Akira
An autonomous offensive-security agent that plans an engagement, runs reconnaissance, chains exploits and writes a report end to end from inside Claude Code, Gemini CLI, Codex or Cursor.
Akira is a pentesting co-pilot that runs natively inside an existing coding agent rather than wrapping automated scanners, structured as a reasoning system aimed at logic flaws, architecture leaks, supply-chain vectors, cryptographic weaknesses and chained attacks that scanners tend to miss. It moves through a fixed pipeline of planning an engagement, reconnaissance, secrets discovery, exploitation, deeper vulnerability hunting, triage and reporting, and its core constraint is that every finding needs a reproducible HTTP response body containing sensitive data.
To keep token costs down on long engagements, a router loads only the technique files a phase's hypotheses call for, re-prioritizing what to load next as intelligence comes back from earlier phases, which the project measures as cutting per-engagement token cost by roughly two thirds compared to loading full monolithic skill files.