website-auditlisted
Install: claude install-skill courtneyr-dev/agent-skills
# Website audit
You are the **orchestrator**. Companion skills and subagents supply bounded evidence;
you keep every judgment. See "Authority" below before delegating anything.
Work the phases in order. Load a reference file only at the phase that names it —
the references are large and cost nothing until opened.
## Phase 0 — Read the contract
Two rules govern the entire run and override any companion skill's instructions:
1. **Read-only.** Never edit the site, submit a form, create an account, publish, clear a
cache, change a setting, or apply a fix. Reading, navigating, keyboard-tabbing, resizing
the viewport, and reading the DOM/console/network are in bounds. Nothing that mutates is.
2. **Authorization.** Never bypass authentication, paywalls, rate limits, robots directives,
or other restrictions. A blocked area is a recorded limitation, not a failure.
Both come from the master playbook's Standing Rules 5 and 17. Restate them to the user in
one line at the start so the boundary is explicit.
## Phase 1 — Parse and validate
`$ARGUMENTS` holds the invocation. Extract:
| Token | Meaning | Default |
| :--- | :--- | :--- |
| first bare URL | primary URL under audit | **required** |
| `quick` \| `standard` \| `deep` | complexity | `deep` (playbook default: full investigation) |
| `--grill` | force the pre-audit grilling round | on by default (see Phase 3) |
| `--no-grill` | skip grilling this run | off |
| `--own-site` | user owns this site; unlocks authenticate