What Setup Questions Does build-iterated-agentic-loop Ask? Complete Configuration Guide

The build-iterated-agentic-loop skill asks nine targeted setup questions covering coding agent selection, schedule cadence, task definition, scope boundaries, validation commands, PR limits, metadata formatting, response templates, and iteration behavior.

The build-iterated-agentic-loop skill in the humanlayer/skills repository streamlines the creation of autonomous coding agents that run on GitHub Actions. Before generating any files, the skill conducts an interactive configuration session through nine precise questions. These questions capture everything needed to produce a working agent loop tailored to your repository.

Where the Questions Live

The complete question set is defined in plugins/build-iterated-agentic-loop/skills/build-iterated-agentic-loop/SKILL.md, specifically lines 34–58. The skill uses these prompts to gather parameters for generating three core artifacts: a local skill definition (SKILL.md), a GitHub Actions workflow (.github/workflows/agent-<task>.yml), and an agent memory file (.github/agent-memory/<task>.md).

The Nine Setup Questions Explained

1. Coding Agent Selection

Question: "Claude Code, Codex, OpenCode, or CodeLayer. Explain the required secret and headless command for the recommended choice."

This determines which LLM backend powers your agent loop. Each option requires a specific environment secret—such as CLAUDE_API_KEY for Claude Code or corresponding credentials for alternatives. The skill explains the exact headless command needed to run your chosen agent non-interactively.

2. Cadence Configuration

Question: "daily, weekly, weekdays, monthly, manual-only, or a custom cron expression. Recommend a cadence based on task risk and review burden."

This sets the schedule.cron value in the generated workflow file. The skill factors in how risky the automated changes are and how much review bandwidth your team has to suggest an appropriate frequency.

3. Task Definition

Question: "What concrete task should the agent loop accomplish? (Check for existing skills; optionally ask the user which parts of the codebase to inspect.)"

This captures the high-level objective—such as "fix ESLint violations" or "update outdated dependencies"—that shapes the entire agent prompt and workflow logic.

4. Scope Boundaries

Question: "Which directories or packages may the loop modify, and which should it only inspect?"

This establishes filesystem guardrails, preventing the agent from touching sensitive areas like infrastructure configurations or generated build artifacts.

5. Validation Requirements

Question: "Which command(s) must succeed before the agent is allowed to commit? (Propose defaults based on repo evidence.)"

The skill inspects your repository for common validation patterns—test runners, linters, type checkers—and proposes appropriate commands. These become the run steps that gate every agent commit.

6. PR Bounding

Question: "Should scheduled runs be skipped when an open PR from this agent already exists? If so, how many open PRs are allowed?"

This prevents PR explosion from overlapping agent runs. The default configuration caps the agent at one open PR, with logic implemented in the generated workflow to check existing PRs before creating new ones.

7. PR Metadata

Question: "Define the label name, PR title prefix, and branch prefix (e.g., agent-<task-slug>)."

This ensures consistent, recognizable PRs from your agent. Typical values might include labels like agent-fix-eslint, title prefixes with dates, and branch patterns like agent/eslint-$(date +%Y%m%d).

8. Response Format

Question: "How should the CI agent format its final response (the PR body)? Show the response-template.md reference and ask which details reviewers need."

The skill references plugins/build-iterated-agentic-loop/skills/build-iterated-agentic-loop/references/response-template.md to show example formats. You select which sections reviewers need—summary, risk assessment, verification steps, changed files, or full diffs.

9. Iteration Behavior

Question: "Should /iterate comments update the existing PR? If yes, ask where to place the helper script (agent-iteration.ts). If no, remove the iteration trigger from the workflow."

This enables or disables comment-driven refinement. When enabled, the skill places agent-iteration.ts in your requested location (typically .github/scripts/) and configures workflow triggers for issue comments containing /iterate.

Example Installation and Configuration

Install the skill and run the interactive configuration:

# Add the skill to your repository

npx skills add humanlayer/skills --skill build-iterated-agentic-loop

# Launch the interactive setup with all nine questions

skills run build-iterated-agentic-loop

A complete configuration session flows as follows:

? Choose a coding agent (default: CodeLayer) › CodeLayer
? Choose a cadence (default: weekdays) › daily
? What task should the loop perform? › Fix ESLint violations
? Which directories may be modified? › src/
? Validation command to run before committing? › npm run lint && npm test
? Bound PRs to 1 open PR? › Yes
? PR label name (default: agent-fix-eslint) › agent-eslint-fix
? PR title template › [${date}][Agent: ESLint Fix]: ${summary}
? Desired PR body format? › Show response-template.md example
? Enable /iterate support? › Yes
? Where to place iteration script? › .github/scripts/

Reference Files Supporting the Questions

File	Purpose
plugins/build-iterated-agentic-loop/skills/build-iterated-agentic-loop/SKILL.md	Contains the nine setup questions and skill logic
plugins/build-iterated-agentic-loop/skills/build-iterated-agentic-loop/references/workflow-template.yml	Base GitHub Actions workflow structure
plugins/build-iterated-agentic-loop/skills/build-iterated-agentic-loop/references/prompt-template.md	Template for embedded agent prompts
plugins/build-iterated-agentic-loop/skills/build-iterated-agentic-loop/references/response-template.md	Example PR body formats
plugins/build-iterated-agentic-loop/skills/build-iterated-agentic-loop/references/memory-template.md	Persistent agent memory structure
plugins/build-iterated-agentic-loop/skills/build-iterated-agentic-loop/references/agent-iteration.ts	/iterate comment handler script
plugins/build-iterated-agentic-loop/.claude-plugin/plugin.json	Claude marketplace metadata

Summary

• The build-iterated-agentic-loop skill poses nine setup questions to configure autonomous coding agents.
• Questions cover: agent selection, cadence, task, scope, validation, PR limits, metadata, response format, and iteration behavior.
• All questions are defined in SKILL.md lines 34–58 and feed into three generated artifacts: local skill, GitHub Actions workflow, and agent memory file.
• Reference templates in the references/ directory provide sensible defaults for each configuration area.

Frequently Asked Questions

What happens if I skip answering a setup question?

The skill provides sensible defaults for most questions—such as CodeLayer for agent and weekdays for cadence. You can accept defaults by pressing Enter, though critical parameters like task description require explicit input.

Can I reconfigure an existing agent loop without starting over?

Yes. Re-run skills run build-iterated-agentic-loop and answer the nine questions again. The skill regenerates all files, overwriting previous configurations. Commit history in your repository preserves prior iterations.

Does the skill detect my repository's existing validation setup?

The skill examines your repository structure—package.json scripts, Makefile targets, CI configurations—to propose validation commands in question 5. You can accept these proposals or specify custom commands.

What is the difference between manual-only and custom cron cadences?

Manual-only removes all schedule triggers from the workflow, requiring workflow_dispatch or API calls to run. Custom cron accepts any valid cron expression for precise scheduling beyond the presets (daily, weekly, weekdays, monthly).

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}