Files
skillshare/internal/audit/audit.go
T
Rusty Shackleford 65af9c17b7 feat(audit): detect narrow prose risk directives (#711)
* feat(audit): detect narrow prose risk directives

* fix(audit): honor link confidence and narrow refusal rule

Apply configured confidence to Markdown link findings and require safety-related evidence in anti-refusal directives to avoid ordinary compliance false positives.

* fix(audit): address prose and confidence review feedback

Narrow prose matches, preserve overrides when enabling rules, and expose effective confidence to rule consumers. Expand Markdown coverage and synchronize the audit guidance.

* fix(audit): retain scheduled payload detection

Keep detecting directives that schedule a payload while excluding ordinary payload processor terminology.

* fix(audit): preserve prose detection recall

Recognize concise disclosure and safety override directives while retaining benign exclusions. Share the confidence default and cover replacement overlays.

* fix(audit): detect quantified refusal directives

Retain detection when a directive refers to any or every user request while preserving benign security discussion.
2026-11-02 16:31:56 +08:00

86 lines
3.4 KiB
Go

package audit
import (
"errors"
"strings"
)
// ErrBlocked is a sentinel error indicating that an operation was blocked
// by the security audit. Use errors.Is(err, audit.ErrBlocked) to check.
var ErrBlocked = errors.New("blocked by security audit")
const (
maxScanFileSize = 1_000_000 // 1MB
maxScanDepth = 6
analyzabilityThreshold = 0.70
)
// Analyzer IDs — each maps to one analysis source.
const (
AnalyzerStatic = "static"
AnalyzerDataflow = "dataflow"
AnalyzerTier = "tier"
AnalyzerIntegrity = "integrity"
AnalyzerStructure = "structure"
AnalyzerCrossSkill = "cross-skill"
AnalyzerMetadata = "metadata"
)
// Category classifies what a finding is about.
const (
CategoryInjection = "injection"
CategoryExfiltration = "exfiltration"
CategoryCredential = "credential"
CategoryObfuscation = "obfuscation"
CategoryPrivilege = "privilege"
CategoryIntegrity = "integrity"
CategoryStructure = "structure"
CategoryRisk = "risk"
CategoryTrust = "trust"
)
// categoryForPattern maps a rule pattern name to a broad category.
func categoryForPattern(pattern string) string {
switch {
case strings.Contains(pattern, "injection") || strings.Contains(pattern, "hidden-comment") || pattern == "prompt-leakage" || pattern == "anti-refusal":
return CategoryInjection
case strings.Contains(pattern, "exfiltration") || strings.Contains(pattern, "fetch-with-pipe") || pattern == "agent-snooping":
return CategoryExfiltration
case strings.Contains(pattern, "credential") || strings.Contains(pattern, "hardcoded-secret"):
return CategoryCredential
case strings.Contains(pattern, "obfuscation") || strings.Contains(pattern, "escape-obfuscation") ||
strings.Contains(pattern, "invisible-payload") || strings.Contains(pattern, "hidden-unicode") ||
strings.Contains(pattern, "data-uri"):
return CategoryObfuscation
case strings.Contains(pattern, "destructive") || strings.Contains(pattern, "shell-execution") ||
strings.Contains(pattern, "dynamic-code-exec") || strings.Contains(pattern, "shell-chain"):
return CategoryPrivilege
case strings.Contains(pattern, "content-"):
return CategoryIntegrity
case strings.Contains(pattern, "dangling-link"):
return CategoryStructure
case strings.Contains(pattern, "tier-") || strings.Contains(pattern, "low-analyzability"):
return CategoryRisk
default:
return CategoryRisk
}
}
// Finding represents a single security issue detected in a skill.
type Finding struct {
Severity string `json:"severity"` // "CRITICAL", "HIGH", "MEDIUM", "LOW", "INFO"
Pattern string `json:"pattern"` // rule name (e.g. "prompt-injection")
Message string `json:"message"`
File string `json:"file"`
Line int `json:"line"`
Snippet string `json:"snippet"` // trimmed matched line (no truncation)
// Phase 2 fields — analyzer traceability and deduplication.
RuleID string `json:"ruleId,omitempty"` // stable rule identifier
Analyzer string `json:"analyzer,omitempty"` // static|dataflow|tier|integrity|structure|cross-skill
Category string `json:"category,omitempty"` // injection|exfiltration|credential|obfuscation|...
Confidence float64 `json:"confidence,omitempty"` // 0~1
Fingerprint string `json:"fingerprint,omitempty"` // sha256 stable hash for deduplication
Acknowledged bool `json:"acknowledged,omitempty"` // previously accepted via --force; excluded from blocking
}