auditlisted

# Full audit A thin orchestrator over the five domain audits. It exists so the user reviews **one** ranked list and approves **one** batch of issues, instead of five of each. It adds no criteria of its own. Every judgement lives in the domain skills and `audit-loop.md`; this file only decides what runs where, and how the results merge. ## Run order — the two browser audits cannot run in parallel The split is driven by one hard constraint: **there is one Browser pane per session**, and two audits drive it. - **`legal-audit`, `security-audit` and `claude-file-audit` run as three parallel subagents.** They are independent, read-only sweeps over disjoint file sets — exactly the case parallel subagents are for. Launch all three in a single message. (`security-audit` may itself invoke the built-in `/security-review` and `npm audit`; that stays inside its own subagent.) - **`accessibility-audit` and `ui-audit` run in the main agent, one after the other — never as subagents, and never at the same time as each other.** Both drive the Browser pane and both need the seeded `dev-temp-data` instance, so a subagent (or a concurrent sibling) would contend for the same tabs and preview server. Run them **sequentially**, sharing the one preview session: bring up `dev-temp-data` once, do the accessibility pass, then the UI pass (or vice versa), then tear the preview down once at the end. Start the three subagents first. Then, while they work, do the two browser passes