DualContext exposes 2 tools with basic descriptions but critical gaps in schema completeness and parameter documentation. run_splunk_query has a partial input schema (query, earliest) but lacks output schema documentation. query_context similarly lacks output schema. Neither tool has parameter descriptions beyond the bare names. Descriptions are present but generic (194-char baseline for A+ tools; these are ~80-100 chars). No error handling guidance, no idempotency hints, no permission declarations. Tools are read-only (good) but lack the structured output and chaining metadata needed for reliable agent composition. Schema visibility is limited to input parameters only, output structures are not documented in the source.

• Document output schemas for both tools. E.g., run_splunk_query returns {rows: [{field1, field2, ...}], total_events: int, sources: [string]}. query_context returns {files: [{name, signature, relevance_score}], tokens_used, tokens_raw, reduction_pct}.
• Add parameter descriptions with format/constraint hints. E.g., 'earliest: Time range for search (e.g., -24h, -7d, 2024-01-15). Default: -24h.' and 'topK: Number of top-ranked files to return (1 - 50, default 5).'
• Add error recovery guidance. E.g., 'If query syntax error: check SPL documentation. If Splunk unavailable: retry after 30s. If no results: broaden time range or simplify query.'
• Declare idempotency and side effects. E.g., 'run_splunk_query is read-only and idempotent. query_context ranking is deterministic given the same query and codebase snapshot.'
• Rename query_context to search_code_context or rank_files_by_relevance to clarify the action and distinguish from run_splunk_query.

Inferred effective spec: <=2025-11-25.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained