Works with the AI Agents you already use
Secret Leak Guard: Stop Claude Code Printing Your API Keys
A secret does not leak because an agent is careless.
$5
Secret Leak Guard: Stop Claude Code Printing Your API Keys
Example session with this skill installed
Install the Secret Leak Guard hook for this machine, then check whether bash -x deploy.sh and grep -l KEY .env would be allowed.
- Read your context and instructions
- Compiled the secret leak guard
selftest: 94 of 94 passed. Hook added to ~/.claude/settings.json (matcher Bash|PowerShell|Read|Grep). bash -x deploy.sh -> DENY: shell tracing (sh -x / -v prints every line, sourced secrets included). grep -l KEY .env -> ALLOW.
Connects securely to your tools. The creator never sees your data.
About this skill
A secret does not leak because an agent is careless. It leaks because the agent is doing its job: debugging. It runs bash -x deploy.sh to see where the script fails, the script sources .env, and the trace prints every assignment. Four live keys in one command, now sitting in a transcript that is stored, synced and sometimes shared.
A rule in CLAUDE.md saying "never trace scripts that load secrets" does not stop this. The agent is not reading the rule at the moment it decides to trace. A rule that lives in front of the tool cannot be skipped.
This skill installs that rule as a Claude Code PreToolUse hook.
An agent was asked why a scheduled job failed silently. It ran the job with bash -x. The job sourced an env file holding a payment provider's live secret key, a telephony auth token, an email API key and a chat bot token. All four were printed into the session in plain text, and all four had to be rotated the same day. The note warning against exactly this already existed. After the hook went in, the same mistake was refused before it ran, with a reason the agent could act on.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.
A pattern guard, not a sandbox: a determined process can still read a file. It stops the common honest accident, an agent printing a secret while debugging. It cannot undo a leak that already happened (rotate the key). On real work it refused 6 of 7,454 safe commands (0.08%); each refusal names the safe alternative.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 9 days ago
- Software
Python 3 (standard library only)Claude Code (PreToolUse hook)
Git
Permissions
File Scopes
No network calls. Reads the hook JSON on stdin and prints a deny decision. Writes only one log line when a human-approved override is used.