Author SHA1 Message Date
Claude 34fafaefda Fix auto-detect false positives for configured values inside compound matches
Home path, shell prompt, git remote, and env var assignment patterns match
a compound string (e.g. "/home/johnsmith") that only *contains* a configured
identity/mapping value ("johnsmith"), not equals it. The exact-match
configured.has() check never recognized these as already handled, so the
"potential PII" warning fired on every message containing one of these
shapes regardless of configuration, and the fallback auto-redact step
overwrote already-substituted text with a generic placeholder.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3nGKyoLaRghi1okiNyRKD
2026-09-15 23:15:42 +00:00
2 changed files with 34 additions and 2 deletions
+17 -1
View File
@@ -349,6 +349,22 @@
const CONTEXT_WORDS_RE = /\b(?:born|birthday|dob|birth|passport|license|driver|ssn|social\s*security|address|zip|postal|date\s+of\s+birth)\b/i; const CONTEXT_WORDS_RE = /\b(?:born|birthday|dob|birth|passport|license|driver|ssn|social\s*security|address|zip|postal|date\s+of\s+birth)\b/i;
// Some patterns (home paths, shell prompts, git remotes, env var assignments)
// match a compound string that only *contains* a configured value rather
// than equaling it exactly (e.g. "/home/johnsmith" contains the configured
// username "johnsmith"). A plain Set.has() exact-match check never fires
// for these, so the same configured value gets flagged as "unconfigured"
// every time. Length-gated to avoid short tokens suppressing unrelated
// findings via accidental substring collision.
function matchesConfiguredValue(value, configured) {
const lower = value.toLowerCase();
if (configured.has(lower)) return true;
for (const c of configured) {
if (c.length >= 3 && lower.includes(c)) return true;
}
return false;
}
// Common English words that are capitalized but aren't proper nouns. // Common English words that are capitalized but aren't proper nouns.
// Used by the proper noun heuristic to reduce false positives. // Used by the proper noun heuristic to reduce false positives.
const COMMON_CAPITALIZED = new Set([ const COMMON_CAPITALIZED = new Set([
@@ -487,7 +503,7 @@
let m; let m;
while ((m = pat.re.exec(text)) !== null) { while ((m = pat.re.exec(text)) !== null) {
const val = m[0]; const val = m[0];
if (configured.has(val.toLowerCase())) continue; if (matchesConfiguredValue(val, configured)) continue;
if (pat.skip && pat.skip.test(val)) continue; if (pat.skip && pat.skip.test(val)) continue;
findings.push({ name: pat.name, value: val, hint: pat.hint, category: pat.cat }); findings.push({ name: pat.name, value: val, hint: pat.hint, category: pat.cat });
} }
+17 -1
View File
@@ -129,6 +129,22 @@ const PII_PATTERNS = [
// Context words that make ambiguous patterns more likely to be PII // Context words that make ambiguous patterns more likely to be PII
const CONTEXT_WORDS = /\b(?:born|birthday|dob|birth|passport|license|driver|ssn|social\s*security|address|home|live|lives|reside|zip|postal)\b/i; const CONTEXT_WORDS = /\b(?:born|birthday|dob|birth|passport|license|driver|ssn|social\s*security|address|home|live|lives|reside|zip|postal)\b/i;
// Some patterns (home paths, shell prompts, git remotes, env var assignments)
// match a compound string that only *contains* a configured value rather than
// equaling it exactly (e.g. "/home/johnsmith" contains the configured
// username "johnsmith"). A plain Set.has() exact-match check never fires for
// these, so the same configured value gets flagged as "unconfigured" every
// time. Length-gated to avoid short tokens (e.g. a 2-char substitute)
// suppressing unrelated findings via accidental substring collision.
function matchesConfiguredValue(value, configured) {
const lower = value.toLowerCase();
if (configured.has(lower)) return true;
for (const c of configured) {
if (c.length >= 3 && lower.includes(c)) return true;
}
return false;
}
const AutoDetect = { const AutoDetect = {
/** /**
* Scan text for potential unconfigured PII. * Scan text for potential unconfigured PII.
@@ -178,7 +194,7 @@ const AutoDetect = {
const value = match[0]; const value = match[0];
// Skip if already configured // Skip if already configured
if (configured.has(value.toLowerCase())) continue; if (matchesConfiguredValue(value, configured)) continue;
// Skip excluded values (like 127.0.0.1) // Skip excluded values (like 127.0.0.1)
if (pattern.exclude && pattern.exclude.test(value)) continue; if (pattern.exclude && pattern.exclude.test(value)) continue;