OmniAudit defines 17 tools across three modules (AuditLens, SiteLint, ReleaseButler, Core) with explicit JSON schemas and descriptions. All tools have input schemas with typed properties and most have descriptions (10 - 150 chars). However, descriptions are often terse and lack actionable context for LLM selection. Parameter descriptions are sparse or missing entirely, most params lack guidance on format, constraints, or dependencies. Output schemas are not documented. Error handling is minimal. Tool naming follows verb_noun convention but some names are generic (e.g., 'get_job', 'list_receipts'). No confirmation patterns for destructive operations (create_issue, create_release, export_report). Composition is reasonable, tools chain via IDs, but some tools conflate multiple concerns (e.g., create_release uploads assets in one call).

• Add parameter descriptions to all tools. For each param, specify format (e.g., 'repo: GitHub repository in owner/name format'), constraints (e.g., 'pr_number: positive integer'), and dependencies (e.g., 'branch: optional; if omitted, defaults to default branch'). Target 50 - 100 chars per param description.
• Document output schemas for all tools. Include field names, types, and what IDs/references are available for chaining. E.g., list_runs should document that it returns run_id, status, and created_at so agents know what to pass to fetch_evidence.
• Add confirmation patterns to destructive tools. Implement confirmation_token validation for create_issue, create_release, and export_report. Require agents to call a separate 'confirm_*' tool or pass a token obtained from a dry-run preview.
• Expand tool descriptions to 100 - 150 chars. Include WHEN to use (e.g., 'Call after start_scan to poll job progress'), prerequisites (e.g., 'Requires a valid job_id from start_scan'), and what the tool returns (e.g., 'Returns job status, progress percentage, and error details if failed').
• Rename generic tools for clarity. Change 'core.get_job' → 'core.get_scan_job', 'core.list_receipts' → 'core.list_audit_receipts'. This reduces LLM confusion when many tools are available and improves intent inference from tool names alone.