Frequently Asked Questions

Getting Started

What exactly is DevSpark?

A set of markdown files -- prompt templates and helper scripts -- that give your AI coding assistant a repeatable workflow. It is not a program, a SaaS subscription, or an extension. Just files you drop into your repo.

What is "spec-driven development"?

You write a short specification (the what and why) before jumping to code. Your AI assistant uses that spec to generate a plan, break it into tasks, and implement -- all traceable back to the original requirements.

Isn't this overkill for small changes?

DevSpark has right-sized workflows:

Task Size Workflow Overhead
Typo / one-liner Just fix it None
Bug fix or small change /devspark.quickfix ~2 minutes
New feature, multi-file Full spec workflow ~10-15 minutes
Architectural change Full spec + /devspark.critic ~20 minutes

/devspark.specify is the default intake command. It will recommend the lighter or heavier path before creating artifacts.

Do I need to learn a new tool?

No. DevSpark is pure markdown. The AI agent does the heavy lifting.

If you want terminal-driven execution, there is an optional CLI runtime as well. That runtime adds devspark doctor, devspark harness ..., and devspark adapter ..., but the prompt-first workflow remains the default way to use DevSpark.

AI-Assisted Coding

Where do I start with AI coding assistants?

  1. Pick an agent (GitHub Copilot, Claude Code, Cursor, Gemini CLI, or any of 17+ supported agents).
  2. Bootstrap DevSpark with the Quick Start Guide.
  3. Follow the Implementation Lifecycle.

How do I know the AI-generated code is any good?

Three mechanisms: the constitution (your project's rules, enforced by every command), PR review (/devspark.pr-review checks code against the constitution), and site audit (/devspark.site-audit scans the full codebase). None replace your own review -- they augment it.

What if the AI makes a mistake?

It will. Specs catch misunderstandings before coding starts. The constitution defines hard requirements. The critic command (/devspark.critic) performs adversarial risk analysis. You review everything -- DevSpark never auto-commits or auto-deploys.

The Constitution

What is a "constitution"?

A markdown file (.documentation/memory/constitution.md) that defines your project's non-negotiable principles: coding standards, security requirements, testing expectations. Every DevSpark command reads and enforces it.

I already have coding standards in a wiki. Do I need this too?

The constitution is your coding standards in a format your AI assistant can use. If you have existing standards, use /devspark.discover-constitution to generate a draft from your codebase.

Can the constitution change over time?

Yes. Use /devspark.evolve-constitution to propose amendments based on PR review findings. See the Constitution Guide for details.

What goes in the constitution?

Architecture, code quality, testing, security, dependencies, and documentation rules. Example:

  • "All public functions MUST have typed parameters"
  • "Unit test coverage MUST exceed 80%"
  • "No hardcoded secrets. All SQL queries MUST be parameterized"

See the Constitution Guide for full examples.

Workflow and Commands

Where do I type slash commands?

In your AI agent's chat interface -- the same place you normally talk to your assistant. They are not terminal commands.

Do I have to use all 27 commands?

No. Most projects use a small subset of the 27 stock commands:

Everyday: /devspark.quickfix, /devspark.specify -> /devspark.plan -> /devspark.tasks -> /devspark.implement, /devspark.pr-review

Occasional: /devspark.constitution, /devspark.site-audit, /devspark.release

Specialized: /devspark.critic, /devspark.harvest, /devspark.evolve-constitution

What is the spec status lifecycle?

Every spec has a **Status**: field that transitions through three states:

Status Set by Meaning
Draft /devspark.specify Spec created, not yet implementing
In Progress /devspark.implement (start) Implementation underway
Complete /devspark.implement (all tasks done) Ready for PR review and merge

PR review blocks approval unless the spec is Complete with all tasks checked off. Site audit flags incomplete specs on main as critical anti-patterns.

What's the difference between quickfix and the full workflow?

