Cross Document Analyzer

cross-document-analyzer

Internal sub-agent. This agent is not user-invokable. It is orchestrated automatically by document-accessibility-wizard when auditing multiple documents at once. You do not need to invoke it directly.

What It Does

cross-document-analyzer receives aggregated scan findings from multiple document audits and does three things:

1. Pattern detection — finds accessibility rules that fail across many files, identifies template-level issues, and classifies issues as systemic (all files), folder-level (one directory), or isolated (single document)
2. Severity scoring — computes a weighted 0–100 accessibility risk score for each document and assigns an A–F grade
3. Remediation tracking — when a baseline report is available, classifies every finding as Fixed, New, Persistent, or Regressed

When It Runs

The document-accessibility-wizard calls this agent whenever a multi-document audit is in progress:

• audit-document-folder prompt — after scanning all documents in the folder
• audit-changed-documents prompt — after scanning the git-changed files
• compare-audits prompt — when comparing two audit reports across time
• generate-vpat prompt — when aggregating conformance data across a library of documents

Scoring Formula

The agent computes a risk score for each document:

Score = 100 − (sum of weighted findings)
Minimum score: 0

Finding Type	Confidence	Deduction
Error	High	−10 points
Error	Medium	−7 points
Error	Low	−3 points
Warning	High	−3 points
Warning	Medium	−2 points
Warning	Low	−1 point
Tip	Any	0 points

Grade Table

Score	Grade	Meaning
90–100	A	Excellent — minor or no issues
75–89	B	Good — some warnings, few errors
50–74	C	Needs Work — multiple errors
25–49	D	Poor — significant accessibility barriers
0–24	F	Failing — critical barriers, likely unusable with assistive technology

Pattern Detection

The agent analyzes findings across all documents in the batch and classifies them:

• Systemic — the same rule fails in 90%+ of documents (e.g., no document has a title set, or all documents lack the language property). Fix the template or the authoring process, not individual files.
• Folder-level — the pattern is confined to a directory (e.g., all files under /legacy/ have image alt text missing). Suggests a batch from the same source or author.
• Isolated — the issue appears in only one or two documents. Address individually.

The pattern summary in the audit report groups issues in this order: systemic → folder-level → isolated. This ordering matches remediation priority: fix what affects the most people first.

Template Analysis

If document metadata reveals shared templates (Word's Template property, PowerPoint slide master names), the agent groups documents by template and calculates the proportion of documents per template that share each failing rule.

A "template-level fix" is flagged when the same rule fails across all documents sharing a template — this means fixing the template source will remediate all child documents automatically.

Remediation Tracking

When baseline report data is provided (by the compare-audits or audit-changed-documents prompts), every finding is classified:

Classification	Meaning
Fixed	Present in baseline, absent in current scan
New	Absent in baseline, present in current scan
Persistent	Present in both scans
Regressed	A Fixed issue has returned

Progress metrics computed:

• Total findings: baseline count → current count → delta
• Score: baseline average → current average → delta
• Velocity: projected cycles to reach zero at current fix rate

Output Format

The agent returns a structured analysis ready for inclusion in the audit report markdown:

• Cross-document pattern summary with frequencies and classifications
• Per-document severity scores and grades (scorecard table)
• Overall average score and grade
• Template analysis section (if templates detected)
• Remediation progress summary (if baseline provided)

Connections

Component	Role
document-accessibility-wizard	Orchestrating wizard that calls this agent and incorporates its output into the final report
document-inventory	Runs before this agent — provides the file list the wizard scans
report-generation skill	Scoring formulas and grade table this agent implements
compare-audits prompt	User-facing prompt that triggers remediation tracking analysis