Implementation Planner
SkillDev toolsUse after requirements, contracts, and visual gates pass to create a TDD-ready implementation plan.
Use Implementation Planner in Claude, ChatGPT or Ahel Desktop
Free. Sign in, add Implementation Planner and connect your AI. About a minute.
Also: Claude Code · Cursor · Codex
Then ask your AI: use the Implementation Planner skill
Details
Instructions available. Your AI can read the instructions. Execution depends on the setup they require.
Account requirements not reviewed. Check the skill instructions before use; Ahel provides instructions and does not run this skill.
No other account needed.
Add Ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.
What this skill tells your AI
The instructions your AI receives, as published by hashgraph-online/awesome-codex-plugins in plugins/kevin592/codex-full-stack-workflow/skills/implementation-planner/SKILL.md and read by Ahel’s review.
This skill creates a task-by-task implementation plan that can be executed with TDD and reviewed after each task.
It is also the bridge from full-stack product contracts into Superpowers execution discipline. Every implementation plan must declare its lifecycle intensity and include a concrete handoff to the execution skills that will carry the work.
Input Artifacts
00-stage.json00-artifact-registry.json- Product, IA, interaction, state, backend/API/frontend, visual, and acceptance contracts
- Visual confirmation when UI applies
Output Artifact
docs/full-stack-development/requirements/<feature-slug>/05-implementation/implementation-plan.md
Required MCP Tool
review_stage_gatereview_artifact_contractreview_visual_evidencewhen UI appliesgenerate_frontend_verification_checklistwhen UI appliesselect_development_flow_profileplan_superpowers_execution_handoff
Hard Stops
- Do not plan implementation when requirements are unapproved.
- Do not plan UI implementation when Pencil/Figma evidence is missing, incomplete, stale, or not reviewed.
- Do not create tasks without test commands and expected failure reasons.
- Do not create an implementation plan without one flow profile:
strict-fullstack,strict-ui,light-change, ordebug-fix. - Do not begin implementation until
plan_superpowers_execution_handoffhas attached TDD, subagent-driven or executing-plans, requesting-code-review, verification-before-completion, and finishing-a-development-branch. - Do not let
light-changebypass change records or completion evidence when requirements or code changed. - Do not let
debug-fixskip root-cause evidence or a failing regression test before the fix.
Pressure Test Responsibility
- Plans without failing tests block.
- Plans without exact files block.
- Plans without verification commands block.
- Plans without a flow profile block.
- Plans without Superpowers execution handoff block.
Flow Profiles
| Profile | Use When | Required Emphasis |
|---|---|---|
strict-fullstack | Product, UI, backend, API, data, auth, or acceptance contracts may change. | Full requirements workspace, visual evidence when UI applies, backend/API audit, code review, completion gate. |
strict-ui | React Web UI, HeroUI, visual design, responsive state, or interaction behavior changes without backend/API changes. | HeroUI official docs, Pencil/Figma evidence, visual inspection metrics, component graph audit. |
light-change | Small docs, copy, config, or narrowly bounded code changes that do not alter product scope, IA, UI, backend, API, data, auth, or acceptance behavior. | Change record when requirements changed, targeted verification, completion gate. |
debug-fix | Bug, failed test, broken build, regression, or unexpected behavior. | Root-cause investigation, failing regression test, targeted fix, fresh verification. |
Superpowers Execution Handoff
After the implementation plan passes this skill, attach this handoff:
superpowers:test-driven-development: production code requires a previously observed failing test.superpowers:subagent-driven-development: default for independent tasks; use one fresh implementer and one reviewer per task.superpowers:executing-plans: fallback only when subagents are unavailable or tasks are tightly coupled.superpowers:requesting-code-review: required after each substantial task and before final completion.superpowers:verification-before-completion: fresh evidence before any completion claim.superpowers:finishing-a-development-branch: structured branch finish after tests pass.
Required execution artifacts: task brief, task report, review package, .superpowers/sdd/progress.md, final code review report, and final acceptance report.
Next Gate
implementation-in-progressmay start only after the implementation plan artifact passes review, a flow profile is selected, and the Superpowers execution handoff is attached.
Signals
- GitHub stars
- 1k
- Forks
- 316
- Last commit
- Oct 2026
Advanced
- Item type
- skill
- Key
implementation-planner- Source
- github.com/hashgraph-online/awesome-codex-plugins
github.com/hashgraph-online/awesome-codex-plugins