Configuration

Configuration Reference

All inputs and outputs for the Prompt2PR action.

---

Inputs

Inputs are configured via the with: block in your workflow step.

prompt (required)

The instruction describing what changes to make. This is sent to the LLM along with the contents of matching files.

with:
  prompt: 'Add input validation to all public functions'

For scheduled workflows, you can provide a default prompt with an optional manual override:

on:
  schedule:
    - cron: '0 9 * * 1'
  workflow_dispatch:
    inputs:
      prompt:
        description: 'Custom prompt (optional)'
        required: false
        default: ''

# In the step:
with:
  prompt: >-
    $

provider (required)

The LLM provider to use. Supported values:

Value	Provider	API Key Variable
mistral	Mistral AI	MISTRAL_API_KEY
openai	OpenAI	OPENAI_API_KEY
anthropic	Anthropic	ANTHROPIC_API_KEY
github	GitHub Models	GITHUB_TOKEN

Providers are interchangeable. Choose based on your preference — the action behaves the same regardless of provider.

model (optional)

Override the provider’s default model. If omitted, the default is used:

Provider	Default Model
mistral	mistral-large-latest
openai	gpt-4o
anthropic	claude-sonnet-4-20250514
github	openai/gpt-4o

For GitHub Models, use the publisher/model-name format:

with:
  provider: github
  model: openai/gpt-4o

paths (optional)

Default: ** (all files)

Comma-separated glob patterns for files to include as LLM context. Only matching files are scanned, and only matching files can be modified.

# Only TypeScript source files
paths: 'src/**/*.ts'

# Multiple patterns
paths: 'src/**,lib/**,*.md'

# Specific file types across the project
paths: '**/*.{ts,tsx,js,jsx}'

Note: The .github/ directory is always excluded regardless of the paths setting. Binary files are also automatically skipped. {: .notice–info }

max_files (optional)

Default: 10

Maximum number of files the LLM is allowed to modify in a single run. If the LLM’s response includes more file changes than this limit, the entire response is rejected and the run fails.

max_changes (optional)

Default: 200

Maximum total lines changed across all files. Like max_files, exceeding this limit rejects the entire response.

label (optional)

Default: prompt2pr

Comma-separated labels to apply to the created PR. The prompt2pr label is always included automatically even if you specify custom labels.

label: 'prompt2pr,automated,cleanup'

branch_prefix (optional)

Default: prompt2pr/

Prefix for the branch name. The full branch name is {prefix}{timestamp}.

branch_prefix: 'ai-fix/'
# Creates branches like: ai-fix/1707900000

dry_run (optional)

Default: false

When set to true, runs the full pipeline (scan, prompt, LLM call, parse, guardrails) but skips branch creation and PR submission. Useful for:

• Testing prompts before enabling real changes
• Auditing what the LLM would modify
• CI checks that validate LLM behavior

The files_changed and lines_changed outputs are still set, so you can inspect what would have happened.

base_url (optional)

Default: (empty — uses the provider’s default endpoint)

Override the LLM provider’s API base URL. Useful for:

• API proxies
• Self-hosted LLM endpoints
• Testing with mock servers

---

Outputs

Outputs are set after the action completes and can be referenced in downstream steps.

Output	Description
pr_url	URL of the created Pull Request. Empty if skipped.
pr_number	Number of the created Pull Request. Empty if skipped.
files_changed	Number of files modified. Set even in dry-run mode.
lines_changed	Total lines changed across all files. Set even in dry-run mode.
skipped	"true" if no PR was created (no changes or dry run). "false" otherwise.

Output Scenarios

Scenario	pr_url	pr_number	files_changed	lines_changed	skipped
PR created	https://github.com/.../pull/42	42	3	47	false
Dry run	(empty)	(empty)	3	47	true
No changes	(empty)	(empty)	0	0	true

Using Outputs

- uses: davd-gzl/Prompt2PR@v1
  id: p2pr
  with:
    prompt: 'Add missing JSDoc comments to all exported functions'
    provider: openai
  env:
    OPENAI_API_KEY: $
    GITHUB_TOKEN: $

# Only runs if a PR was created
- if: steps.p2pr.outputs.skipped != 'true'
  run: |
    echo "PR: $"
    echo "Changed $ files ($ lines)"

# Useful for dry-run reporting
- if: steps.p2pr.outputs.skipped == 'true'
  run: |
    echo "No PR created."
    echo "Would have changed $ files"

---

Next: Providers