In the agent-skills ecosystem, two names come up constantly:
- grill-me — Matt Pocock (mattpocock/skills)
- Superpowers — Jesse Vincent / obra (obra/superpowers)
People often treat them as an either/or. A better frame: they overlap on pre-implementation alignment, but differ sharply in shape, enforcement, and lifecycle coverage.
This post compares positioning, mechanisms, workflows, trade-offs, and practical selection. Structure follows public repos as of mid-2026 (star counts move; check GitHub for live numbers).
One-line positioning #
| grill-me | Superpowers | |
|---|---|---|
| What it is | A (family of) relentless interview skill(s) | A full software methodology + composable skills |
| Core question | “Have you actually thought this through?” | “How do we ship from idea to a mergeable branch with discipline?” |
| Output | Shared understanding (optionally no files) | Design → plan → implement → review → finish |
| Philosophy | Small, hackable, composable; human keeps control | Process-first; skills force agents not to cut corners |
| Author | Matt Pocock | Jesse Vincent (obra) |
| Repo | mattpocock/skills | obra/superpowers |
Where each came from #
grill-me: three sentences that force alignment #
grill-me went viral as a tiny skill. The essence:
Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk each branch of the design/decision tree, resolving dependencies one by one. If the codebase can answer a question, explore the code instead of asking. For each question, give your recommended answer.
It later split into a clearer stack:
| Entry | Role |
|---|---|
| grilling | Interview primitive (callable by other skills) |
| grill-me | User-facing front door: pure interview, writes nothing |
| grill-with-docs | Same interview + writes CONTEXT.md glossary / ADRs |
The broader mattpocock/skills repo markets Skills for Real Engineers: many small composable skills (domain-modeling, to-spec, to-tickets, tdd, implement…). It argues against frameworks that own the process and take your control (README contrasts approaches like GSD / BMAD / Spec-Kit).
Install sketch:
npx skills@latest add mattpocock/skills --skill=grill-me -y -g
# or install the set, then run /setup-matt-pocock-skillsSuperpowers: weld engineering process into the agent #
Superpowers describes itself as:
A complete software development methodology for your coding agents.
It is not “one skill.” It is:
- A library of composable skills (brainstorming, writing-plans, TDD, subagent-driven-development, …)
- A session bootstrap via using-superpowers: check and invoke relevant skills before acting
- Multi-harness packaging (Claude Code, Codex, Cursor, OpenCode, Copilot CLI, Pi, …)
The enforcement tone from using-superpowers:
If you think there is even a 1% chance a skill might apply… you ABSOLUTELY MUST invoke the skill.
In short: Superpowers fights the failure mode where agents rationalize skipping TDD, planning, or review under time pressure.
Claude Code install sketch:
/plugin install superpowers@claude-plugins-officialWorkflow comparison #
grill-me covers the alignment entrance #
Idea / rough plan
│
▼
/grill-me or /grill-with-docs or /grilling
│
├─ one question at a time
├─ recommended answer each time
├─ explore code when facts live there
└─ resolve decision-tree deps in order
│
▼
Shared understanding (+ optional CONTEXT.md / ADRs)
│
▼
(optional) to-spec → to-tickets → implement → …Notes:
- Does not implement by default
