More screenshots
Works with the AI Agents you already use
Railway Post-Breach Check
This skill turns that advice into a checklist your agent runs with you on any Railway stack.
$12
Railway Post-Breach Check
Example session with this skill installed
Run the Railway post-breach check on my project acme-shop (production only). Start with the read-only part and don't change anything yet.
- Read your context and instructions
- Compiled the railway post-breach check
(Excerpt from a longer session. Fictional demo data.)
How I'll work: I won't print any part of a secret value, only names, kinds, lengths and fingerprints. You create new keys and paste them into the Railway dashboard yourself. Nothing changes until you approve that one step, and each step comes with its undo. I can't tell you whether your keys were taken; Railway says it can't tell either. I can help you rotate them and record what was checked.
Scope: acme-shop / production (api, worker, Postgres). Secrets stored before Oct 22, 03:32 UTC: yes. Rollback targets recorded for api and worker.
INVENTORY (names only)
| scope | variable | secret? | kind | issuer (guess) | len | fp |
|---|---|---|---|---|---|---|
| production/api | OPENAI_API_KEY | YES | api-key | openai [AI] | 56 | ab18c28a |
| production/worker | OPENAI_API_KEY | YES | api-key | openai [AI] | 56 | ab18c28a |
| production/api | STRIPE_SECRET_KEY | YES | api-key | stripe (live) | 107 | c62c7281 |
| production/worker | DATABASE_URL | YES | db-url with password (PUBLIC proxy) | postgres | 94 | - |
| production/api | APP_ENCRYPTION_KEY | YES | opaque-token | your app (ENCRYPTION key: data!) (by name) | 44 | 6c21b58d |
EXPOSURE
- HIGH: production/Postgres has a public TCP proxy on 5432.
- worker reaches the database through that public URL. Move it to the private URL before the proxy is removed.
DEPLOYMENTS IN THE WINDOW (Oct 20 13:58 to Oct 22 03:32 UTC)
- api: 0 in window (12 before, 3 after). worker: 1 in window (Oct 20 18:47 UTC, later replaced by a newer deploy). Is that one yours?
PLAN (nothing has changed yet)
- P1 OpenAI key (2 services, same value): you create a new key, keep the old one live, paste it as a new sealed shared variable. I switch worker, redeploy, verify; then api. You revoke the old key last, once OpenAI's key list shows the new key in use.
- P1 Postgres: move worker to the private URL, then you click Regenerate on the database (it restarts and drops open connections) and I redeploy api and worker. Then remove the public proxy. External tools on that proxy will lose access; tell me which ones use it.
- P1 Stripe live key: roll it in Stripe with the old key kept valid for a while.
- APP_ENCRYPTION_KEY: not rotating here. Data encrypted with it would become unreadable; it needs its own re-encryption plan.
Next step: step 1, worker only. Shall I go ahead?
Connects securely to your tools. The creator never sees your data.
About this skill
Railway reported unauthorised access to its systems between Oct 20 and Oct 22, 2026, and asked users to rotate sensitive credentials stored in Railway variables during the incident, remove public network access to databases, and review GitHub and container registry audit logs. This skill turns that advice into a checklist your agent runs with you on any Railway stack. Anything stored on Railway before the attack window ended (Oct 22, 03:32 UTC) is treated as in scope.
What your agent does
- Inventory (read-only). Every project, environment, service and variable, by name. Each variable gets a secret yes/no, what it looks like (API key, database URL, signing secret…), a guess at the issuer, its length, and an optional fingerprint that spots the same secret in several places. AI provider keys and database URLs on a public proxy are flagged. It also lists secrets that aren't variables: registry credentials, webhook headers, bucket keys, Railway tokens in CI.
- Crypto wallets first. A seed phrase or private key can't be rotated, so it goes first. The script spots common shapes (seed phrases, also in lists or JSON; EVM, Solana and Bitcoin keys; keystore files), but a key under an ordinary name can slip past, so the agent also asks you. You move the funds to a new wallet yourself, after checking for a sweeper bot; the agent never touches the seed, key or funds.
- Rotation plan. One row per secret, in priority order, with where to create the new key, which services use it, how to verify, when to revoke the old key, and the undo.
- Guided rotation. New key at the issuer first, then Railway (sealed shared variables preferred), redeploy one service, verify, next service, then revoke the old key. Covers database passwords (Regenerate on Railway, the provider's reset elsewhere), issuers that allow only one key, signing secrets, and a hard stop for data-encryption keys.
- Database exposure. Finds public TCP proxies on databases, lists services still using a public URL, and walks you through removing the proxy, warning you what loses access.
- Access review (read-only). Service logs for the incident window (you read them in the dashboard), Railway members, tokens, audit log (retention varies by plan) and deployments in the window; GitHub security and audit logs, installed apps, deploy keys, webhooks and commits in the window; container registry logs; issuer-side usage.
- Final report. What was rotated, what's pending, what was checked and what couldn't be. It never says you're "safe".
Safety contract (the agent states it before running anything)
- Never prints any part of a secret value, not even a prefix. Listings go through the bundled offline script, which outputs names, kinds, lengths and fingerprints.
