PandaWard
Local, evidence-based application security assessment CLI and Claude Code plugin that reports undecided controls as plainly as failed ones.
At a Glance
Free and open source CLI and Claude Code plugin under AGPL-3.0-only. Runs locally with no account and no cloud backend. Currently alpha software.
Engagement
Available On
Listed Nov 2026
About PandaWard
PandaWard is an open-source (AGPL-3.0) Go CLI and Claude Code plugin that scans your own repository locally and keeps an evidence ledger across sessions. Its README describes it as alpha software. It is passive by default and reports controls it could not decide instead of presenting them as clean.
What It Is
PandaWard is an application security assessment engine. It inventories a repository read-only (routes, manifests, workflows, IaC, mobile manifests, AI/MCP usage). It maps a control catalog of eighteen domains (A to R) onto what it finds, runs built-in offline checks plus optional scanners, and writes redacted, hash-addressed evidence into a local SQLite ledger. Reports come as HTML, Markdown, JSON, SARIF and CSV.
Honest statuses and the ledger
Only PASS and FAIL are decisive. Other outcomes (SUSPECTED, BLOCKED, UNSUPPORTED, INCONCLUSIVE, ERROR, STALE) are shown as gaps. Findings are fingerprinted across runs, so a fix counts only when the check runs again and no longer sees the problem, and a moved file keeps its findings. Agents can only record hypotheses, and PandaWard reads the cited lines itself. Manual controls such as the threat model or rate limiting can be attested by a person after a reviewer proposes a verdict with cited evidence.
Safety model
The default profile never runs install scripts, builds or hooks and sends no traffic to targets. Network tests need a scope manifest that a person approves from an interactive terminal, bound to the file's hash, with a request budget and no redirects. Secrets are masked before storage, printing or return to a model. There is no account or cloud backend, and telemetry is off unless enabled.
Cross-domain edges and tool adapters
When two domains share an asset, such as login routes and tokens, or webhooks and payments, PandaWard creates an edge and tracks it separately. Optional adapters include gitleaks, semgrep, osv-scanner, trivy, apktool, ZAP and a local MobSF server. SARIF from other analyzers can be imported.
Setup path
Install as a Claude Code plugin from the ratomirjovanovic/pandaward marketplace and run /pandaward:start, or install the CLI via Homebrew, Scoop, a release binary or go install and run pandaward assess. The project states it is not a certification, attestation or signed penetration test, and cannot prove the absence of vulnerabilities.
Community Discussions
Be the first to start a conversation about PandaWard
Share your experience with PandaWard, ask questions, or help others learn from your insights.
Pricing
Open Source (AGPL-3.0)
Free and open source CLI and Claude Code plugin under AGPL-3.0-only. Runs locally with no account and no cloud backend. Currently alpha software.
- Local CLI and Claude Code plugin
- Passive repository scanning
- Local SQLite evidence ledger
- No account and no cloud backend
- Telemetry off unless enabled
Capabilities
Key Features
- Passive-by-default local repository scanning
- Local SQLite evidence ledger persisting across sessions
- Honest statuses (PASS, FAIL, SUSPECTED, BLOCKED, UNSUPPORTED, INCONCLUSIVE, ERROR, STALE)
- Secret redaction before storage, output or model access
- Human-approved scope manifests for network tests
- Cross-domain edges between related controls
- Attestation of manual controls with cited evidence
- Reports in HTML, Markdown, JSON, SARIF and CSV
- Ordered fix plan and retest with delta
- Route and entrypoint discovery across common frameworks
- CI GitHub Action and pre-commit hook
- Built-in rule benchmark corpus
- Opt-in telemetry, off by default
