AlgoBrain exposes 5 wiki tools with basic descriptions and parameter schemas. All tools have descriptions (10-200 chars) and input schemas with type definitions. However, schemas lack depth: no enums for constrained inputs (e.g., source_type, check type), no output schemas documented, and parameter descriptions are minimal (10-30 chars). Error handling is absent, tools return raw subprocess output or generic 'not found' messages without recovery guidance. No tool annotations (readOnlyHint, destructiveHint). wiki_ingest modifies state but lacks confirmation/dry-run pattern. Naming is clear (verb_noun), but parameter descriptions do not explain constraints, formats, or valid values. Output is unstructured text/JSON without pagination or field documentation.

• Add JSON Schema enums for source_type (article, pdf, video, tweet, screenshot, data, book, misc) and check (frontmatter, links, orphans, stale, empty, tags). This prevents hallucinated values and self-documents valid options.
• Document output schemas for all tools. E.g., wiki_search returns {data_instruction: {...}, results: [{path, title, type, status, relevance_score}]}. wiki_stats returns {total_pages, by_type, by_status, by_category}. LLMs need this to chain calls.
• Expand parameter descriptions to 50-100 chars and include constraints. E.g., 'Filter by tag (e.g., crypto, trading, risk-management; see wiki_stats for available tags)' or 'Specific check: frontmatter (YAML validity), links (broken refs), orphans (unreferenced pages), stale (>6mo old), empty (<100 chars), tags (missing/invalid).'
• Add tool annotations: mark wiki_search, wiki_lint, wiki_read, wiki_stats as readOnlyHint=true; mark wiki_ingest as destructiveHint=true. This helps agents reason about side effects.
• Implement error recovery guidance. E.g., if wiki_read returns 'Page not found', append 'Try wiki_search(query=...) to find similar pages.' If wiki_ingest fails, return 'Invalid source_type. Must be one of: article, pdf, video, tweet, screenshot, data, book, misc.'