agentfootprint
Version:
The explainable agent framework — backtrack a wrong answer to the exact context that caused it (evidence, not guesses). Built on footprintjs.
40 lines • 1.57 kB
TypeScript
/**
* defineSteering — sugar for always-on system-prompt Injections.
*
* Steering docs are the simplest form of context engineering: a fixed
* piece of guidance the LLM should follow on every iteration. Style
* guides, output format rules, persona statements, safety policies.
*
* Produces an `Injection` with:
* - flavor: `'steering'`
* - trigger: `{ kind: 'always' }`
* - inject: `{ systemPrompt: prompt }`
*
* @example
* const jsonOnly = defineSteering({
* id: 'json-only',
* description: 'Always respond with valid JSON.',
* prompt: 'Respond with JSON only. No prose. No markdown.',
* });
*/
import type { Injection } from '../types.js';
import type { CachePolicy } from '../../cache/types.js';
export interface DefineSteeringOptions {
readonly id: string;
readonly description?: string;
/** Always-on text appended to the system-prompt slot. */
readonly prompt: string;
/**
* Cache policy for this steering injection. Defaults to `'always'`
* — steering is by definition always-on stable content, ideal for
* provider-side caching. Override with `'never'` if the prompt
* contains volatile content (timestamps, per-request IDs).
*
* See `CachePolicy` in `agentfootprint/src/cache/types.ts` for all
* variants. The CacheDecision subflow reads this from
* `injection.metadata.cache` each iteration.
*/
readonly cache?: CachePolicy;
}
export declare function defineSteering(opts: DefineSteeringOptions): Injection;
//# sourceMappingURL=defineSteering.d.ts.map