/**
 * The Karpathy-guidelines context block, machine-injected into every sub-agent
 * prompt by the dispatch prompt composer
 * (`src/cli/commands/dispatch-prompt-composition.ts`), the same way the Test
 * Tool Detection block is injected by `./test-tool-detection.ts`.
 *
 * Why this module exists: the obligation to deliver this block lived only as
 * prose in `skills/bee/peaks-rd/references/rd-sub-agent-dispatch.md`, addressed
 * to the orchestrator. Measured 2026-10-12: four RD dispatch prompts in one
 * session carried zero occurrences of `Karpathy`, and the one block that WAS
 * appended had a word wrong — which the verbatim requirement makes a defect.
 * An obligation a busy caller can forget is not an executor; this is.
 *
 * SOURCE OF TRUTH: the fenced block in the reference file's
 * §"Karpathy-guidelines context" section is the canonical human-facing text;
 * `KARPATHY_GUIDELINES_BLOCK` below is its machine copy, and
 * `tests/unit/services/dispatch/karpathy-block.test.ts` asserts the two are
 * byte-equal, so the copy cannot drift silently. A second copy that CAN drift
 * is the same defect class as the dropped block; a pinned one is not.
 *
 * Why a static constant rather than a per-call read of the reference file
 * (same reasoning as `./test-tool-detection.ts`): dispatch is a hot path, and
 * the skill tree is not guaranteed to be present in an installed bundle.
 */
/**
 * The canonical block, verbatim. Byte-for-byte the fenced block under
 * §"Karpathy-guidelines context" in the RD dispatch reference, including its
 * trailing newline.
 */
export declare const KARPATHY_GUIDELINES_BLOCK = "Karpathy-guidelines context (REQUIRED \u2014 read before any code action)\n\nThe 4 Karpathy guidelines are mandatory for every planning, implementation, refactor, and review action you take. If 200 lines could be 50, rewrite. Touch only what the request requires. Surface assumptions, do not hide them. Define verifiable success criteria.\n\n1. Think Before Coding\n\nDon't assume. Don't hide confusion. Surface tradeoffs. Before implementing: state assumptions explicitly; if multiple interpretations exist, present them; if a simpler approach exists, say so; if unclear, stop and ask.\n\n2. Simplicity First\n\nMinimum code that solves the problem. Nothing speculative. No features beyond what was asked. No abstractions for single-use code. No \"flexibility\" or \"configurability\" that wasn't requested. No error handling for impossible scenarios. If you write 200 lines and it could be 50, rewrite it. Ask yourself: \"Would a senior engineer say this is overcomplicated?\" If yes, simplify.\n\n3. Surgical Changes\n\nTouch only what you must. Clean up only your own mess. When editing existing code: don't \"improve\" adjacent code, comments, or formatting; don't refactor things that aren't broken; match existing style. When your changes create orphans: remove imports/variables/functions that YOUR changes made unused. The test: every changed line should trace directly to the user's request.\n\n4. Goal-Driven Execution\n\nDefine success criteria. Loop until verified. \"Add validation\" \u2192 write tests for invalid inputs, then make them pass. \"Fix the bug\" \u2192 write a test that reproduces it, then make it pass. For multi-step tasks, state a brief plan with verify checkpoints. Strong success criteria let you loop independently. Weak criteria require constant clarification.\n";
/** True when `text` already carries the canonical block (see the boundary note above). */
export declare function hasKarpathyBlock(text: string): boolean;
/**
 * Ensure `text` carries the canonical block. Pure and idempotent.
 *
 * - `text` already carries the block (per {@link hasKarpathyBlock}) → returned
 *   unchanged, byte for byte. This is the compliant caller's guarantee: no
 *   double-append and no reflow of a prompt that already has the block.
 * - otherwise → `text` with the block appended, and `text` itself untouched.
 *
 * Near-miss policy, stated rather than left implicit: when the caller carries a
 * copy whose body differs, the canonical block is injected anyway and the
 * caller's copy is left in place. Deleting text a caller wrote is not this
 * function's job; delivering the canonical text is. The cost is two
 * "Karpathy-guidelines context" headings in that prompt — visible, and the one
 * the child must obey is the canonical one.
 *
 * The caller applies this to the assembled prompt, not to the caller's draft —
 * see `composeDispatchPrompt`, which runs it on `effectivePrompt` (the exact
 * string handed to the sub-agent as `toolCall.args.prompt` and recorded as the
 * dispatch record's `prompt`). The caller's draft is inside that string
 * verbatim, so a draft-side copy is matched too, and the check also catches a
 * block some other contributor put in the assembled text.
 */
export declare function ensureKarpathyBlock(text: string): string;