implementkit

Implement plans, specs, or issues into working code with mode resolution and a test-build gate.

1|Updated Jul 12, 2026
One-click install
npx skills add https://github.com/mimukit/skills --skill implementkit-mimukit
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: implementkit
Source: https://github.com/mimukit/skills/tree/main/skills/implementkit
Command: npx skills add https://github.com/mimukit/skills --skill implementkit-mimukit

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning a settled plan or issue into code often goes wrong in two ways: the agent guesses at an underspecified design, or it declares done without running the repo's own checks. This Skill enforces a disciplined build step that resolves how to build (straight-through vs TDD), bounces thin inputs back for hardening, and proves the work with the repo's test and build gate before calling it done. ## Core Features & Use Cases - Explicit input handling: Builds from a named plan file, GitHub issue, freeform spec, or a fix round of review findings, with support for narrowing to a single phase. - Mode resolution by precedence: Picks straight-through or TDD via prompt, then agent instructions (CLAUDE.md), then repo habit, then a single ask, never imposing TDD silently. - Done-gate enforcement: Discovers the repo's own test, build, typecheck, and lint commands and requires them all to pass, with bounded fix attempts on failure. - Plan stamping and handoff: Marks built phases with a date stamp and leaves all changes unstaged for the commit step. - Use Case: You say "implement phase 2 of docs/plans/plan-sso-2026-07-23.md the TDD way" and get red-green-refactor implementation, a green test suite, a stamped plan, and unstaged changes ready to commit. ## Quick Start Ask the agent to implement a specific plan file, issue number, or spec, optionally naming a phase or a TDD preference, and let it run the repo's test and build gate before reporting done.

Frequently Asked Questions about implementkit

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I implement a plan or GitHub issue into code with an AI agent?

Name the input explicitly, such as a plan file path, an issue number fetched with gh issue view, or a freeform spec. The agent reads it in full, builds the code in the resolved mode, and runs the repo's test and build commands before declaring done.

How does the skill decide between TDD and straight-through implementation?

Mode resolution follows a fixed precedence: an explicit prompt instruction wins first, then the repo's agent instructions like CLAUDE.md, then inferred repo habit requiring both test infrastructure and real test-shipping practice, and finally a single ask. Non-interactive runs default to straight-through.

Can I implement only one phase of a larger plan?

Yes, both plan files and issues carry phases as headings, so you can narrow a run to a single phase. The agent still reads the whole document to know what earlier phases provide and what later phases must not be broken.

What happens if the spec or issue is too vague to implement?

The agent stops and bounces the input back instead of guessing the design, naming the specific gaps and pointing to plan-hardening steps. A fix round of concrete review findings always passes this bar because the defects are already named.

Does the skill commit the code it writes?

No, it never stages or commits. All changes are left unstaged with a summary of the mode used, files changed, gate results, and stamped phases, so a separate commit step can group and commit the work.

What happens when the test or build gate keeps failing?

The agent attempts to fix its own output with a bound of roughly three attempts. If the gate is still red, it stops, reports the failure and what was tried, and never declares done on a failing gate.