SDD Observatory

Superpowers established

Skill-based agent discipline with hard gates: a mandatory brainstorm → spec → plan pipeline with human approval before any code, then per-task subagent execution under strict test-driven development.

Reviewed last month · Jul 18, 2026

Website GitHub · obra/superpowers Compare with other frameworks →

Stars
278.3K
Contributors
44
Open issues
322
Age
11 months
Last push
last week
Latest release
v6.3.0 Aug 12, 2026
Week of Aug 31, 2025: 0 commits Week of Sep 7, 2025: 0 commits Week of Sep 14, 2025: 0 commits Week of Sep 21, 2025: 0 commits Week of Sep 28, 2025: 0 commits Week of Oct 5, 2025: 57 commits Week of Oct 12, 2025: 35 commits Week of Oct 19, 2025: 6 commits Week of Oct 26, 2025: 14 commits Week of Nov 2, 2025: 1 commit Week of Nov 9, 2025: 0 commits Week of Nov 16, 2025: 33 commits Week of Nov 23, 2025: 45 commits Week of Nov 30, 2025: 5 commits Week of Dec 7, 2025: 20 commits Week of Dec 14, 2025: 3 commits Week of Dec 21, 2025: 9 commits Week of Dec 28, 2025: 0 commits Week of Jan 4, 2026: 0 commits Week of Jan 11, 2026: 2 commits Week of Jan 18, 2026: 14 commits Week of Jan 25, 2026: 9 commits Week of Feb 1, 2026: 14 commits Week of Feb 8, 2026: 8 commits Week of Feb 15, 2026: 4 commits Week of Feb 22, 2026: 0 commits Week of Mar 1, 2026: 19 commits Week of Mar 8, 2026: 40 commits Week of Mar 15, 2026: 30 commits Week of Mar 22, 2026: 23 commits Week of Mar 29, 2026: 5 commits Week of Apr 5, 2026: 2 commits Week of Apr 12, 2026: 14 commits Week of Apr 19, 2026: 1 commit Week of Apr 26, 2026: 1 commit Week of May 3, 2026: 30 commits Week of May 10, 2026: 12 commits Week of May 17, 2026: 9 commits Week of May 24, 2026: 5 commits Week of May 31, 2026: 15 commits Week of Jun 7, 2026: 81 commits Week of Jun 14, 2026: 17 commits Week of Jun 21, 2026: 7 commits Week of Jun 28, 2026: 12 commits Week of Jul 5, 2026: 35 commits Week of Jul 12, 2026: 12 commits Week of Jul 19, 2026: 4 commits Week of Jul 26, 2026: 1 commit Week of Aug 2, 2026: 0 commits Week of Aug 9, 2026: 1 commit Week of Aug 16, 2026: 0 commits Week of Aug 23, 2026: 0 commits
Commits per week, last 12 months · 655 total
View weekly commit data
Week startingCommits
Aug 23, 20260
Aug 16, 20260
Aug 9, 20261
Aug 2, 20260
Jul 26, 20261
Jul 19, 20264
Jul 12, 202612
Jul 5, 202635
Jun 28, 202612
Jun 21, 20267
Jun 14, 202617
Jun 7, 202681
May 31, 202615
May 24, 20265
May 17, 20269
May 10, 202612
May 3, 202630
Apr 26, 20261
Apr 19, 20261
Apr 12, 202614
Apr 5, 20262
Mar 29, 20265
Mar 22, 202623
Mar 15, 202630
Mar 8, 202640
Mar 1, 202619
Feb 22, 20260
Feb 15, 20264
Feb 8, 20268
Feb 1, 202614
Jan 25, 20269
Jan 18, 202614
Jan 11, 20262
Jan 4, 20260
Dec 28, 20250
Dec 21, 20259
Dec 14, 20253
Dec 7, 202520
Nov 30, 20255
Nov 23, 202545
Nov 16, 202533
Nov 9, 20250
Nov 2, 20251
Oct 26, 202514
Oct 19, 20256
Oct 12, 202535
Oct 5, 202557
Sep 28, 20250
Sep 21, 20250
Sep 14, 20250
Sep 7, 20250
Aug 31, 20250

