Why You Can Ditch Most Superpowers Skills in the GPT‑5.6 Era

Many basic steps no longer need to be taught

Older Skills were often long checklists—how to read a project, find call chains, run tests, and verify before a PR. With models like GPT‑5.6, K3, Claude Fable5, the agent can read the project, locate bugs, add tests, and run verification automatically, so these explicit steps are unnecessary.

For longer tasks, approvals, or multi‑agent collaboration, Codex can still use Skills, Plugins, MCP, and Hooks, but the earlier problem of “incomplete step lists causing chaos” has largely disappeared.

Below I use Codex as a coding agent example; the same reasoning applies to other agents.

Skills compete for work

Codex does not read the entire SKILL.md at the start. It first loads each Skill’s name, description, and path, then loads the full content only when the task matches—a mechanism called progressive disclosure.

According to Codex’s Skills documentation[1], the Skill list consumes at most 2 % of the model’s context window (up to 8 000 characters). When the list exceeds this budget, descriptions are shortened and some Skills are omitted from the initial list.

When 100 Skills are installed, the agent may see only truncated descriptions, causing rule conflicts and making it hard to spot which Skill is causing a mis‑run.

Skills are similar to browser bookmarks: you keep many at first, but after months only a few remain truly useful.

Separate rules from reusable tasks

Project‑wide conventions that apply to every task belong in AGENTS.md or a project rules file.

Procedures needed only for specific tasks should stay as Skills.

Long‑running investigations that generate lots of intermediate data belong to a Sub‑agent.

Reusable scripts, templates, or personal preferences should be packaged as a Plugin.

Mechanical constraints that must never be missed belong to Hooks, CI, linters, or tests—not to natural‑language reminders.

Mixing rule files and Skills creates a bloated SKILL.md that is hard to maintain.

Why I rarely use Superpowers

Large, all‑in‑one Skill suites like Superpowers[3] tend to over‑complicate workflows. In group discussions, teammates noted that Superpowers’ “grilling” mode asks dozens of questions, which clarifies requirements but adds overhead.

The full software‑development process (brainstorm → plan → design → execute → review → verify) is still valuable for complex, high‑risk changes, but for small tweaks the process overhead can dominate execution time.

Strong models consume many tokens; unnecessary Skills waste context and can introduce security risks if third‑party Skills contain dangerous commands. Always review SKILL.md, scripts/, and references/ before installation.

Which Skills still deserve keeping in the strong‑model era

I now keep three categories of Skills:

Personal preferences or fixed deliverables that the model cannot infer (e.g., article style, chart standards, internal templates, release procedures).

Tasks that require professional judgment, scripts, or reference material (e.g., security reviews, complex document handling, framework migrations, production checks).

Skills that prevent directional errors by asking clarifying questions before execution, such as the grilling Skill[5].

In my open‑source project “SpringAI Intelligent Interview Platform” (≈2.8 k Stars), I used grilling to decide whether to integrate a knowledge base or keep the existing interview flow, and to choose single‑library versus multi‑library setups.

My Skill pruning standards

Before adding a Skill, I ask:

Will the model already perform this task?

Do I repeatedly fail at this point without the Skill?

Does the Skill encapsulate scripts, templates, or personal preferences?

Can I detect and remove the Skill when it becomes obsolete?

Skills that merely repeat “read project → write code → run tests” are now omitted; such logic belongs in AGENTS.md. Only Skills that address recurring, high‑cost failures are retained, and they should stay small, easy to modify, and composable.

References

[1] Codex Skills documentation: https://learn.chatgpt.com/docs/build-skills

[2] AIGuide – AI Application Development, AI Programming Practice & Interview Guide: https://mp.weixin.qq.com/s/le3RzJsaAH22auUoB05y1Q

[3] Superpowers repository: https://github.com/obra/superpowers

[4] mattpocock/skills repository: https://github.com/mattpocock/skills

[5] grilling Skill definition: https://github.com/mattpocock/skills/blob/main/skills/productivity/grilling/SKILL.md

[6] Codex Best Practices – Configuration, AGENTS.md & Agentic Workflow: https://javaguide.cn/ai-coding/practices/codex-best-practices.html

[7] AI Programming Practice Guide: https://javaguide.cn/ai-coding/