SEIS

Key-parity report for the 5-locale translations.json: per-locale key counts, keys missing from any locale, referenced-but-undefined keys, and empty values. Call this before and after any translation edit.

Key-parity report for the COPY and PROOF dictionaries inside script.js — the i18n layer index.html actually renders from via data-copy-key. Reports per-locale key counts, keys missing from a locale, empty values, and data-copy-key attributes with no dictionary entry. Distinct from i18n_status, which…

Look up one translation key and return its value in every locale (tr/en/fr/it/de). Null marks a locale where the key is missing.

Search translation keys and values (case-insensitive substring) across all locales. Use to find the right key before editing copy.

Add a new translation key with a value for every locale. Writes translations.json. Fails if the key exists unless overwrite is true.

Rename a translation key across ALL locales and rewrite every reference: data-i18n* attributes in index.html and getT() calls in script.js. Fails if the old key is missing anywhere or the new key already exists. Run web_contract_check + i18n_status afterwards.

Audit index.html for SEO/PWA requirements: title, meta description, canonical, Open Graph, Twitter card, hreflang alternates, JSON-LD, manifest, robots.txt, sitemap.xml.

Verify every #id/.class selector referenced from script.js exists in index.html (or is created by the script). Catches markup rewrites that silently break site behaviour — run after ANY edit to index.html or script.js.

List drawing images under public/media/drawings with sizes, and cross-check that every image referenced from index.html exists on disk.

Static CSS audit of style.css: FAILS when a var(--x) custom property is used but never defined (in CSS, inline HTML style, or JS setProperty). Also reports statically-unused CSS classes and unstyled HTML classes as informational.

Read site-config.json (contact endpoint + contact email used by the contact form).

List translation keys that exist in translations.json but are never referenced by a data-i18n attribute in index.html or a getT() call in script.js. Useful for cleaning up dead translation copy.

Summary of the monorepo workspace: lists all packages under packages/ with their name, version, and whether a package.json exists. Also reports the apps/ directory contents.

Read the single-public-plugin SEIS-Agent suite and canonical integration manifest for the embedded SEIS source-module inventory, non-destructive legacy personal aliases, embedded lanes, helper plugin universe, quality gates, and runtime/app integration surfaces.

Read the one-public-SEIS-Agent distribution and lifecycle status for unified-suite availability, embedded module discovery, canonical alias resolution, SEIS AI connection, install/MCP smoke gates, release channel, support tiers, and approval boundaries.

Read the SEIS AI Core provider registry for zero-key Local Demo, supervised Codex, optional server-only cloud providers, local-provider candidates, public provider states, and security invariants. Read-only; performs no live provider calls, credential validation, network checks, SSH, deployment, or …

Read the SEIS AI Core model scaling hardware profile, parameter ladder, and model-scaling sub-agent council for the planned 20B target on 16GB+ RAM, compatibility profiles, benchmark manifest contract, memory budget contract, quantization lanes, local runtime candidates, future 70B ladder, 150B fron…

Read the SEIS AI Core version registry for SEIS AI Core v0.1, SEIS Language v0.1, model-router, prompt-engine, agent-runtime, sub-agent lane bindings, truth boundaries, and five-year promotion gates. Read-only; never claims trained model ownership or live autonomous execution.

Dry-run a SEIS AI Core version promotion gate against repository-local evidence for the embedded SEIS plugin lanes. Read-only; never approves a release, mutates files, calls providers, accesses credentials, deploys, or runs autonomous execution.

Read the SEIS AI Core bounded sub-agent operating model, five-year plan linkage, permission levels, lane quality gates, cadence, and current approval boundaries. Read-only; does not execute sub-agents or claim autonomous runtime.

Evaluate one SEIS AI Core sub-agent dry-run fixture task against role, permission, approval, cancellation, tool, and path-scope rules. Read-only; returns a decision and never mutates files, queues, GitHub, cloud, SSH, providers, or credentials.

Read the SEIS AI Core five-year sub-agent quarterly review ledger. Read-only; shows current/planned quarters, evidence, approval boundaries, and next safe actions without mutating queues, files, GitHub, cloud, SSH, providers, or credentials.

Static accessibility audit of index.html: FAILS when an <img> is missing alt=, an interactive input/select/textarea has no associated label, or a <button> has no accessible name (text, aria-label, data-i18n, etc.). Advisory: positive tabindex values.

Static performance budget for apps/web: file sizes (HTML ≤ 100 KB, CSS ≤ 100 KB, JS ≤ 150 KB, total ≤ 300 KB), render-blocking <script src> in <head>, and images missing loading=lazy or width/height. Budget violations and blocking scripts FAIL the check; missing lazy/dimensions are advisory.

Static security audit of index.html: FAILS on target="_blank" without rel="noopener noreferrer" (tab-napping), javascript: hrefs (XSS), and http:// src/href attributes (mixed content). Advisory: missing CSP meta tag and external resources without integrity=.

Run the full audit suite (i18n, SEO, HTML/JS contract, drawings, CSS style, performance, accessibility, security) and return one aggregate report with a top-level ok flag.