Module: pipeline/prompts/promptTiers

Single source of truth for tiered prompt content (MNT-009).

Local models (Ollama / mistral:7b) have effective context windows of ~4K–8K tokens. Cloud models (Anthropic, OpenAI, Google) have 128K+. This module centralises ALL tier-dependent prompt text so other modules import from one place instead of scattering local/cloud variants across files.

Architecture

  • selfHealing.js owns self-healing rules (CORE_RULES, SELF_HEALING_PROMPT_RULES) and exposes getPromptRules(tier).
  • promptTiers.js (this file) owns everything else that varies by tier: assertion rules, stability rules, code requirements, and config.
  • outputSchema.js assembles the final system prompt by importing from both.

Exports

  • getTier — Returns "cloud" or "local" based on the active provider.
  • TIER_CONFIG — Per-tier configuration (maxElements).
  • getAssertionRules — Tier-aware assertion rules.
  • getStabilityRules — Tier-aware stability rules.
  • getCodeRequirements — Tier-aware code requirement block.
Source:

Members

(static, constant) TIER_CONFIG :Object.<string, TierConfig>

Type:
  • Object.<string, TierConfig>
Source:

Methods

(static) getAssertionRules(tier) → {string}

Parameters:
Name Type Description
tier "cloud" | "local"
Source:
Returns:
Type
string

(static) getCodeRequirements(tier) → {string}

Parameters:
Name Type Description
tier "cloud" | "local"
Source:
Returns:
Type
string

(static) getStabilityRules(tier) → {string}

Parameters:
Name Type Description
tier "cloud" | "local"
Source:
Returns:
Type
string

(static) getTier() → {"cloud"|"local"}

Determine the prompt tier based on the active AI provider.

Source:
Returns:

The tier name.

Type
"cloud" | "local"

Type Definitions

TierConfig

Type:
  • Object
Properties:
Name Type Description
maxElements number

Max DOM elements to include in prompts.
Source: