Skip to content

armgit / ruflo

Public Imported from GitHub
0

🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated

fix(aidefence): quick-mode PII detection + wrapper-level stats counters

Reuvenea312c8508aa
diff --git a/v3/@claude-flow/cli/src/mcp-tools/security-tools.ts b/v3/@claude-flow/cli/src/mcp-tools/security-tools.ts
index a3c2c8cd83..b9eba583d6 100644
--- a/v3/@claude-flow/cli/src/mcp-tools/security-tools.ts
+++ b/v3/@claude-flow/cli/src/mcp-tools/security-tools.ts
@@ -27,6 +27,21 @@ let aidefenceInstance: AIDefenceInstance | null = null;
// Track if we've attempted install this session
let installAttempted = false;
+// ADR-093 follow-up: wrapper-level counters for the lightweight aidefence
+// tools (has_pii, is_safe, scan-quick) that bypass the package's own
+// stats tracking. The audit flagged that aidefence_stats stayed at zero
+// regardless of detections from these paths. Stats now combine the
+// underlying defender.getStats() with our wrapper counters so the user
+// sees a true reflection of how often each tool category fired.
+const wrapperStats = {
+ hasPiiCalls: 0,
+ hasPiiHits: 0,
+ isSafeCalls: 0,
+ isSafeUnsafeVerdicts: 0,
+ quickScanCalls: 0,
+ quickScanThreats: 0,
+};
+
/**
* Get or create AIDefence instance (throws if unavailable)
*/
@@ -117,13 +132,23 @@ const aidefenceScanTool: MCPTool = {
if (quick) {
const result = defender.quickScan(input);
+ // Audit-flagged: quickScan focuses on prompt-injection/jailbreak
+ // patterns and missed obvious PII (email + API key). Layer a fast
+ // PII check so quick mode catches both threat classes.
+ let piiPresent = false;
+ try { piiPresent = !!defender.hasPII(input); } catch { /* hasPII unavailable */ }
+ const threatDetected = result.threat || piiPresent;
+ wrapperStats.quickScanCalls++;
+ if (threatDetected) wrapperStats.quickScanThreats++;
return {
content: [{
type: 'text',
text: JSON.stringify({
- safe: !result.threat,
- threatDetected: result.threat,
+ safe: !threatDetected,
+ threatDetected,
confidence: result.confidence,
+ piiDetected: piiPresent,
+ promptInjectionDetected: result.threat,
mode: 'quick',
}, null, 2),
}],
@@ -264,11 +289,24 @@ const aidefenceStatsTool: MCPTool = {
content: [{
type: 'text',
text: JSON.stringify({
+ // Underlying defender stats (from full detect() calls via aidefence_scan/aidefence_analyze)
detectionCount: stats.detectionCount,
avgDetectionTimeMs: stats.avgDetectionTimeMs,
learnedPatterns: stats.learnedPatterns,
mitigationStrategies: stats.mitigationStrategies,
avgMitigationEffectiveness: stats.avgMitigationEffectiveness,
+ // ADR-093 follow-up: wrapper-level stats for the lightweight
+ // tools that bypass the defender's own counters. Audit found
+ // these stayed at 0 even after has_pii/is_safe/scan-quick
+ // calls — these counters surface real activity.
+ wrapper: {
+ hasPiiCalls: wrapperStats.hasPiiCalls,
+ hasPiiHits: wrapperStats.hasPiiHits,
+ isSafeCalls: wrapperStats.isSafeCalls,
+ isSafeUnsafeVerdicts: wrapperStats.isSafeUnsafeVerdicts,
+ quickScanCalls: wrapperStats.quickScanCalls,
+ quickScanThreats: wrapperStats.quickScanThreats,
+ },
}, null, 2),
}],
};
@@ -400,13 +438,26 @@ const aidefenceIsSafeTool: MCPTool = {
const input = args.input as string;
try {
+ // Route through the singleton so wrapper stats track this call,
+ // and so a single defender controls both the prompt-injection model
+ // and the PII detector.
+ const defender = await getAIDefence();
const { isSafe } = await import('@claude-flow/aidefence');
- const safe = isSafe(input);
+ const promptSafe = isSafe(input);
+ let piiPresent = false;
+ try { piiPresent = !!defender.hasPII(input); } catch { /* hasPII unavailable */ }
+ const safe = promptSafe && !piiPresent;
+ wrapperStats.isSafeCalls++;
+ if (!safe) wrapperStats.isSafeUnsafeVerdicts++;
return {
content: [{
type: 'text',
- text: JSON.stringify({ safe }, null, 2),
+ text: JSON.stringify({
+ safe,
+ promptSafe,
+ piiPresent,
+ }, null, 2),
}],
};
} catch (error) {
@@ -444,6 +495,8 @@ const aidefenceHasPIITool: MCPTool = {
try {
const defender = await getAIDefence();
const hasPII = defender.hasPII(input);
+ wrapperStats.hasPiiCalls++;
+ if (hasPII) wrapperStats.hasPiiHits++;
return {
content: [{