Redact secrets from agent session logs

How to do it

Logged commands carry what the agent typed, and agents paste keys into commands, such as curl -H "Authorization: Bearer …". Log only what the activity log needs. Keep prompts, tool output and raw API bodies off unless you are debugging. In Claude Code these are off by default. Run every export through a collector that masks known key shapes, such as the OpenTelemetry Collector's redaction processor. Mask by value patterns, such as sk_live_, ghp_ and AKIA, and by key names that contain token or secret. Then scan the stored logs with a secret scanner each day, and rotate any key it finds. The same idea for app logs is Keep personal data and secrets out of logs.

Examples

YAML example

processors:
  redaction:
    allow_all_keys: true
    blocked_key_patterns: [".*token.*", ".*secret.*", ".*password.*"]
    blocked_values:
      - "sk_live_[A-Za-z0-9]+"
      - "gh[pousr]_[A-Za-z0-9]{36,}"
      - "AKIA[0-9A-Z]{16}"

Python example

import re

SECRET_SHAPES = re.compile(r"sk_live_[A-Za-z0-9]+|gh[pousr]_[A-Za-z0-9]{36,}|AKIA[0-9A-Z]{16}")

def mask_secrets(line: str) -> str:
    return SECRET_SHAPES.sub("****", line)

JavaScript example

const SECRET_SHAPES = /sk_live_[A-Za-z0-9]+|gh[pousr]_[A-Za-z0-9]{36,}|AKIA[0-9A-Z]{16}/g;

const maskSecrets = line => line.replace(SECRET_SHAPES, '****');

How to check

Start a session that runs echo sk_live_TESTONLY000000000000. Within five minutes the log store shows the command with the key masked. A daily secret scan of the log store finds no key, or opens a ticket naming it.

Sources

• Claude Code monitoring
• OpenTelemetry Collector redaction processor
• OWASP Logging Cheat Sheet

Related rules

• AID-34 Log what every agent session does in AI-assisted development
• OPS-9 Keep personal data out of logs in Operations and observability