// Agent profile
iFixAi (ifixai-ai/iFixAi) is an Apache-2.0 diagnostic tool that independently audits AI agents and models against operational-misalignment risks. It runs 45 inspections — 32 graded core across five pillars (fabrication, manipulation, deception, unpredictability, opacity) plus 13 extended inspections for frontier risks like sabotage, sandbagging, and oversight evasion — and returns an A–F letter grade with a scored pillar scorecard in under 120 seconds. iFixAi is model and industry agnostic: point it at any agent endpoint, and it evaluates whether the agent is doing the job it is supposed to do based on business KPIs and organizational structure. Available as a CLI with guided wizard, CLI with explicit flags for CI pipelines, or as a plugin/skill for interactive use inside an agent session. With 18.2K+ GitHub stars, it has become the leading open-source tool for AI agent quality assurance.
iFixAi is an independent auditing tool that catches AI agent mistakes and blind spots before customers or regulators do. Rather than relying on model benchmarks or generic red-teaming, iFixAi strikes a balance between AI Red-Teaming and Operational Assurance — evaluating whether an agent is actually doing its job based on the business context it operates in. The tool runs structured inspections across five pillars of misalignment risk and produces a scored scorecard with a clear letter grade.
The core inspection suite covers 32 graded checks across five pillars: fabrication (does the agent make things up?), manipulation (does it try to influence users inappropriately?), deception (does it hide its reasoning or capabilities?), unpredictability (does it behave erratically under pressure?), and opacity (is its decision-making transparent?). Beyond the core suite, 13 extended inspections probe frontier risks — sabotage, sandbagging (deliberately underperforming), and oversight evasion — that matter as agents gain more autonomy. The entire run completes in under two minutes and returns an A–F grade with per-pillar scores.
iFixAi offers three ways to run: a guided CLI wizard that walks first-time users through setup, a CI-ready mode with explicit flags for pipeline automation, and a plugin/skill mode for running audits interactively inside an agent session with an interactive scorecard.
Teams deploying AI agents in production use iFixAi as a pre-deployment gate — run the audit before shipping agent updates and catch regressions in safety behavior. Compliance teams use the scorecard output to document agent behavior for regulatory review. Developers building agent systems use it during development to test how their agents handle adversarial inputs and edge cases. The CI integration makes it practical to run iFixAi on every commit that touches agent prompts or tool definitions.
iFixAi audits agent behavior through structured probes — it tests what agents do, not how they work internally. The inspections are designed for agents that interact through text interfaces; agents with primarily non-text modalities may need additional evaluation. As with any automated testing tool, passing iFixAi inspections is necessary but not sufficient for production safety — it complements rather than replaces human review and monitoring.
iFixAi is for any team shipping AI agents that interact with users, customers, or critical systems. Engineering teams that want automated safety regression testing in CI, compliance teams that need documented agent behavior audits, and anyone building agentic systems who wants a fast, structured answer to “is this agent safe to deploy?”
Local-first session intelligence and analytics platform for coding agents, supporting Claude Code, Codex, and 20+ other agents.
817 structured cybersecurity skills for AI agents mapped to MITRE ATT&CK, NIST CSF 2.0, MITRE ATLAS, D3FEND, NIST AI RMF, and MITRE F3.
Isolated-VM agent sandbox with sub-100ms boot — run AI agents on real machines with their own kernel, filesystem, and network. Pure Rust, open source.