Drawing the Ownership Line

Superpowers v2: own integration, borrow methodology

The Frame

Amplifier wrapped a proven methodology so its discipline travels with the platform

Superpowers is Jesse Vincent's (obra) disciplined-development methodology: TDD's "Iron Law: NO CODE WITHOUT FAILING TEST FIRST," systematic debugging, plan-first work.

Amplifier's bundle turned that methodology into Amplifier-native agents. Did the wrapping actually ship?

Proof — It Shipped

v1 shipped: a git submodule import plus 9 thin wrapper agents

The primary repo imports obra/superpowers via a git submodule (url = github.com/obra/superpowers.git) and exposes 9 wrapper agents. Each references exactly one distinct upstream skill — 9 skills in all.

The import is real and committed. But which version did it freeze?

9
wrapper agents, one per upstream skill
9
distinct upstream skills referenced
31
commits; default branch master
The Tension — Drift

The submodule is frozen at v3.6.2 while upstream raced to v6.1.1

v1 pinned obra/superpowers to commit 154d664 = tag v3.6.2, dated 2025-12-03. Upstream's latest tag is v6.1.1, dated 2026-07-02 — three major versions of drift.

That's the first structural weakness. It isn't the only one.

Pinned in v1
v3.6.2
2025-12-03
Upstream latest
v6.1.1
2026-07-02
The Tension — Scatter

Enforcement wasn't centralized: the "Iron Law" text lives in 4 separate files

The TDD "Iron Law: NO CODE WITHOUT FAILING TEST FIRST" text is repeated across 4 markdown files rather than centralized — duplication the refresh targets.

Drift plus scatter. So what's the fix that resolves both?

The Mechanism

The fix is one clean split: knowledge stays a skill, workflows become recipes

The v2 staging analysis examined 14 upstream skills. It keeps only the 3 knowledge/reference ones as skills and maps 6 multi-step workflows to Amplifier recipes.

Apply that split and the whole bundle collapses to something thin.

14
upstream skills analyzed in v2 staging
3
kept as knowledge "True Skills"
6
workflows mapped to recipes
The Turn

The split collapses to a thin bundle: 3 skills, 7 recipes, one expert agent

The v2 staging area (Inactive/superpowers-2) keeps just 3 copied knowledge-skills — test-driven-development, systematic-debugging, verification-before-completion — plus 7 recipes over a behavior and one superpowers-expert agent.

That thin shape is the ownership line made concrete.

3
copied knowledge-skills retained
7
recipes for multi-step workflows
1
expert agent over one behavior
The Payoff

Amplifier owns integration and enforcement — methodology flows live from upstream

The README states the design intent: a single source of truth in obra/superpowers, with git submodule update --remote so improvements flow through automatically — no frozen pin to chase.

The Takeaway

The direction is set: borrow the wisdom, own only the integration

v2 is a staged design, not yet shipped on the primary repo's master branch. But the pattern generalizes: pull methodology live from its source, and keep your own layer thin — integration and enforcement.

Own the seam, not the whole library.

Status: v2 staging — not yet shipped
Sources

Research Methodology

Data as of: July 21, 2026

Feature status: v1 SHIPPED on ramparte/amplifier-bundle-superpowers (master); v2 STAGING / not yet shipped (Inactive/superpowers-2, not a functioning git repo; no v2 branch/PR/issue).

Research performed:

Gaps: "Pull methodology live" is a stated design goal (README + v2 bundle design), not an observed running behavior — the submodule remains pinned to v3.6.2. The "7→2" skill figure was not verifiable; grounded counts are 9 skills referenced in v1 and 3 copied skills in v2 staging. v2 commit dates/authors could not be established (staging dir is not a functioning git repo).

Contributors: Jesse Vincent (@obra) — upstream methodology author; Sam Schillace — author of the primary repo's current HEAD commit (073253a); ramparte — owner of the Amplifier integration repo (Adapter: Ramparte).

More Amplifier Stories