People ask how this relates to the two other collections they hear about most - Superpowers and Matt Pocock’s skills. All three are good, share DNA, and are worth learning from. This is an honest map of how they’re shaped differently, so you can pick what fits - or borrow across all three.
They optimize for different moments. agent-skills organizes the whole lifecycle with review personas, anti-rationalization guards, and an in-repo eval framework. Superpowers leans into autonomous, reasoning-heavy runs. Matt Pocock’s skills are a sharp, opinionated daily toolkit with a signature “grill me” loop.
| agent-skills | Superpowers | Matt Pocock’s skills | |
|---|---|---|---|
| Core idea | Encode the full senior-engineering lifecycle as skills | A complete development methodology built on composable skills | One expert’s Claude Code workflow, open-sourced |
| Organizing principle | SDLC phases (Define → Ship) behind a meta-skill router | A single disciplined loop: brainstorm, plan, execute, review | A curated toolbox of focused, composable commands |
| Catalog size | 24 skills across the whole lifecycle | ~14 skills, deep on the inner build loop | ~30 skills, grouped by area |
| Lifecycle coverage | Broad: idea → API/UI → security → perf → CI/CD → observability → launch | Deep but narrow: TDD, debugging, planning, review, skill authoring | Define & Build heavy: grilling, PRDs, issues, TDD, triage |
| Distinctive mechanisms | Anti-rationalization tables + Red Flags in every skill; parallel review personas; reference checklists; a three-tier eval framework in CI | Subagent-driven development with a task reviewer and fix loop; git-worktree isolation; skills-that-write-skills | The grilling primitive; seam-based TDD; explicit user- vs model-invoked split |
| Quality measurement | Trigger, routing & behavioral evals against the catalog (some in CI) | Pressure-testing is core; the eval suite lives in a separate repo | None shipped in-repo |
| Best for | Driving a feature through every phase with a checkpoint at each | Long, autonomous, reasoning-heavy or exploratory work | A pragmatic, battle-tested daily loop - strongest at requirements & TDD |
We deliberately leave out star counts and adoption figures - they’re cited wildly differently and change weekly. All three are actively used and maintained.
A full development methodology that bets on autonomy and upfront reasoning. A session runs a deliberately linear pipeline - Socratic brainstorming, executable plans, then subagent-driven execution with a task reviewer signing off before each task closes. Git worktrees isolate parallel work. Strong when you hand off a sizable chunk and want a reviewed result back; heavier on small changes.
View repoThe Claude Code skills one very good engineer actually uses, developed in public with visible in-progress/ and deprecated/ folders. The center of gravity is grilling - a reusable one-question-at-a-time interrogation loop. Authentic and sharp; Claude-Code-first in practice, with thinner coverage past Build and no in-repo evals.
View repoOm Mishra ran a controlled experiment - same model, same repo, same prompt in Claude Code, changing only the skill framework. agent-skills moved to code faster (~8 min vs ~12) and ran more validation passes (7 vs 5, including the full test suite), which caught a compatibility issue outside the immediate feature. Superpowers invested more upfront architectural reasoning. Token efficiency was effectively identical. His honest conclusion: pick the tool to the task.
Read the write-upThe complete write-up covers how to decide in practice, phase by phase.