CASE 04 Claude Code, agents, plugins, hooks, Node 2026 48 skills, 19 agents, 11 hooks

Nothing ships until review has run, and the record is a file a gate checks.

5 SHOTS verdicts, stamp, gate refusal, validator, estate map

1 / 5
Review tier, 2026-09-15 12:47, anthony-edward.com, verdict PASS
itemas recorded
anthony-web:html-page-validatorPASS, 1 warning
anthony-web:accessibility-auditorPASS, predicted 97-100 (was 93 live)
anthony-web:performance-auditorPASS, 2 warnings, predicted desktop ~93-95 / mobile 72-82 (was 87 / 61)
anthony-web:security-auditorPASS after fix, 0 critical
anthony-web:js-behavior-auditorPASS, 2 warnings WAIVED; focused re-check of setBackgroundInert PASS
anthony-web:visual-qaPASS
anthony-web:voice-auditor (routed, Grammar A)PASS 0/0/0
impeccable detect.mjs0 findings (34 before waivers)
fig. 1 Review-tier verdicts for my acting and teaching site, typeset from the receipt dated 2026-09-15 12:47. Paths, session fields, notes and commit hashes removed; auditor names and verdict strings as written.

I build with Claude Code: agents I configure, packaged as plugins, with hooks on the tools and a review tier that a web page must pass, or carry a logged waiver, before it is committed. The map of that setup lists 48 skills, 19 agents and 11 hooks.

I keep the rule in a file a program reads, not in my memory: nothing ships until review has run. The review tier runs the page past its auditors, and a stamp records the pass against the file’s exact bytes. A commit gate refuses staged HTML with no matching stamp, and one byte edited after the stamp brings the refusal back.

This site’s own build commits carry a logged waiver, and the full tier runs before it deploys. The estate validator reports 2 red and 60 amber findings today. The figure shows 9 of the 48 skills and 15 of the 19 agents, the ones joined to a project this page can show. No hook is, so there is no hook column.

Review to commit, my site, 2026-09-15