One install gives your agent a bounded release preflight for a site you own: a fixed 19-path set, hard request and byte limits, and verdicts that separate a real .env leak from a single-page app serving its usual shell — or a blanket WAF block from a hardened origin.
install
Detects Claude Code, Cursor, Codex, Qoder and more, then writes the skill into the right folder. Add -l to see what the repo ships before anything is installed.
npx skills add arm-pz/web-preflight
Quality record
Version, date and limits are read from the repo on every sync, and the check runs every night. Real-world runs are listed only with their evidence.
the audit
A preflight that expands into a scan defeats its own purpose. The path list is fixed, the budget is hard, and nothing sensitive is written to disk.
Every call is bounded, and the bound is enforced by the client, not by trusting the server.
Config, repository metadata, and admin or debug surfaces. Never a wordlist, never a query string.
Read first. A rule you can fetch is not a wall around a secret.
Calibrated so a finding means what it says.
Apex and www are assessed separately; a clean canonical host does not vouch for a host that only redirected to it. The overall verdict is the weakest host's.
house rules
If a firewall swallowed the probes, the budget ran out, or the tooling could not enforce the byte cap, the answer is INCONCLUSIVE. "Couldn't assess" is not folded into PASS or PASS WITH FIXES.
CRITICAL needs a confirmed credential or key. A reachable source map is MEDIUM, and a bare Git reference is content-free metadata — not proof that repository contents are exposed. Escalate only when the body says so.
Bounded fingerprints instead of saved bodies, sanitized headers instead of raw ones, cookie values removed before anything is recorded. The report carries the minimum evidence needed to act.
Free, open, and it runs entirely inside the agent you already use.
Install web-preflight Read the source