Configuration

PromptLint is configured via a .promptlintrc YAML file in your project root. All settings are optional — sensible defaults are applied automatically.

Minimal Config

# .promptlintrc
model: gpt-4o
token_limit: 800

Full Reference

# .promptlintrc

# ── Global ──────────────────────────────────────────────────────
model: gpt-4o           # tiktoken model for exact token counts
token_limit: 800        # CRITICAL if prompt exceeds this
cost_per_1k_tokens: 0.005  # USD per 1k input tokens
calls_per_day: 10000    # for cost projection

structure_style: auto   # "xml" | "headings" | "markdown" | "auto" | "none"

display:
  preview_length: 60    # chars to show in finding preview
  context_width: 80     # terminal width for wrapping

# ── Rules ────────────────────────────────────────────────────────
rules:
  # Cost
  cost: true
  cost_limit: true

  # Security
  prompt_injection:
    enabled: true
    patterns:
      - ignore previous instructions
      - system prompt extraction
      - you are now a [a-zA-Z ]+
      - disregard all prior
      - forget everything
      - new instructions:
  jailbreak_pattern: true
  secret_in_prompt: true
  pii_in_prompt:
    enabled: true
    check_email: true
    check_phone: true
    check_ssn: true
    check_credit_card: true
  context_injection_boundary: true

  # Structure
  structure_sections: true
  role_clarity: true
  output_format_missing: true
  hallucination_risk: true

  # Quality
  clarity_vague_terms: true
  specificity_examples: true
  specificity_constraints: true
  politeness_bloat:
    enabled: true
    allow_politeness: false
  verbosity_redundancy: true
  verbosity_sentence_length: true
  actionability_weak_verbs: true
  consistency_terminology: true
  completeness_edge_cases: true

# ── Auto-fix ────────────────────────────────────────────────────
fix:
  enabled: true
  prompt_injection: true
  politeness_bloat: true
  verbosity_redundancy: true
  structure_scaffold: true
  normalize_spacing: true

Global Settings

model

Type: string · Default: gpt-4o

The tiktoken model used for exact token counting. Any model supported by tiktoken works.

model: gpt-4o
# model: gpt-4
# model: gpt-3.5-turbo
# model: claude-3-5-sonnet-20241022  # falls back to cl100k_base

token_limit

Type: integer · Default: 800

Triggers cost-limit (CRITICAL) when the prompt exceeds this count. Set to 60–70% of your model's context window to leave room for completions.

token_limit: 800   # GPT-4o has 128k context; 800 keeps system prompts lean

cost_per_1k_tokens

Type: float · Default: 0.005

Input token price in USD per 1,000 tokens. Used by the cost rule to project spend.

Model	Input price
GPT-4o	0.005
GPT-4o mini	0.00015
GPT-4	0.03
GPT-3.5 Turbo	0.0015
Claude 3.5 Sonnet	0.003
Claude 3 Haiku	0.00025

calls_per_day

Type: integer · Default: 1000000

Used to project daily and monthly cost savings. If ≥ 100,000, cost projections are suppressed (considered a load test scenario).

structure_style

Type: string · Default: auto

Controls what format the structure-sections auto-fix scaffolds:

Value	Effect
xml	Wraps in <task>, <context>, <output_format> tags
headings	Adds Task:, Context:, Output Format: headings
markdown	Adds ## Task, ## Context, ## Output Format sections
auto	Detects existing style and matches it
none	Disables auto-fix scaffolding

Rule Configuration

Each rule can be enabled/disabled and have its severity overridden:

rules:
  # Simple toggle
  completeness_edge_cases: false

  # Override severity
  specificity_examples:
    enabled: true
    level: warn   # promote from INFO → WARN

  # Custom patterns (prompt-injection)
  prompt_injection:
    enabled: true
    patterns:
      - ignore previous instructions
      - admin override:
      - bypass safety

Disable a Rule

rules:
  actionability_weak_verbs: false
  verbosity_sentence_length: false

Severity Override

rules:
  specificity_examples:
    enabled: true
    level: warn    # or "critical", "info"

Team Presets

Security-First

model: gpt-4o
token_limit: 1000

rules:
  prompt_injection:
    enabled: true
    patterns:
      - ignore previous instructions
      - system prompt extraction
      - you are now a [a-zA-Z ]+
      - admin mode
      - bypass restrictions
  jailbreak_pattern: true
  secret_in_prompt: true
  pii_in_prompt:
    enabled: true
    check_email: true
    check_phone: true
    check_ssn: true
    check_credit_card: true
  context_injection_boundary: true
  cost: false
  specificity_examples: false
  specificity_constraints: false
  completeness_edge_cases: false

Cost-First

model: gpt-4o
token_limit: 400
cost_per_1k_tokens: 0.005
calls_per_day: 50000

rules:
  cost: true
  cost_limit: true
  politeness_bloat: true
  verbosity_redundancy: true
  verbosity_sentence_length: true
  prompt_injection: true
  structure_sections: false
  specificity_examples: false
  completeness_edge_cases: false

fix:
  enabled: true
  politeness_bloat: true
  verbosity_redundancy: true

Quality-First

model: gpt-4o

rules:
  structure_sections: true
  role_clarity: true
  output_format_missing: true
  hallucination_risk: true
  clarity_vague_terms: true
  specificity_examples:
    enabled: true
    level: warn
  specificity_constraints:
    enabled: true
    level: warn
  completeness_edge_cases:
    enabled: true
    level: warn
  actionability_weak_verbs:
    enabled: true
    level: warn
  cost: false
  cost_limit: false

Config File Location

PromptLint searches for .promptlintrc starting from the current directory, walking up to the filesystem root (similar to ESLint). You can also pass a config explicitly:

promptlint --file prompt.txt --config /path/to/custom.yaml