Development Workflow

Hyprlayer gives you a spec-driven development workflow. Each step writes an artifact to the thoughts directory, where it survives the session and your teammates can read it.

The Workflow

Research → Plan → Implement → Commit → Validate → Review → Ship

1. Research

/research_codebase

Explore and document how existing code works. The agent spawns parallel sub-agents to analyze different parts of the codebase and writes findings to the thoughts directory, where a future session can pick them up.

When to use: Before starting any significant work, when onboarding to unfamiliar code, or when you need to understand how components interact.

2. Plan

/create_plan

Build a phased implementation plan through interactive research and iteration. The agent works with you to understand the problem, research the codebase, and produce a detailed technical specification with success criteria for each phase.

The agent saves plans to thoughts/shared/plans/, and you refine them with /iterate_plan.

When to use: Before implementing any non-trivial feature or change.

3. Implement

/implement_plan thoughts/shared/plans/2026-03-08-my-feature.md

Execute the plan phase-by-phase. The agent reads the plan, implements each phase, verifies success criteria, and ticks the plan’s checkboxes as it finishes each section.

When to use: After you’ve reviewed and approved a plan.

4. Commit

/commit

Create atomic git commits for the changes. The agent reviews what changed, groups related files, drafts commit messages, and asks for confirmation before committing.

5. Validate

/validate_plan thoughts/shared/plans/2026-03-08-my-feature.md

Spawns an inspector sub-agent per phase group to audit the diff and run every automated criterion, then promotes the plan’s status to implemented only if no phase, check, or manual item is outstanding.

When to use: After /commit, since validation reads the git history to find the implementation diff.

6. Review

/code_review

Run an adversarial review of the branch’s diff against its base. The skill picks the codex CLI for a cross-model second opinion when it finds one, and otherwise spawns the adversarial-reviewer subagent in a fresh context window. It reads and never writes. The reviewer hunts for the ways your code will fail in production.

When to use: Right before opening a PR, after /commit has produced the diff you want reviewed.

7. Ship

/describe_pr

Generate a PR description from your repository’s template (thoughts/shared/pr_description.md on git/obsidian; a PR Description Template page/object on notion/anytype). The agent analyzes the diff, asks whether to run the template’s verification commands, hands the write-up to the herald agent, and updates the PR.

Supporting Commands

Command	Purpose
/iterate_plan	Refine an existing plan based on feedback
/create_handoff	Write a handoff document to transfer context to another session
/resume_handoff	Pick up work from a handoff document
/local_review	Set up a worktree to review a colleague’s branch
/founder_mode	Retroactively create a JIRA ticket and PR for already-implemented work
/cost_estimate	Estimate what the current codebase would have cost to build, and the AI-assisted ROI
/hyprlayer_doctor	Verify the configured thoughts backend is reachable before writing to it