- Never carries a new secret through a tool call. You create it at the issuer and paste it into the Railway dashboard; the agent writes only
${{shared.NAME}}references. - Every change is shown with what it changes, what may break and how to undo it, and waits for your approval. One change at a time; approval never carries over.
- Old keys are revoked last, after every service is verified on the new one, so most steps can be undone; the ones that can't are flagged and approved separately.
- Free text (commit messages, audit-log payloads, webhook URLs, start commands) is never printed. Value lengths are shown.
What's in the pack
| File | Purpose |
|---|---|
SKILL.md | Safety contract, scope, phase overview, pre-flight and inventory |
phases/ | Rotation plan, rotation, database exposure, access review, final report |
gotchas.md | 29 known traps, such as staged variables that resolve empty |
issuers.md | Per-issuer notes: can old and new keys overlap |
sources.md | The incident sources the skill relies on, with dates |
report-template.md | The value-free report your agent fills in |
scripts/ | Offline scripts, plus a leak self-test |
What's tested
The scripts are tested on fake data, including a self-test that plants random fake secrets (also where a name, ID or author belongs) and checks none leaks into the output. The read-only steps were run on real Railway projects through Railway's MCP connector, reading no values. The steps that change things follow Railway's docs but haven't been run end to end. The CLI paths, gh commands, Windows and sealed shared variables are untested too; the README lists everything.
Who it's for
Developers and small teams on Railway who got the "rotate your keys" message and want to do it carefully, one service at a time, with a record of what was done. Most keys rotate with no outage; single-key issuers and database password resets need a short planned one, and the skill says so first.
By Mausoft Skills. Railway is a trademark of its owner; this skill is not affiliated with or endorsed by Railway.
Version 1.0 by Mausoft Skills.
Pack contents: SKILL.md (safety contract, scope, phase overview, pre-flight and inventory), phases/ (salt setup, inventory, rotation plan, rotation, database exposure, access review, final report), gotchas.md (29 known traps), issuers.md, sources.md, report-template.md, scripts/ (inventory.py, rotation_plan.py, exposure_check.py, window_check.py, namescreen.py, selftest.py), README.md.
No packages to install. No API keys of ours, no telemetry, no hosted service.
Install: this zip has no top folder (SKILL.md is at its root), so create the skill folder and unzip into it: mkdir -p ~/.claude/skills/railway-post-breach-check && unzip railway-post-breach-check.zip -d ~/.claude/skills/railway-post-breach-check (Cursor: ~/.cursor/skills/railway-post-breach-check; other agents: their skills folder, often .agents/skills/railway-post-breach-check). If you unzip by double-click, make sure the folder holding SKILL.md is named exactly railway-post-breach-check. The README also covers a zip with one top folder.
Incident facts are limited to Railway's preliminary report (with its Oct 26 update), Railway's changelog, Railway's status page updates (Oct 21-27), Railway's "Recover Failing Deploys", "Redeploy Buckets and Reconnect GitHub", "Check for Anomalous GitHub Access" and per-database rotation guides, its deployment-actions docs, TechCrunch and BleepingComputer, and, for the reported wallet thefts, crypto.news and The Hacker News, as of Oct 27, 2026. Railway's final report, which it says will confirm whether source code was accessed, isn't out yet; check its incident post for later updates.
Railway is a trademark of its owner. This skill is not affiliated with or endorsed by Railway.
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 21 days ago
Requires
Permissions
File Scopes
Security disclosure: this skill handles secrets by design, so it is built never to expose them. Variable listings pass through scripts/inventory.py, which reads stdin only, imports no network modules, opens and writes no files, and prints only the variable name, a kind, an issuer guess, the length and an optional HMAC fingerprint keyed with a random per-check salt (kept in a private temp file, never shown, deleted at the end). No part of any value is printed, not even a prefix. Database URLs and short or low-variety values get no fingerprint. Its names-only mode accepts only JSON names and refuses pasted text. The other scripts print names, ports, IDs, statuses, times and counts, and the deployment check has a "triggered by" column that shows - for CLI input (the CLI's JSON has no actor; who triggered a deploy comes from the dashboard, or from Railway MCP output read by eye); they never print hosts of public proxies, commit messages or other free text. Values are never printed. Text in name, author and label fields is screened, best effort: a token that turns up there is masked as <looks like a value, len=N>, but a short or word-like secret (hunter12, a passphrase) typed where a name goes may still print. The length column does reveal how long each value is. The agent never carries a new secret value through a tool call: you paste values in the Railway dashboard and the agent writes only ${{shared.NAME}} references. The skill tells the agent to avoid tools that return secret values or full webhook URLs. Every change requires your explicit confirmation, one step at a time, with the undo shown first. Old keys are revoked last. Security scan notes: automated scanners may flag a few things here, and they are expected. The scripts read environment variable data on stdin in order to classify it. Strings such as sk_live_, AKIA, ghp_ and whsec_ in scripts/namescreen.py and scripts/inventory.py are detection patterns, not keys. scripts/namescreen.py also bundles the BIP39 English wordlist (MIT licence, notice included) to recognise wallet seed phrases; it holds no seed. scripts/selftest.py generates random fake secrets at run time to check that nothing leaks; no key-shaped literal ships in the pack.