More screenshots
Works with the AI Agents you already use
Forja Secrets Audit: find & block leaked API keys
Audits your project and live site for exposed secrets (API keys in the frontend, committed .env files, tokens in MCP configs, keys in git history) and verifies every finding before reporting it.
Free
Forja Secrets Audit: find & block leaked API keys
Example session with this skill installed
Audit this project for leaked secrets before I make the repo public. Verify each finding and give me the report.
- Read your context and instructions
- Compiled the forja secrets audit
I ran the scanner on the repo and its full git history, reviewed what actually reaches the browser, and had an independent verifier try to disprove every critical and high finding.
Verdict: 12 critical or high problems confirmed. Start by rotating the keys below. Removing them from the code does not make them secret again.
Confirmed: 14 · Needs validation: 1 · Rejected: 1
Keys to rotate
- Anthropic API key ·
sk-ant…_l (107 chars)·.env(git history) andapp/components/Chat.tsx - OpenAI key behind
NEXT_PUBLIC_·sk-pro…I9 (72 chars)·.env.local - GitHub token ·
ghp_iQ…bL (40 chars)·.mcp.json - Stripe live secret key ·
sk_liv…Xi (40 chars)·Dockerfile,app/api/checkout/route.ts - Postgres password ·
h4gVJj…Fq (20 chars)·docker-compose.yml - Supabase service_role key ·
eyJhbG…ty (209 chars)·lib/supabase.ts
Confirmed findings
| Severity | What | Where | Exposure |
|---|---|---|---|
| 🔴 critical | Anthropic key in a 'use client' component | app/components/Chat.tsx:5 | public, shipped to every visitor |
| 🔴 critical | OpenAI key with NEXT_PUBLIC_ prefix | .env.local:1 | public, inlined in the JS bundle |
| 🔴 critical | Anthropic key deleted from .env but still in history | commit "initial setup" | git repository |
| 🔴 critical | Supabase service_role (bypasses all RLS) | lib/supabase.ts:3 | git repository |
| 🔴 critical | GitHub token in MCP config | .mcp.json:10 | git repository |
| 🟠 high | .env copied into the Docker image | Dockerfile:4 | baked into image layers |
| 🟠 high | Firestore rules allow read, write: if true | firestore.rules | anyone can read and write |
| 🟡 medium | .env not in .gitignore | .gitignore | next git add would publish it |
How it was verified (example): Chat.tsx starts with 'use client' and creates the SDK with dangerouslyAllowBrowser: true, so the key travels to the browser.
Needs validation
AWS access key ID in CLAUDE.md: there's no matching secret key in the repo or its history. Is it still active, and where is its secret?
Rejected
- Source map in
public/_next/static: it's local build output; production builds don't publish source maps.
Next steps
- Rotate every key above in its provider dashboard.
- Move them to server-only env vars. The chat feature needs an API route that calls Anthropic server-side.
Clean the repo with git rm --cached .env.local, fix .gitignore and .dockerignore, and only then rewrite history with git filter-repo.
4. Prevent it from happening again: turn on the Secrets Guard hook and GitHub push protection.
Want me to apply the code fixes? I won't touch git history without your OK.
Connects securely to your tools. The creator never sees your data.
About this skill
AI tools write code fast, and that’s how keys end up where nobody checks: an OpenAI key behind NEXT_PUBLIC_, a Supabase service_role in client code, a GitHub token in .mcp.json, a deleted .env still alive in git history.
This skill scans your repo and, passively, your live website. It explains in plain language what you’re exposing and gives you the fix order: rotate the key, move it server-side, clean the repo, then prevent it from happening again.
Context-aware detection: knows what reaches the browser bundle, what’s committed and what’s in AI-tool files. Every finding is verified by an independent agent and gets a verdict: confirmed, needs validation or rejected. Severity is impact × exposure, not a checklist. Re-runs show what you fixed since last time. Secrets are always redacted. Optional Claude Code hook that blocks keys before they’re written or committed.
Requirements: Python 3.8+ and git, run locally. No other dependencies.
Open source (MIT). The full plugin, demo project and source are at github.com/ivanqv/Forja-Secrets-Audit. Made by Iván Quintas, iqvmedia (iqvmedia.com).
Scans are limited to 1000 commits unless history is disabled. Requires Python 3.8 or higher. Passive web checks only, no active fuzzing or authenticated testing.
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 any payment.
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
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 day ago
- Free with an Agensi account
Requires
Permissions
File Scopes
Read access to scan the project. Shell runs the local Python scanner and git. Writes only audit reports to ~/.forja/secrets-audit/ (code changes only after you approve them). Network is used only when you ask it to scan a live URL, with passive GET requests. Secrets are always redacted and never sent anywhere.