Metrics updated Aug 27, 2026 · collected automatically from the GitHub API

Core approach

Fourteen composable Markdown skills, bootstrapped by a session-start hook that makes their use mandatory ("IF A SKILL APPLIES TO YOUR TASK, YOU DO NOT HAVE A CHOICE"). Creative work must pass through brainstorming, which writes a design doc to docs/superpowers/specs/ and hard-gates implementation until the user approves it; writing-plans then produces a plan of 2–5-minute tasks with exact file paths. Execution dispatches a fresh subagent per task, each diff judged first for spec compliance and then for code quality, under the TDD "Iron Law": no production code without a failing test first.

Workflow

  1. superpowers:brainstorming — one-question-at-a-time requirements refinement; writes a design doc to docs/superpowers/specs/ and blocks implementation until the user approves
  2. superpowers:using-git-worktrees — isolated workspace on a new branch with a clean test baseline
  3. superpowers:writing-plans — plan in docs/superpowers/plans/ with 2–5-minute tasks, exact file paths, and a no-placeholders rule
  4. superpowers:subagent-driven-development (or superpowers:executing-plans) — fresh subagent per task; each diff must pass a spec-compliance verdict then a code-quality verdict, tracked in a progress.md ledger under .superpowers/sdd/
  5. superpowers:test-driven-development — the Iron Law: no production code without a failing test first; untested code gets deleted and reimplemented
  6. superpowers:requesting-code-review / superpowers:receiving-code-review — review against the plan, issues by severity
  7. superpowers:finishing-a-development-branch — hard stop unless tests pass, then merge, push for PR, keep, or discard

Supported tools

Strengths

  • Enforcement, not suggestion — hard gates at every phase, from design approval before code to per-task review verdicts and a tests-must-pass stop before merging
  • Covers the whole loop, not just planning: systematic-debugging, verification-before-completion, code review, worktree hygiene, and branch finishing
  • Context-rot resistant by design — fresh subagents per task plus on-disk specs, plans, and a progress.md ledger mean state survives compaction and session resets
  • The largest adoption in the category, distribution through Anthropic's official plugin marketplace, and vendor-neutral skills across ten agent harnesses
  • Unusually empirical maintenance — a dedicated eval harness, with release notes documenting features removed when evals showed cost without quality gain

Limitations

  • Token and time overhead is the dominant user complaint; the project's own v6 releases are partly devoted to cutting bootstrap and review costs
  • The mandatory ceremony (brainstorm interview, TDD on every change, the delete-untested-code rule) is a poor fit for trivial fixes and exploratory or legacy-heavy work
  • Enforcement is prompt-level, not tool-level — the gates live in Markdown instructions, with no CI check or traceability tooling behind the artifacts
  • Specs are date-stamped point-in-time design docs with no update or delta mechanism — third-party hybrid projects exist specifically to graft spec lifecycle management onto it
  • High churn: six major versions in its first nine months, with slash commands added then removed and artifact directories relocated twice

Tracked projects using Superpowers

Notes

Superpowers is this observatory’s borderline case, included deliberately. It never calls itself spec-driven — its self-description is “an agentic skills framework & software development methodology” — but it mandates persistent, human-approved spec and plan artifacts before any code and judges every task’s diff against them, which meets a minimal definition of SDD with unusually hard gates. What it lacks is the spec lifecycle the other tracked frameworks center on: no requirement IDs, no deltas, no traceability — just date-stamped design docs — which is why a small ecosystem of third-party hybrids exists to bolt OpenSpec-style governance onto it. It is also the category’s biggest phenomenon: launched by Jesse Vincent (Request Tracker, Keyboardio) the day Claude Code shipped plugins, it grew into one of GitHub’s most-starred repositories within nine months. Its signature idea is recursion — a writing-skills skill that teaches the agent to author new skills — and the dogfooding is radical: Vincent reports having hand-written no code since October 2025, and the repo’s own plans are checked in under docs/plans/.

Added Jul 18, 2026 · Assessment last reviewed Jul 18, 2026 · How we track