Files
gnoma/internal/security/firewall.go
T
vikingowl 49d80cf847 feat(security): format-aware entropy safelist (Phase F-1)
Add a deterministic pre-extractor that skips known-safe token shapes
before they reach the entropy scorer. Targets the false-positive
regime that bites under lowered entropy_threshold or
redact_high_entropy = true — UUIDs (~3.4 bits), SHA hex digests
(~3.9 bits), ISO-8601 timestamps, and HTTP(S) URLs.

Config knob lives under the existing security section to match
entropy_threshold / redact_high_entropy convention:

  [security]
  entropy_safelist = ["uuid", "sha_hex", "iso8601", "url"]

Empty / unset preserves pre-F-1 behaviour exactly — users opt in.

Per-pattern Debug telemetry fires on every skip (pattern name +
token length, never the token bytes). This is the data F-2's
go/no-go gate depends on; the plan literally specifies it.

NewFirewall validates names at the config boundary and emits a
Warn for unknown entries so a typo like "uid" instead of "uuid"
surfaces loudly instead of silently disabling FP reduction.

Tests cover: UUID/SHA-1/SHA-256 skipped at lowered threshold,
mixed payload (safe shape + real secret) preserves the secret,
secret-adjacent-to-UUID regression guard, empty safelist preserves
pre-F-1 behaviour, unknown name silently dropped at scanner level
but warned at firewall level, end-to-end FirewallConfig wiring,
and the skip-telemetry log line.

F-2 remains gated on real-workload FP-rate observations.
2026-05-22 12:39:10 +02:00

161 lines
4.1 KiB
Go

package security
import (
"encoding/json"
"log/slog"
"somegit.dev/Owlibou/gnoma/internal/message"
)
// Firewall scans outgoing LLM requests and incoming tool results
// for secrets, sensitive data, and dangerous Unicode. Core security
// layer — not a plugin, everyone benefits by default.
type Firewall struct {
scanner *Scanner
incognito *IncognitoMode
logger *slog.Logger
// Config
scanOutgoing bool
scanToolResults bool
}
type FirewallConfig struct {
ScanOutgoing bool
ScanToolResults bool
RedactHighEntropy bool
EntropyThreshold float64
EntropySafelist []string
Logger *slog.Logger
}
func NewFirewall(cfg FirewallConfig) *Firewall {
logger := cfg.Logger
if logger == nil {
logger = slog.Default()
}
scanner := NewScanner(cfg.EntropyThreshold, cfg.RedactHighEntropy)
scanner.SetLogger(logger)
// Validate safelist names at the config boundary so a typo surfaces
// loudly instead of silently disabling FP reduction.
entries, unknown := splitSafelistNames(cfg.EntropySafelist)
for _, name := range unknown {
logger.Warn("ignoring unknown entropy safelist name",
"name", name,
"hint", "valid names: uuid, sha_hex, iso8601, url",
)
}
scanner.safelist = entries
return &Firewall{
scanner: scanner,
incognito: NewIncognitoMode(),
logger: logger,
scanOutgoing: cfg.ScanOutgoing,
scanToolResults: cfg.ScanToolResults,
}
}
// Incognito returns the incognito mode controller.
func (f *Firewall) Incognito() *IncognitoMode {
return f.incognito
}
// Scanner returns the secret scanner for adding custom patterns.
func (f *Firewall) Scanner() *Scanner {
return f.scanner
}
// ScanOutgoingMessages scans all message content before sending to provider.
// Returns cleaned messages with secrets redacted.
func (f *Firewall) ScanOutgoingMessages(msgs []message.Message) []message.Message {
if !f.scanOutgoing {
return msgs
}
cleaned := make([]message.Message, len(msgs))
for i, m := range msgs {
cleaned[i] = f.scanMessage(m)
}
return cleaned
}
// ScanToolResult scans a tool execution result for secrets.
// Returns the cleaned content.
func (f *Firewall) ScanToolResult(content string) string {
if !f.scanToolResults {
return content
}
return f.scanAndRedact(content, "tool_result")
}
// ScanSystemPrompt scans the system prompt for accidentally embedded secrets.
func (f *Firewall) ScanSystemPrompt(prompt string) string {
return f.scanAndRedact(prompt, "system_prompt")
}
func (f *Firewall) scanMessage(m message.Message) message.Message {
cleaned := message.Message{Role: m.Role}
cleaned.Content = make([]message.Content, len(m.Content))
for i, c := range m.Content {
switch c.Type {
case message.ContentText:
cleaned.Content[i] = message.NewTextContent(
f.scanAndRedact(c.Text, "message_text"),
)
case message.ContentToolResult:
if c.ToolResult != nil {
tr := *c.ToolResult
tr.Content = f.scanAndRedact(tr.Content, "tool_result")
cleaned.Content[i] = message.NewToolResultContent(tr)
} else {
cleaned.Content[i] = c
}
case message.ContentToolCall:
// Scan LLM-generated tool arguments for accidentally embedded secrets
if c.ToolCall != nil {
tc := *c.ToolCall
scanned := f.scanAndRedact(string(tc.Arguments), "tool_call_args")
tc.Arguments = json.RawMessage(scanned)
cleaned.Content[i] = message.NewToolCallContent(tc)
} else {
cleaned.Content[i] = c
}
default:
// Thinking blocks — pass through
cleaned.Content[i] = c
}
}
return cleaned
}
func (f *Firewall) scanAndRedact(content, source string) string {
// Unicode sanitization first
content = SanitizeUnicode(content)
// Secret scanning
matches := f.scanner.Scan(content)
if len(matches) == 0 {
return content
}
for _, m := range matches {
switch m.Action {
case ActionBlock:
f.logger.Error("blocked: secret detected",
"pattern", m.Pattern,
"source", source,
)
return "[BLOCKED: content contained a secret]"
default:
f.logger.Debug("secret redacted",
"pattern", m.Pattern,
"action", m.Action,
"source", source,
)
}
}
return Redact(content, matches)
}