Skip to content

chimera.security

chimera.security classifies tool-call risk and decides whether to ask the user before executing.

For the tutorial, see Add Security Policies.

Top-level exports

from chimera.security import (
    SecurityRisk,
    RiskClassifier,
    SecurityAnalyzer,
    LLMSecurityAnalyzer,
    RuleBasedSecurityAnalyzer,
    CompositeSecurityAnalyzer,
    ConfirmationPolicy,
    NeverConfirm,
    AlwaysConfirm,
    ConfirmAboveThreshold,
)

Risk levels

`SecurityRisk`	Value	Meaning
`UNKNOWN`	0	Could not classify. Treated as HIGH for safety.
`LOW`	1	Read-only, harmless.
`MEDIUM`	2	Writes, installs, network to known endpoints.
`HIGH`	3	Destructive ops (`rm -rf`, `DROP TABLE`, force push), credential access.

Analyzers

Class	Module	Speed
`RuleBasedSecurityAnalyzer`	`chimera.security.analyzer`	Fast, free; pattern matches against known-dangerous strings.
`LLMSecurityAnalyzer`	`chimera.security.analyzer`	Smarter; sends the call to a `Provider` for classification.
`CompositeSecurityAnalyzer`	`chimera.security.analyzer`	Runs the rule-based check first; falls through to the LLM only when uncertain.

Every analyzer implements analyze(tool_call) -> SecurityRisk.

Confirmation policies

Class	Module	Behaviour
`NeverConfirm()`	`chimera.security.policy`	Never ask, regardless of risk.
`AlwaysConfirm()`	`chimera.security.policy`	Always ask.
`ConfirmAboveThreshold(SecurityRisk.MEDIUM, confirm_unknown=True)`	`chimera.security.policy`	Ask for risks at or above the threshold; `UNKNOWN` triggers confirmation by default.

See also

Add Security Policies for end-to-end wiring.
chimera.permissions for the policy surface that consumes SecurityRisk decisions.
chimera.events for SecurityEvent.