- grill-me is stateless: alignment lives in the chat; durable artifacts need grill-with-docs
- Matt’s main build chain is roughly:
grill-with-docs → to-spec → to-tickets → implement → code-review - Huge / foggy efforts may start with wayfinder, then return to grilling
Superpowers covers the full delivery pipeline #
You: “Let’s build X”
│
▼
using-superpowers (must check skills first)
│
▼
brainstorming ← HARD-GATE: no code until design approved
│ context → one-by-one Qs → 2–3 options → design sections → design doc
▼
using-git-worktrees
▼
writing-plans (2–5 min tasks with verification)
▼
subagent-driven-development / executing-plans
│ + test-driven-development (red-green-refactor)
│ + requesting-code-review between tasks
▼
finishing-a-development-branch (tests, merge/PR/cleanup)Notes:
- Auto-triggers by design; you often don’t type
/brainstorm - Brainstorming hard-gates implementation skills until approval
- TDD, YAGNI, evidence-before-claims are first-class
- Subagents + worktrees support long unattended runs
Mechanism differences #
1. Invocation: you opt in vs it enforces #
| Dimension | grill-me | Superpowers |
|---|---|---|
| Default trigger | User slash /grill-me (disable-model-invocation: true) | Session-level mandatory skill checks |
| Skip risk | You can simply not call it | Designed so the model can barely skip |
| Fits | Keep your own pace; pressure on demand | Default engineering behavior |
This is the deepest product split:
- grill-me: optional scalpel
- Superpowers: always-on seatbelt + full factory line
2. The interview: similar, not identical #
Both attack “code before clarity”:
| Shared | Detail |
|---|---|
| One question at a time | Avoid questionnaire dumps |
| Understand before building | Stop drive-by implementation |
| Prefer code exploration | Facts from the repo; decisions from you |
| Guided options | Superpowers: 2–3 approaches; grill-me: recommended answers |
Differences:
| grill-me / grilling | Superpowers brainstorming | |
|---|---|---|
| Metaphor | Decision-tree interrogation | Collaborative design + hard gates |
| Artifacts | None by default; with-docs → glossary/ADR | Design doc path + commit habit |
| Visual | None built-in | Optional visual companion (browser mockups) |
| Downstream | Hands control back to you | Chains into writing-plans |
In one line: brainstorming ≈ heavier, productized grilling hooked to a delivery pipeline.
3. Coverage: point vs surface #
| Lifecycle stage | grill-me family | Superpowers |
|---|---|---|
| Clarification / design | ✅ home turf | ✅ brainstorming |
| Ubiquitous language / ADR | ✅ grill-with-docs / domain-modeling | Design docs; weaker DDD framing |
| Implementation plan | Other skills (to-spec/tickets) | ✅ writing-plans |
| Isolated workspace | Not built-in | ✅ git worktrees |
| Execution | implement / tdd, etc. | ✅ SDD + TDD |
| Code review | code-review skill exists | ✅ inter-task review path |
| Branch finish | Loose | ✅ finishing-a-development-branch |
| Debugging method | diagnosing-bugs, etc. | ✅ systematic-debugging |
| Multi-harness | Agent Skills / installer | First-class multi-platform plugins |
4. Artifacts and “memory” #
- grill-me: conversation is the artifact; cross-session reuse → grill-with-docs (
CONTEXT.md+ ADRs) - Superpowers: defaults to specs under
docs/superpowers/specs/..., plans and worktree lifecycle follow the flow
If you care about project-level language, Matt’s CONTEXT.md story is sharper. If you care about feature delivery discipline, Superpowers is thicker.
