Skip to main content
Modes are specialized workflows that guide how Pullfrog handles different types of tasks. When Pullfrog receives a prompt, the agent analyzes the request and selects the most appropriate mode.

​
How mode selection works

Mode selection happens automatically at the start of each agent run:
  1. Pullfrog spawns an agent that reads all configured modes (built-in + custom)
  2. The agent analyzes the prompt against each mode’s description
  3. The agent calls the select_mode tool with the best-matching mode
  4. The agent receives guidance and instructions for that mode and executes the task
You don’t need to specify which mode to use—Pullfrog figures it out from context. For example, a comment like “Please review this PR” will automatically trigger Review mode, while “Add dark mode support” will trigger Build mode.

​
Custom instructions

Set task-specific instructions in your repository console under Automations. Each card ends with a Custom instructions field. Review PRs, Plan, Build, Address reviews, and Fix CI instructions supplement their built-in workflows on both automatic runs and direct requests, and win where the two conflict. Re-reviews share the Review PRs instructions. Mentions, Enrich issues, and Label issues have separate event prompts. Issue prompts supplement the selected Plan or Build behavior; in Enrich mode, the prompt replaces the default enrichment prompt. Mention prompts provide standing guidance without replacing the request. Instructions for every run stay under Agent → Standing instructions. Plan and Build cards do not enable automatic runs. Use Enrich issues to choose what happens when an issue is opened.

​
Built-in modes

Pullfrog comes with eight built-in modes:

Build

Implement, build, create, or develop code changes. Make specific changes to files or features. Execute a plan.Creates a branch, makes code changes, runs tests, pushes with a clean working tree (optional repo prepush hooks may run tests/lint first), and opens a PR.

Address Reviews

Address PR review feedback. Respond to reviewer comments. Make requested changes to an existing PR.Reads each full review thread before responding. On human-authored PRs, questions and discussion receive replies without code changes; an @pullfrog mention alone is not an edit request. Explicit change requests can produce commits and pushes.On PRs Pullfrog opened, enabled review automation still addresses actionable findings automatically. Later replies can limit or withdraw a request. Discussion-only replies leave threads open unless asked to resolve them.These are agent instructions, not an approval gate. Set the action’s push: disabled input to block all code pushes.

Review

Review code, PRs, or implementations. Provide feedback or suggestions. Check code quality, style, and correctness.Analyzes the PR diff and submits a review with inline comments.

Incremental Review

Re-review a PR after new commits are pushed. Focus on new changes since the last review.Generates an incremental diff from the previous HEAD, fetches prior review feedback to avoid duplication, and submits a focused review on only the new changes.

Fix

Fix CI failures. Debug failing tests or builds. Investigate and resolve check suite failures.Fetches failure logs, reads workflow files to understand how CI runs, reproduces locally with exact CI commands, fixes the issue, and pushes.
Pullfrog can start a Fix run on its own when CI fails on a PR. See Auto-fix CI failures.

Plan

Create plans, break down tasks, outline steps. Analyze requirements and understand scope of work.Creates a structured plan with clear milestones and dependencies.

Resolve Conflicts

Resolve merge conflicts in a PR branch against the base branch.Fetches the base branch and merges. If the merge is clean, pushes (optional repo prepush hooks may run) and finishes. If there are conflicts, resolves them, verifies, commits, pushes, and reports.

Task

General-purpose tasks that don’t fit other modes: answering questions, adding comments, labeling, running ad-hoc commands, or any direct request.Reads the request, gathers context if needed, performs the task, and creates a branch/PR if code changes are involved.