- Home
- Skills
- Agents & Orchestration
- Rescue Audit
Works with the AI Agents you already use
Rescue Audit
Take over an abandoned project the way a rescue contractor does: scope, stop the bleeding, tests, first 30 days.
$12
Rescue Audit
Example session with this skill installed
I just took over a courier-booking back office (Next.js + SQLite) from a vendor who left. Run /rescue-audit, start with stopping the bleeding.
- Read your context and instructions
- Compiled the rescue audit
(Example output on a fictional project. Full sample in examples/03-bleeding-example.md)
03 Stop the bleeding: first-hour list
| # | Problem | Evidence | Action |
|---|---|---|---|
| 1 | Sessions can be forged and never expire: cookie is base64(SECRET:timestamp), so one leaked cookie reveals the secret | src/lib/auth.ts:4,30-38 | Server-side random sessions with expiry; logout must revoke |
| 2 | Public endpoint returns customer name, phone, address; ids are date + 4 random chars, enumerable | api/public/booking/[id]/route.ts:7-12 | Return QR + status only; unguessable tokens; rate limit |
| 3 | Dev proxy signing real carrier requests runs on production port 3001; its token is printed in the README | proxy/proxy.js:10, README.md:117 | Human check: is 3001 open? If yes, stop it today |
| 4 | Admin over plain HTTP; basic-auth layer removed | README.md:78 | HTTPS; IP allowlist meanwhile |
| 5 | No database backup anywhere | deploy.sh, crontab | Daily off-server backup, then restore once |
Keys, third-party services and servers each get a keep / rotate / transfer / stop call. Secrets are listed by location only.
Connects securely to your tools. The creator never sees your data.
About this skill
You just inherited a project. The vendor left, the spec is thin, and nobody knows which key opens what. Most "codebase audits" read the code and hand you a list of smells. Real rescues fail somewhere else: nobody agreed what "done" means, and something was already bleeding while everyone read code.
Rescue Audit runs the takeover in the order an experienced rescue contractor does, and writes a client-ready document at every step:
- Understand what was promised. Reads the spec, designs, meeting notes and quote, maps the user flows and every decision point, and grades each feature solid / gap / landable against the code.
- Fence the scope. Pulls the out-of-scope list from the quote (or drafts one) into a one-page sign-off sheet a non-technical client can approve. Then the rescue-or-rebuild gate: it extracts what the existing code actually does as flows and checks each against the confirmed requirements. If most flows conflict, over half the code serves no confirmed flow, or the core data model is wrong, it recommends a rebuild (usually: keep the UI, redesign the data model and logic) instead of patching.
- Stop the bleeding, on day one. Inventories keys, third-party services, servers and open doors (forgeable sessions, routes that skip login, dev proxies left on production). Each gets a call (keep / rotate / transfer / stop), plus a ranked first-hour list.
- Agree on "working", then write the tests. Asks the client's operating conditions and turns each must-not-break flow into a Given/When/Then template, then failing tests. The rescue is done when they pass.
- Rescan and retest. Versions as actually installed per environment (catches the "lockfile never deployed" trap), vulnerabilities judged by whether you use the affected feature, upgrades retested one by one.
- Watch the first month. A monitoring plan with a baseline, and a follow-up mode that reads your exported logs and flags anomalies.
Built from real rescues. Field-tested on a production courier-booking back office, where it found a forgeable session, a public endpoint leaking customer data and a production signing proxy left running. Those were issues a previous cleanup had missed.
Safe by design
- Read-only until you approve phase 4. No commits, installs or deploys on its own.
- Secrets are reported by location only, never printed.
- Every finding cites a file, commit or command output.
- Anything for the client is a draft you send yourself.
- No network calls except your package manager's audit command.
Use it when you're taking over a stalled project, onboarding a vendor's handover, or quoting a rescue job and need to know what you're walking into.
Works with any coding agent that can read files and run shell commands: Claude Code, Codex CLI, Gemini CLI, Cursor and others.
Does not fix code or rotate secrets for you. Phase 2 and 4 need your client's input (quote, operating conditions). Phase 6 needs exported logs or analytics. Dependency audit needs the package registry to be reachable.
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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 8 days ago
Permissions
Reads the repository and runs read-only shell commands (git log, grep, package audit). Writes report files to a rescue/ folder only after you agree. Network: only the package manager's audit command.