CGS-AUTH-001
Authorized scope
Ownership or written authorization is recorded before assessment.
Published standard · version 1.0.0
The open security baseline for AI coding agents.
CapitalGuard Standard turns AI-agent repository risk into twelve testable controls, a deterministic Guardprint, and a public verification protocol.
Normative controls
One baseline. Clear evidence. No hidden pass rules.
Open conformance checks the public controls. CapitalGuard Risk Score calibration remains separate and cannot change whether a required control passed.
CGS-INV-001
Agent inventory
Agents, instructions, tools, commands, permissions, and connected systems are mapped.
CGS-SEC-001
Secret-safe evidence
Secrets are detected without copying values into reports, logs, or public records.
CGS-DATA-001
Data minimization
Local checks do not upload repository source by default.
CGS-INJ-001
Untrusted instructions
Repository text cannot silently override policy or authorize sensitive tools.
CGS-PRIV-001
Least privilege
Agent file, command, connector, and transfer access is limited to the task.
CGS-HUMAN-001
Human approval
High-impact production, data, auth, billing, and deployment actions require approval.
CGS-POL-001
Machine policy
Blocked paths, protected paths, redaction, and approval gates are explicit.
CGS-DRIFT-001
Guardprint drift
A deterministic SHA-256 baseline makes exposure changes visible.
CGS-REPORT-001
Decision-ready report
Findings include impact, precaution, confidence, and scan limits.
CGS-TRACE-001
Evidence traceability
Every result maps to a versioned control and privacy-safe evidence digest.
CGS-ATTEST-001
Signed attestation
Official records use Ed25519, expiry, revocation, and public key verification.
Open conformance
Adopt the standard without surrendering source.
Free checker runs on the repository owner’s runner.
Matched values and source snippets never enter the output.
Results are conformant, non-conformant, or incomplete.
Self-conformance never grants the CapitalGuard Verified badge.
CapitalGuard Verified
Official proof requires a live signature.
A recorded Guardprint proves integrity. An active verified record additionally proves that CapitalGuard completed the authorized review for that exact scope and version.
Ed25519
Signature
SHA-256
Guardprint
Public
Key registry
Live
Revocation
The standard is open. The intelligence is not.
Clear public controls make adoption easier. Protected scoring and data keep the official assessment hard to imitate.
Risk Score weights and calibration
Threat intelligence and detection tuning
Benchmark distributions and peer data
Reviewer playbooks and escalation thresholds
Anti-fraud controls and private signing keys
CapitalGuard Standard 1.0