5. Weight and learning curve #
| grill-me | Superpowers | |
|---|---|---|
| Cognitive load | Tiny: one slash command | Medium–high: methodology + skill graph |
| Customization | Easy to fork small skills | Cross-agent consistency; higher bar to change |
| Trivial tasks | Easy to skip | Can feel heavy (even fans admit trivial work may not need it) |
Pros and cons #
grill-me strengths #
- Extremely light — install and go
- Clear control — you choose when to pressure-test
- Universal — coding, courses, product decisions
- Recommended answers — turn open interrogation into fast yes/no
- DDD-friendly chain — grill-with-docs / domain-modeling turn alignment into assets
grill-me weaknesses #
- Stops at alignment — no implement/finish pipeline
- No enforcement if you forget to run it
- One skill is not enough for real delivery — you need the rest of the set
- Pure grill-me is ephemeral without with-docs
Superpowers strengths #
- End-to-end methodology — brainstorm to merge options
- Forces skill use — fights “skip TDD under pressure”
- Subagents + worktrees + two-stage review — long tasks
- Official multi-harness support — team tool-churn is cheaper
- TDD / verification defaults — higher quality baseline
Superpowers weaknesses #
- Heavy for small edits
- Less agency for people who hate being process-managed
- Stacks badly with other heavy methodologies if you install everything
- Debugging “why did it enter brainstorming again?” has a cost
Community color (anecdotal) #
- Reddit/blog trifectas often list Superpowers / get-shit-done / grill-me
- Some call Superpowers more “elegant and complete,” grill-me a small “decision-driven” tip of the spear
- Others say Superpowers feels like endless questioning — similar to grill-me — but with a full engineering pipe behind it
- Matt’s own coding recommendation evolved: early viral grill-me; later prefer grill-with-docs / domain-model chain when aligning against a real codebase
How to choose #
Pick grill-me (or grilling / grill-with-docs) if… #
- Your main pain is coding before alignment
- You want manual cadence, not a default full process
- You grill non-code decisions often
- You like small skills and will compose
to-spec → tickets → implementyourself - You value CONTEXT.md / ADRs as cross-session assets
Pick Superpowers if… #
- You want long autonomous runs that stay on rails
- The team needs shared engineering defaults (TDD, plans, review)
- You work across Claude Code / Codex / Cursor
- You’ll trade process weight for fewer rewrites
- You hate “simple task → half-baked patch”
The practical answer: combine, don’t oppose #
They overlap on the entrance interview and complement on delivery:
A (light): /grill-me → implement yourself or plan mode
B (mid): /grill-with-docs → to-spec → implement
C (heavy): Superpowers full flow
D (hybrid): grill first for high-stakes design,
then Superpowers plan/execute
(or keep Superpowers TDD/worktree only)Practical tips:
- Install only one side for two weeks and notice whether enforcement helps or irritates
- Already on Superpowers? Still use grill-me for extra pressure tests on risky designs
- Already on Matt’s full set? Don’t stack Superpowers just for stars unless you need worktree/SDD/multi-harness
- Trivial edits: skip both. High-risk architecture: use some form of interview either way
Summary table #
| Dimension | grill-me | Superpowers |
|---|---|---|
| Grain | Point skill (+ primitive/variants) | Methodology + skill library |
| Trigger | User slash primarily | Default auto + hard gates |
| Main value | Alignment, decision trees, shared understanding | Full-process discipline and delivery |
| Artifacts | Optional (with-docs) | Default design/plan paths |
| TDD / subagents / worktrees | Other skills or manual | Built into the chain |
| Multi-platform | Agent Skills / installer | Official multi-harness plugins |
| Best for | Pressure interviews, general decisions | Serious feature delivery |
| Risk | Unused = useless | Over-process |
| Relationship | Entrance scalpel | Factory OS |
Quick install #
grill-me / Matt skills:
npx skills@latest add mattpocock/skills --skill=grill-me -y -g
# preferred for coding:
npx skills@latest add mattpocock/skills --skill=grill-with-docs -y -gClaude plugin:
/plugin marketplace add mattpocock/skills
/plugin install mattpocock-skills@mattpocockSuperpowers:
/plugin install superpowers@claude-plugins-officialSee the Superpowers README for Cursor / Codex / OpenCode and others.
Conclusion #
- grill-me asks: Before we write code, do we share the same understanding?
- Superpowers asks: After we understand, how does the agent ship with engineering discipline?
Treating them as rival products undersells grill-me’s universality and Superpowers’ pipeline depth. A healthier model:
grill-me is the pressure-test primitive; Superpowers is the default engineering OS.
One slot only: alignment pain → grill-me (or grill-with-docs); delivery discipline pain → Superpowers. Two slots: interview and domain language from Matt’s stack, plan/TDD/worktree from Superpowers — many strong workflows already mix them without writing it down.
