Claude Code's 'Don't Ask Again' Setting Leaked Secrets in 1,122 Public Repos — Sell AI Coding Agent Config Audits Before a Client Becomes the Next One.

Apiiro's research into Claude Code's "Don't Ask Again" setting is not just a security story. It is a service business sitting in plain sight.

What actually happened

When a Claude Code user picks "Yes, and don't ask again," the tool can save the approved command word-for-word in a project settings file: .claude/settings.local.json.

If that command contained a password, API key, or token, it gets written straight into the file — and if the repo is public, so is the secret.

Apiiro's breakdown of what it found is specific enough to use in a pitch without rounding anything up:

• 379 database passwords
• 43 SSH logins for public servers
• 42 tokens for GitHub, GitLab, and npm
• 23 AWS keys
• 734 additional secrets still sitting in older versions of the file across 509 repositories
• 211 forks carrying the same secrets as their originals
• 63 organizations affected, mostly software companies

Apiiro also compared 16 coding agents and says Claude Code is "the only one confirmed to save the full approved command" by default.

Why this is a business, not just a headline

Every one of those numbers describes a gap most engineering teams do not know they have. They adopted a coding agent for speed. Nobody added "audit the agent's local settings file" to onboarding, offboarding, or their security checklist.

That gap is exactly what a paid service fills:

1. Scan the client's repos and forks for local agent settings files that contain secrets

2. Check older commits and file versions too — Apiiro's numbers show the risk doesn't end at the current file

3. Rotate every exposed credential and scrub the secret from git history

4. Hand the client a one-page policy: which commands are safe to auto-approve, and a flat rule that secrets never get typed into a command an agent might persist

5. Set up a recurring scan so a new leaked settings file gets caught the week it appears, not the year a researcher finds it

The pitch that sells itself

Open with the numbers, not a generic security warning: "Apiiro found secrets in 1,122 public repos through one coding agent setting. We'll scan your org for the same exposure — free, for one repo — before you decide if you want the full audit."

That is concrete, it is free to try, and it is scoped to something the client can verify in minutes.

Bottom line

This story works as a playbook because the fix is cheap to deliver and the risk is expensive to ignore. A few hundred dollars of scanning time against database passwords, AWS keys, and SSH logins sitting in public repos is an easy yes for almost any engineering lead.

Source: https://thehackernews.com/2026/11/claude-codes-dont-ask-again-file.html