Tool Surface Audit & Decision Quality

What's Changed in v1.2.0

🔧 MCP Tool Surface Reduction

• Removed 3 tools from the public MCP surface: search_project_glossary_fts, get_engrams_schema, and get_workspace_detection_info — all were redundant wrappers or diagnostic tools not intended for normal agent workflows.
• Merged 7 tools into 3 via routing parameters: verify_bindings now subsumes get_stale_bindings (via mode= param); manage_budget_config unifies the old get/set config pair; delete_item replaces four type-specific delete tools.
• Merged 3 more tools via flag parameters: dry_run=True on get_relevant_context replaces estimate_context_size; staleness_only=True and sections= on get_project_briefing replace get_briefing_staleness and get_section_detail.

📝 Tool Description Overhaul

• Replaced all 42 tool descriptions with audit-optimized versions following a three-question pattern: what it does, when to use it over alternatives, and what it returns.
• Standardized workspace_id field descriptions and documented all enum valid values and mutual exclusivity rules (e.g., content vs patch_content, __DELETE__ sentinel) across all 30+ tools.

📋 Strategy Template Updates

• Updated all 7 agent strategy templates (Roo Code, Cline, Cursor, Windsurf, Claude Code, Claude Desktop, Generic) to remove tool-selection guidance now embedded in descriptions and add merged_tools_note sections.
• Added decision quality criteria to all templates and rewrote post_task_completion steps to distinguish strategic decisions from implementation work.

🧠 Decision Quality Governance

• Established a formal quality gate: decision summaries must be prescriptive, not descriptive, and must pass the litmus test "Would this still be relevant in a completely new coding session?"
• Updated log_decision and log_progress tool descriptions to enforce the strategic vs. implementation distinction, preventing decision log pollution.