fix(logging): neutralize control characters in RedactingFormatter (fixes #61409)

[kuangmi-bit]

Summary

RedactingFormatter.format() redacts secrets but never sanitized control characters, allowing log record injection. An attacker-controlled value in a log argument could embed a newline to forge additional log records with attacker-chosen levels and components.

Fix

Add control-character neutralization to RedactingFormatter.format(). After redact_sensitive_text(), replace control characters (\x00-\x1f, DEL, NEL, LINE SEPARATOR, PARAGRAPH SEPARATOR) with ? — matching the existing _log_safe_path pattern in gateway/platforms/base.py.

Verification

• Embedded newline in interpolated message → single physical line, no forged record
• Existing redact test still passes
• New test test_control_characters_neutralized covers the injection scenario

[alt-glitch]

This was generated by AI during triage.

Duplicate of #61411 — same fix: both add the identical control-character neutralization regex to RedactingFormatter.format in agent/redact.py and both close #61409. #61411 was opened earlier (09:51Z vs 12:07Z) and sanitizes the message text before formatting so multi-line tracebacks stay intact, which is the cleaner variant. Consolidating here; #61411 is the canonical fix.

[tonydwb]

Code Review Summary

Verdict: Approved (LGTM)

Overview

Small, well-scoped security fix. Adds control character sanitization to RedactingFormatter to prevent log injection attacks where attacker-controlled data embedded in log arguments could forge additional log records.

Changes

• agent/redact.py: Added _LOG_UNSAFE_CHARS regex pattern covering control chars (0x00-0x1F), DEL, NEL, LINE SEPARATOR, PARAGRAPH SEPARATOR. Applied in format() after redaction.
• tests/agent/test_redact.py: New test test_control_characters_neutralized covering malicious input with embedded newline.

Security

• Log injection prevention — attacker-controlled data cannot forge additional log records
• Matches existing _log_safe_path pattern in gateway/platforms/base.py
• No secrets or credentials in diff

Code Quality

• Clean implementation, minimal diff (~20 lines)
• Good documentation explaining the security rationale
• Test is thorough: checks replacement char, legitimate prefix survives, single physical line result

Testing

• Dedicated unit test covers the specific attack vector

---

Reviewed by Hermes Agent

[tonydwb]

Code Review Summary

Verdict: LGTM

What the PR Does

Neutralize control characters (including newlines) in log output via RedactingFormatter, preventing log-forging attacks where attacker-controlled data in log arguments injects fake log records.

Assessment

• Security: This is a genuine security fix. A malicious actor controlling log argument content could inject newlines to forge ERROR-level log entries. The regex covers control chars, DEL, NEL, LINE SEPARATOR, and PARAGRAPH SEPARATOR.
• Tests: New test test_control_characters_neutralized covers the newline-forging scenario directly.
• Correctness: Matches the existing _log_safe_path pattern in gateway/platforms/base.py.

Note: This is the formal review verdict. PR 61450 previously had a COMMENT activity.

---

Reviewed by Hermes Agent