Panda Red v0.4.0
Panda Red maps authorized security surfaces, runs a requested repository scan
workflow using available Panda Security analysis, validates invariants in an
approved lab and prepares bug-bounty cases. Active work always needs an
engagement that names exact targets, techniques and budgets.
Panda Red and Panda Blue were split from Panda Recon in
v0.4.0. Red keeps the offensive and assessment modes; Blue keeps hunting, incident
reconstruction, fix verification and threat intelligence. Case, asset, evidence and
finding records use the shared panda-case.*/v1 schemas, so a case can move
between them.
Quickstart · Command reference · Installation · Workflows · Troubleshooting
| Agent mode | Purpose | Example |
|---|---|---|
| scan | Code/configuration security concerns, with manual validation | Source scan |
| recon | Authorized assets, interfaces and trust boundaries | Surface map |
| simulate | Approved synthetic invariant tests | Local lab |
| bounty | Policy-bound validation/submission preparation | Program policy |
Standalone Codex prompt: $panda-red scan ./project --depth quick --no-tools.
Codex plugin: $panda-suite:panda-red scan ./project --depth quick --no-tools.
Claude Code plugin: /panda-suite:panda-red scan ./project --depth quick --no-tools.
These are agent request examples; interactive model runs of these prompts remain
Not Run. Natural language works too: “Use Panda Red to scan this local
project, quick depth, static analysis only; no scanner execution, network or fixes.”
For deterministic help from a source checkout:
.venv/bin/python skills/panda-red/scripts/red.py help
.venv/bin/python skills/panda-red/scripts/red.py help scan
Help does not perform analysis. Default scan behavior is quick/offline; missing
scanners are skipped visibly. The optional scan-report helper imports local
results and validates source locations; it never executes a scanner or a model.
Panda Red makes no network lookups; for current CVE, KEV or EPSS status use Panda
Blue intel and pass its dated output in.
Source/configuration auditing reuses panda-security when available. Write can
compose substantial reports from structured facts. Red remains usable alone.
Active testing, remote scans/uploads, installations, production changes and
external submissions require authorization covering those actions. A manifest
grants no authority.