Quickfix is for small, well-understood changes (one sentence, fewer than 3 files). Full spec workflow is for larger work where scope or design isn't immediately clear.

A middle path also exists: quick-spec. That route keeps intent, scope, constraints, and an action plan without the full specification overhead.

Can I customize commands for my team?

Yes. DevSpark uses three-tier resolution:

  1. Personal -- .documentation/{git-user}/commands/
  2. Team -- .documentation/commands/
  3. Stock -- .devspark/defaults/commands/

Use /devspark.personalize to create personal overrides.

Agent Compatibility

Which AI assistants work with DevSpark?

17+ agents including GitHub Copilot, Claude Code, Cursor, Gemini CLI, Windsurf, Amazon Q Developer, and more. See the Quick Start Guide for the full list.

Can I use multiple agents on the same project?

Yes. Stock prompts live in .devspark/defaults/commands/, repo overrides live in .documentation/commands/, and each agent gets thin shims that resolve personal, team, then stock prompts.

Project Setup

How do I add DevSpark to an existing project?

Run the matching quickstart prompt from the Quick Start Guide. Then use /devspark.discover-constitution to generate a constitution from your existing code patterns.

What files does DevSpark add?

  • .devspark/ -- Framework files (prompt defaults, scripts). The "engine."
  • .documentation/ -- Your project artifacts (specs, plans, constitution). "Your stuff."
  • Agent shims -- Platform-specific command files such as .claude/commands/ or .github/agents/

Uninstall removes .devspark/ without touching your work.

When you run the harness runtime, it also writes execution artifacts under .documentation/devspark/runs/ by default. Those are generated run records, not stock framework files.

What is the harness runtime?

An optional CLI execution layer for declarative engineering workflows. You define a YAML or JSON spec, validate it with devspark harness validate, execute it with devspark harness run, and inspect its event log with devspark harness trace.

See Harness Engineering for the full model.

Is the harness runtime part of the slash-command workflow?

No. The 27 stock slash commands remain prompt-driven. The harness runtime is additive and lives in the optional CLI.

What adapters are built in?

noop, manual, claude_code, copilot, and cursor.

  • noop is the safe default for dry runs, CI, and contract testing.
  • manual pauses for a human and requires a TTY.
  • claude_code, copilot, and cursor call their corresponding local CLIs when available.

Use devspark adapter list to inspect availability on the current machine and devspark adapter default <name> to save a local default.

Will DevSpark conflict with my existing tooling?

No. It only adds .devspark/, .documentation/, and optionally a few agent config files. It doesn't modify your build system, CI/CD, or linter config.

Can I use DevSpark with any programming language?

Yes. Specs and constitutions are markdown. The AI agent handles language-specific implementation.

Is my code sent anywhere?

DevSpark itself sends nothing anywhere. When you use an AI assistant, your code is processed by that AI provider per their terms. DevSpark adds no telemetry or data transmission.

Multi-App Monorepo Support

Do I need multi-app support?

Most projects do not. Consider it only when your monorepo has applications with different tech stacks, different governance rules, or needs per-app constitutions.

How do I enable it?

  1. Run /devspark.add-application for each application.
  2. Use --app <id> with any command to scope it.
  3. Run /devspark.validate-registry to verify consistency.

Multi-app is additive -- it doesn't change existing single-app behavior. See the Monorepo Guide.

Troubleshooting

For upgrade-specific issues, see the Upgrade Guide.

Slash commands aren't working

  1. Restart your IDE/editor completely (not just reload window).
  2. Verify agent files exist (e.g., ls .claude/commands/ or ls .github/prompts/).
  3. Re-run the quickstart prompt from the Quick Start Guide to refresh files.

The AI isn't following my constitution

  • Verify .documentation/memory/constitution.md exists and isn't empty.
  • Make principles specific and testable: "All public API endpoints MUST return proper HTTP status codes" not "write good code."
  • For large codebases, try working in smaller scopes.

Still Have Questions?

Edit this page