AI-Assisted Website Audit: How We Quality-Checked 22 Pages in One Session
Session Date: 2026-02-13
Project: ReportsHub (integritystudio.io)
Focus: Readability & Accessibility Audit Across All Report Pages
Session Type: Quality Audit + Documentation
Session ID: 01af120d-c2c7-4a0b-8e65-1c0276987bc0
What Happened
We ran an AI-assisted audit of the entire Integrity Studio reports hub – a collection of 22 HTML pages spanning client research reports, opportunity analyses, market studies, and operational dashboards. The goal: find every readability and accessibility issue that could make these reports harder to use on phones, in dark mode, or for readers who need larger text.
The session produced a comprehensive backlog document (docs/BACKLOG.md) with 17 prioritized issues, a graded scorecard for every project directory, and a phased plan to bring all pages up to a consistent quality bar.
What We Found
The Good News
One report – the Capital City Village analysis – already met every standard we checked for. It became our “gold standard” reference: proper dark mode support, responsive tables, print-friendly styles, keyboard navigation, and accessible font sizing. This proved the template was achievable; most other pages just hadn’t caught up yet.
The Issues (By Priority)
We categorized findings into four severity levels:
| Priority | Count | Examples |
|---|---|---|
| Critical | 3 | No dark mode on 15 pages; no mobile breakpoints on 11 pages; tables overflow on small screens |
| High | 4 | Fixed sidebar breaks mobile layout; external font dependency violates self-contained rule; inconsistent color scheme conventions |
| Medium | 6 | Missing semantic landmarks; small source text; low-contrast opacity values; paragraphs too wide for comfortable reading |
| Low | 4 | No skip-to-content links; no print styles; CSS typo in one file |
Template Families
The audit revealed that most pages fall into one of five template families, which means fixes can be applied in batches rather than file-by-file:
- Research Report template (11 pages) – the most common, used across balloon-collective, edgar_nadyne, holliday_lighting, integrity-studio-ai, trp-austin, and zoukmx
- Capital City template (2 pages) – the reference standard
- Leora Forms/Dashboard (4 pages) – form and dashboard layouts with pixel-based sizing
- Hub/Portal pages (4 pages) – navigation portals linking to reports
- Unique layouts (1 page) – the TRP competitor analysis with its inverted dark-first design
Scorecard Snapshot
| Directory | Grade | Key Issue |
|---|---|---|
| capital_city | A | Reference implementation |
| leora_research | B+ | Portal is solid; forms use px sizing |
| holliday_lighting | B- | Mixed quality; one file has external font dependency |
| edgar_nadyne | C+ | No dark mode; inline grid layouts break on mobile |
| balloon-collective | C | No dark mode or responsive design |
| integrity-studio-ai | C | No dark mode or responsive design |
| trp-austin | C- | Inconsistent conventions between two files |
| ngo-market | D+ | Fixed sidebar unusable on mobile |
| zoukmx | C | No dark mode; otherwise clean |
How the AI Session Worked
The entire audit was conducted through a Claude Code session with OpenTelemetry instrumentation tracking every operation. Here’s what the telemetry captured:
Session Telemetry Overview
| Metric | Value |
|---|---|
| Session ID | 01af120d-c2c7-4a0b-8e65-1c0276987bc0 |
| Working Directory | /Users/alyshialedlie/reports |
| Git Branch | main |
| Repository | integritystudio/reports |
| Node.js | v25.6.0 |
| npm | 11.8.0 |
What the AI Did (Tool Operations)
The session executed 39 tool operations with a 100% success rate – no failed operations across the entire audit.
| Tool | Count | What It Did |
|---|---|---|
| Bash | 34 | Directory listing, file discovery, telemetry queries |
| TaskCreate | 1 | Created the audit tracking task |
| TaskUpdate | 2 | Updated task progress and completion |
| Edit | 1 | Modified existing file content |
| Write | 1 | Created the BACKLOG.md deliverable |
Specialized Agents
The session launched an Explore agent to quickly scan the repository structure and identify all HTML files while excluding submodule directories (ai-news, ai-observability, calender-updates, isabel_budenz, john_skelton).
Observability Pipeline
Every tool call, agent launch, and skill check was recorded as an OpenTelemetry span and exported to both local storage and SigNoz Cloud:
| Signal | Destination | Status |
|---|---|---|
| Traces | Local JSONL + SigNoz Cloud | Active |
| Logs | Local JSONL + SigNoz Cloud | Active |
| Metrics | SigNoz Cloud | Active |
The session generated 54 trace spans and 73 log entries across the audit workflow:
| Span Type | Count | Purpose |
|---|---|---|
builtin-post-tool | 37 | Tracked every tool completion |
skill-activation-prompt | 7 | Checked for matching skills on each prompt |
session-start | 3 | Session initialization and resumptions |
builtin-pre-tool | 3 | Pre-tool validation checks |
agent-pre-tool / agent-post-tool | 2 | Explore agent lifecycle |
plugin-pre-tool / plugin-post-tool | 2 | Plugin integration hooks |
Log Activity
| Severity | Count | Examples |
|---|---|---|
| INFO | 50 | Tool completions, session starts, context tracking |
| DEBUG | 23 | Skill activation checks, environment validation |
Broader Context: Today’s Platform Activity
This audit session was part of a busy day across the platform:
- 86,400 total hook executions recorded in the performance log
- 23,180 trace spans and 45,701 log entries across all sessions today
- 35 session starts across multiple projects
- 22 agent launches including Explore, general-purpose, code-reviewer, and webscraping-research-analyst agents
What Happens Next
The BACKLOG.md document organizes fixes into four phases:
- Phase 1 – Critical Fixes: Add dark mode and responsive breakpoints to the 11 research-template pages (highest impact, moderate effort)
- Phase 2 – High Priority: Fix the ngo-market sidebar, remove external font dependency, standardize color scheme conventions
- Phase 3 – Medium Priority: Add semantic HTML landmarks, improve contrast ratios, constrain paragraph widths
- Phase 4 – Polish: Add skip-to-content links, print styles, fix the CSS typo
The phased approach means the most impactful improvements – dark mode and mobile support – ship first, with diminishing-priority refinements following.
Key Takeaway
An AI-assisted audit found 17 distinct issues across 22 pages, categorized them by severity, identified that batch fixes across template families could address most problems efficiently, and produced a structured backlog – all in a single session. The OpenTelemetry instrumentation made it possible to verify that every operation completed successfully and to reconstruct the exact workflow after the fact.
Generated from OpenTelemetry session data exported to SigNoz Cloud. All telemetry signals (traces, metrics, logs) are available for further analysis in the SigNoz dashboard.