has been reviewed and what hasn't. A prompt gives no sign of what it never looked at.
- 04Findings are checked before you see them. Candidates are investigated and challenged in a separate validation step. A single prompt reports its first impression with the same confidence whether it's right or wrong.
- 05Results stay stable. Findings get stable fingerprints, so the same issue isn't reported again under a new name, and a finding is never dropped just because a later run didn't spot it. Only a developer can retire one.
- 06It runs unattended. It paces itself, survives reboots and keeps working for weeks or months without anyone babysitting a terminal.
- 07It works with your choice of model. Dokima runs on Claude, Codex, OpenCode, Pi or open models, and switching runners doesn't lose your history.
Capability comparison
CapabilityDokimaA prompt or skill
RunsContinuous, scheduled reviewOne session, when someone remembers to run it
CoverageTracked per component and perspectiveWhatever fit in context that time; not recorded
MemoryFindings, maps and run history persistNone between runs
ConsistencyStable fingerprinted findings with deduplicationDifferent results each run; duplicates and silent misses
ValidationSeparate investigation and validation stagesReports its first impression
OutputEvolving report and private team publishingA chat transcript
Use them together
Use Claude directly for quick questions and fixes. Use Dokima for the systematic review that keeps track of everything.
See what Dokima finds
Put an AI security reviewer to work on your codebase.
Start a full-product 21-day trial for your engineering team. No payment card required.