INGCapaDev
Maintainer of this workflow, its core skills, and the OpenCode orchestration.
RepositoryA human-in-the-loop AI harness
Prompt less.
Engineer more.
Good software starts with clear thinking. Give your agent a process for understanding the problem, planning the work, and reviewing the code—with you in control.
Built for OpenCode / Reusable skills / Human judgment
[ capa / workflow ]PLAN.md
From an idea to code you understand.
A deliberate stopping point.
Each task ends with a reviewable candidate. You decide what happens next.
01 / FRAME the problem02 / PLAN the work03 / BUILD in small steps04 / PROVE the result
01 / The process
Think first. Build in small steps.
A lightweight loop for the decisions that matter. Run each stage to see what it produces.
~/ai-harness
interactiveworkflow stages
capa@workflow:~/ai-harness$/plan
Agree on the work before doing it.
Investigate the codebase, surface the meaningful decisions, and turn the ask into small, coherent tasks.
output: An approved PLAN.md with scope, decisions, evidence, and progress.
✓ ready for your decision
Small, understood change? Work directly. Planning and delegation scale with the problem.
02 / The skill library
Small skills. Clear responsibilities.
Core workflow skills, plus complementary skills linked to their original authors.
01 / Core
/planning
Frame decisions and turn an agreed approach into proportionate work.
02 / Core
/review
Assess a frozen candidate against quality standards and the plan.
03 / Core
/code-quality
Keep code readable, maintainable, and no more complex than needed.
04 / Core
/coding-conventions
Route work through the relevant TypeScript, React, Hono, Go, and architecture guidance.
05 / Core
/improve-ai
Learn from session evidence and propose focused workflow improvements.
06 / Core
/conventional-commits
Describe the actual change without implying permission to commit.
07 / Upstream
/writing-for-agents
Write agent-facing documents with precise context and triggers.
Original author credited in repository
08 / Upstream
/pr
Make pull requests easier to assess with concise summaries and evidence.
Original author credited in repository
03 / Credits & references
Shared ideas, clearly credited.
Matt Pocock / AI Hero
Composable skills and agent-facing writing that influence the complementary skills here.
Original repositoryGentleman Programming
Inspiration for orchestration, memory, fresh-context delegation, and review lifecycle patterns.
Gentle AIContext7 & Engram
The visual-summary approach referenced by the pull-request skill.
Original skillsStart with the source