Flounder Security Auditor

What does this agent do, and when should you use it?

Flounder is an end-to-end security audit system built around modern coding agents, with a CLI, React dashboard, REST API, pi extension, and execution daemon. It can prepare material from a repository, source tree, package, project link, transaction, or address, then map scope, dig into code, compose cross-scope candidates, and verify them through local execution. Model-generated code, PoCs, dependency installs, and tests run in a copied workspace; the default OCI sandbox restricts capabilities and can disable networking for audit commands. The system tracks scopes, evidence, findings, and reports, and its confirmation phase can reproduce findings against real-target ground truth. Flounder is intended for public-source or authorized targets; project code and provider credentials stay on the executor host, while the chosen model provider may require authentication.

With flounder run <clue>, Flounder can prepare target source and materials, then run the map, dig, synthesize, verify, confirm, and report workflow; --source starts directly from staged code. The model reads source and optional project documentation, specifications, and historical materials, creates a scope inventory, and writes and runs PoCs, tests, or harnesses in an isolated workspace. Findings need passing local command evidence; the confirmation phase can reproduce them against a real target or local fork. The CLI, dashboard, and REST API expose run status, scopes, findings, execution evidence, and Markdown reports.

1. A protocol or contract team can run a blind audit of public-source or authorized code and assess findings through execution evidence.
2. A bug bounty researcher can use Prepare to collect public project materials, deployments, and scope information, then assess findings against program requirements.
3. An incident responder can start from a suspicious transaction, address, or incident link and test whether the root cause reproduces against local ground truth.
4. An auditor can import existing claims or select a file, region, or scope for targeted verification with verify or audit --scope.
5. A Flounder maintainer can explicitly enable maintainer mode to run constrained candidate experiments based on Evaluation failures.

How do you install or deploy this agent?

Use Node.js 24 LTS. To install from source:

nvm use
npm install
npm run build
npm run sandbox:build

You can also install the Flounder skill for Codex or Claude Code:

npx skills add adshao/flounder --skill flounder -g -a codex -a claude-code

Audit runs need an available sandbox runtime. The default Docker backend needs Docker or a compatible runtime; Apple container is also supported on Apple silicon macOS. Authenticate the chosen model provider on the execution daemon.

How do you use this agent?

Start the control plane and dashboard:

node dist/cli.js ui

Log in to and check the provider on the execution daemon:

flounder daemon provider login openai-codex
flounder daemon provider check openai-codex

Then create a dashboard project, enter an audit task or clue, choose its execution daemon and provider profile, and start a run. To run the full workflow from a clue:

flounder run <tx-or-address-or-project-or-repo-or-link>

To audit local source directly:

flounder run --target my-target --source ./src --build-root . --corpus ./docs

openai-codex uses pi subscription/OAuth authentication. For other providers, configure OAuth or an API key on the execution daemon using the model setup guide.

What are this agent's strengths and limitations?

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

How does FollowAgents rate this agent?

FAQ