Are some skills deliberalty not used ? — Skill Arborescence Review & Feedback #642
Description
First off, love the plugin — it's been a game changer for my workflow. I'm a young engineer working on a Rails migration and the structured brainstorming → planning → execution pipeline has genuinely improved how I build features. Thank you for that.
That said, I spent some time mapping the full skill arborescence and I spotted a few things I wanted to flag. Not sure if they're intentional design choices or gaps so i want to point that i might be mistaken over this subject.
- executing-plans is thinner than subagent-driven-development
When writing-plans offers the two execution paths, subagent-driven-development explicitly integrates test-driven-development (subagents should use it) and requesting-code-review (two-stage review after each task). executing-plans references neither. TDD is partially covered because the plan document itself contains TDD steps, but the skill's guardrails aren't active. Code review is the bigger miss — subagent-driven gets automated review after every task, executing-plans just says "Ready for feedback" and relies on the human. Would it make sense to wire requesting-code-review into executing-plans between batches, and explicitly reference test-driven-development?
- dispatching-parallel-agents is orphaned
No skill in the entire chain references it. Not brainstorming, not writing-plans, not subagent-driven, not executing-plans. It exists as a standalone utility for debugging multiple independent failures. But it could be useful inside executing-plans or subagent-driven when plan tasks are independent of each other. Right now executing-plans says "execute batch of 3 sequentially" with no decision point about parallelisation. Is the sequential-only approach intentional?
- verification-before-completion is orphaned
Same situation — no skill invokes it. The skill itself is emphatic about never claiming completion without fresh verification evidence, but finishing-a-development-branch (the natural integration point) doesn't reference it. finishing-a-development-branch does run tests in Step 1, which provides partial coverage, but not the full rigour of verification-before-completion. Should finishing-a-development-branch list it as a required sub-skill?
- using-superpowers can't compensate for these gaps
I initially thought using-superpowers ("if even 1% chance a skill applies, invoke it") would catch these. But it's a soft nudge to Claude, not structural wiring. It can't force dispatching-parallel-agents into executing-plans if executing-plans explicitly says "sequential batches." It can't force verification-before-completion if no skill creates a trigger point for it.
- executing-plans bypasses the subagent-driven choice
writing-plans generates plan headers with "REQUIRED SUB-SKILL: Use superpowers:executing-plans". If you open a fresh session with that plan, it goes straight into batch execution with no opportunity to choose subagent-driven instead. The choice point only exists inside writing-plans. Should executing-plans have a small gate re-offering the option?
None of this takes away from how useful the plugin is. These feel like small wiring gaps in an otherwise excellent system. Happy to discuss or help test if you decide to change anything.
I ve linked a more detailed view of the subject in a file (made with claude of course).
I would like to emphasize that i miht be mistaken on the subject here so please feel free to tell me if so.