SimplePower is designed to get your work done a lot faster even without /fast.
By using SimplePower, you accept a faster AI-Human turn around time, and you are expected to put more effort to guide the AI.
SimplePower is a Codex-only fork of Superpowers by Jesse Vincent / Prime Radiant. Thanks to Jesse Vincent / Prime Radiant for the upstream project this fork is based on. Huge thanks and respect for them, SimplePower is not trying to replace SuperPower, it is here just as an alternative. Right now I am only maintaining a codex version as I wanna keep this project focus for start.
This table explains what SimplePower is trying to achieve (times are just estimate from my experience):
| Goal | Superpowers-style path | SimplePower path |
|---|---|---|
| AI Time | 3x | 0.3x |
| Turnaround Time | Slow | Fast |
| Human Input | Less | More |
| Tokens | 3x | 2x |
| Result Accuracy and Quality | 95% | 90% |
| Pharse | SuperPower | SimplePower |
|---|---|---|
| Spec / Plan | brainstorming -> approve spec -> spec.md (commit) -> plan.md (approve and commit) |
brainstorming -> approve spec -> plan.md (approve and commit) too lazy to check spec.md and plan.md |
| Subagent Implementation this is why SimplePower is fast |
Task1 impl agent -> Task1 planning check -> Task1 quality agent -> Task2 impl agent -> Task2 planning check -> Task2 quality agent -> .... |
Many subagents in parallel for multiple files -> Quick tests runner subagent (spark) -> Single final reviewer + fixer |
| Git Commits? | every steps | all at once after parallel subagent + final after review |
Install Simple Power from the Codex plugin marketplace:
codex plugin marketplace add garyfpga/codex-plugins
codex plugin marketplace upgradeUse codex plugin marketplace upgrade again whenever you want to pull
marketplace updates.
Simple Power uses two configurable model tiers:
SIMPLEPOWER_BEST_MODEL="gpt-5.5-high"
SIMPLEPOWER_FAST_MODEL="gpt-5.4-mini-high"Parse each value as <model>-<reasoning_effort> by taking the final
dash-delimited segment as reasoning_effort and the preceding string as
model. For example, gpt-5.4-mini-high resolves to
model="gpt-5.4-mini" and reasoning_effort="high".
Simple Power skills use the simplepower:* namespace. Mention a skill by name,
such as simplepower:brainstorming, when you want Codex to use it.
The brainstorming skill can use a temporary localhost visual companion for
mockups, diagrams, and other visual questions. Generated implementation plans
are saved under docs/simplepower/plans/.
After simplepower:writing-plans saves and reviews a plan, it asks which
implementation handoff to use. The recommendation comes from current Codex context usage
when available: below 55% continues in the current session, and
55% or higher recommends a fresh session. If context measurement is
unavailable, Simple Power falls back to saved plan size. The handoff still shows
both commands; for fresh context, run /clear first.
Use simplepower:brainstorming and start your plan.
simplepower:systematic-debugging also follows the Simple Power flow.
MIT License. See LICENSE for details.