Skip to content

symbi-codered

Governed AI source-code auditor. Multi-language SAST + LLM-driven reasoning + sandboxed PoC validation + devil's-advocate rebuttal, with an evidence chain you can audit.

Produces SARIF + Markdown + a signed engagement-seed.json handoff that downstream consumers (e.g. symbi-redteam) can ingest to drive exploit validation.

What it is, in one paragraph

Run codered hunt against a repo. Static analyzers (semgrep, bandit, clippy, gosec, eslint, checkov, trivy, …) produce raw findings. A tree-sitter dataflow extractor builds dataflow_edges; a mechanical taint tracer walks them. Four LLM agents — pattern_scout, chain_builder, poc_forge, devils_advocate — then reason over the evidence: pattern_scout composes citation-gated findings, chain_builder maps them onto the seven-stage Agent Kill Chain, poc_forge synthesizes reproducer scripts that run in network-isolated sandboxes, and devils_advocate runs an inverted-prompt rebuttal pass (optionally on an independent, non-mirroring model). A reflector agent distills the engagement into reusable knowledge triples. Every step is Cedar-policy-gated, hash-chained in an audit journal, and signed with a per-engagement Ed25519 key. codered report then emits SARIF, Markdown, and an Ed25519-signed JSON handoff.

Why it's different

Most code scanners stop at "here is a risky-looking pattern." codered treats a raw finding as a hypothesis and makes it earn promotion to a reported finding:

  • Citation-gated — no finding is stored without a structural witness (analyzer output, code reference, or a reachability hypothesis). Enforced by Cedar, not convention.
  • Reachability, not shapes — a mechanical taint tracer walks real dataflow edges from source to sink. A finding must cite that path.
  • Sandboxed proofpoc_forge writes a reproducer and runs it in a network-isolated container; the result (reproduced / refuted / inconclusive) is recorded.
  • Adversarial review — a read-only devils_advocate tries to refute each finding, optionally on an independent model, and can only suppress one if it cites a structural witness of its own.
  • Auditable — every tool call and policy decision is hash-chained and signed. You can prove the journal wasn't edited after the fact.

Core capabilities

Capability What it does
Multi-language SAST Per-language scanner sidecars (semgrep, bandit, clippy, gosec, eslint, checkov, trivy, progpilot, …)
Dataflow + taint tree-sitter dataflow extraction + mechanical BFS taint tracer (no LLM) producing source→sink chains
LLM reasoning Citation-gated pattern_scout, kill-chain chain_builder, reproducer poc_forge, rebuttal devils_advocate, reflector
Sandboxed PoC LLM-synthesized reproducers run in network_mode: none, read-only /repo, time-boxed containers
Cedar governance Policy gates at every finding/advocate/poc/triple/seed write
Tamper-evident audit SHA-256 hash-chained journal; audit::verify_chain proves no tampering
Signed handoff Ed25519-signed, Cedar-filtered engagement-seed.json for downstream red-team tooling

Get started

License

The codered core — the analysis engine and CLI plus the agents/, policies/, tools/, and scanners/ definitions — is licensed under the Apache License 2.0. Copyright © ThirdKey.

Some additional features, including the multi-tenant client portal, are a separate enterprise offering and are not covered by this license. Contact ThirdKey for enterprise licensing.