/**
 * MCP server instructions for non-plugin clients.
 * Condensed version of the 2 mandatory gates from docs/CLAUDE_TEMPLATE.md.
 * Sent in the MCP initialize response so any MCP client gets workflow guidance.
 */
export declare const TWINING_INSTRUCTIONS = "# Twining \u2014 Agent Coordination\n\nTwining provides persistent project memory: decisions survive context resets, new sessions start informed, and multi-agent work stays coordinated. State lives in `.twining/` as plain files.\n\n## 2 Mandatory Gates\n\n### Gate 1: Context Assembly (BEFORE working)\nYour FIRST tool call MUST be `twining_assemble` with your task description and scope. Do not read files or make changes until assemble returns. Also call `twining_why` on files you plan to modify.\n\n### Gate 2: Record (BEFORE committing or ending)\nCall `twining_record` with a summary and any decisions before every `git commit` or session end. Write decisions as natural sentences: \"Chose X over Y \u2014 reason\". The server handles routing to the decision store and blackboard.\n\n## Key Conventions\n- **Scopes** use path-prefix semantics: `\"src/auth/\"` not `\"project\"` \u2014 use the narrowest scope that fits\n- Never skip `twining_assemble` before work or `twining_record` before committing\n\n## Core Tools\n- **twining_assemble** \u2014 orient before working (Gate 1)\n- **twining_record** \u2014 record what you did before committing/ending (Gate 2)\n- **twining_post** \u2014 share findings, warnings, or needs during work\n- **twining_why** \u2014 check what decisions constrain a file before modifying it\n";
//# sourceMappingURL=instructions.d.ts.map