SkillVariants

Let your coding agent study how Agent Skills have been adapted across GitHub.

Deterministic GitHub evidence. Agent-powered analysis. Real implementations.

hundreds of GitHub matches
   ↓  collapse copies, group near-clones
mutation groups
   ↓  behavior-equivalence clustering with a strict guardrail
recurring adaptations
   ↓  representative implementations with exact source links

Captured counts below are from 2026-08-29; GitHub results change over time.

Use from your Agent

Install the bundled Agent Skill (copy skills/skillvariants/ into your agent's skills directory), then just ask:

Study how this Skill has been adapted across GitHub:
https://github.com/obra/superpowers/blob/main/skills/systematic-debugging/SKILL.md

The agent runs the deterministic study runtime, analyzes mutation groups, and answers with recurring adaptations, strict counts, and real source links — then compares any variant with the target on request. Details: docs/agent-skill.md.

Try the CLI

No SkillVariants installation is required when using uvx (the uv tool must be installed). GitHub Code Search authentication is required: GITHUB_TOKEN or gh auth login.

macOS / Linux:

uvx skillvariants related \
  https://github.com/obra/superpowers/blob/main/skills/systematic-debugging/SKILL.md

Windows PowerShell:

uvx skillvariants related `
  https://github.com/obra/superpowers/blob/main/skills/systematic-debugging/SKILL.md

Real output (2026-08-29 capture; full version in examples/systematic-debugging.txt):

SYSTEMATIC-DEBUGGING
Candidate matches found: 272

  Exact copies              0
  Unique related variants   175
  Detected mutation archetypes 4

COMPACT REWRITES
34 groups · 38 unique variants · 54 occurrences
─────────────────────────────────────────────
GuicedEE/ai-rules
  relatedness: 0.58
  length changed by -91%
  18 headings added/removed

ROUTING SPECIALIZATIONS
8 groups · 18 unique variants · 22 occurrences
─────────────────────────────────────────────
bg-szy/TOP-SKILLS
  relatedness: 0.82
  new routing-boundary language
  +6 cross-skill references

Why this exists

Agent Skills get copied between repositories constantly — and the copies are rarely identical. They get compressed into checklists, wrapped in thin redirects, rerouted to sibling skills, or specialized for one project. None of that is visible from GitHub search, and git diff can't help because you'd need to already know which two files to compare.

SkillVariants is not a registry ("what Skills can I install?") — it answers a different question: what happened to this Skill as different repositories adapted it?

Web explorer

A static explorer over the three validated studies (home → study → motif → compare) is in web/ — precomputed data, no backend, no auth. See docs/web-explorer.md.

Commands

skillvariants inspect  <url>            # frontmatter, body stats, signals
skillvariants related  <url> [--mode mutations|closest] [--json]
skillvariants evidence <url> --json     # agent-facing evidence payload
skillvariants compare  <url-a> <url-b>  # similarity + structural changes + diff
skillvariants study-*                   # persistent study runtime (for agents)

• related --mode mutations (default) shows the archetype map above.
• related --mode closest is pure textual-nearest order after exact-copy collapsing — deliberately no story logic.
• All commands support --json; stdout is clean JSON, warnings go to stderr.

How it works

Deterministic pipeline, fully inspectable:

1. same-name code search on GitHub ("name: x" filename:SKILL.md)
2. normalized SHA-256 collapse of exact copies (body-only variants kept separate)
3. conservative relatedness gate (name match requires content corroboration)
4. mutation feature vectors (plain regex + RapidFuzz)
5. near-copy grouping (union-find ≥ 0.90 body ratio + hub partition)
6. archetype classification by fixed signal rules
7. per-archetype representative scoring with absorber/deletion/placeholder penalties
8. agent semantic layer: PASS A group analysis → behavior-equivalence consolidation → per-group verification → deterministic acceptance

No LLM inside the engine; the agent layer is your own coding agent.

What it does not claim

No ancestry proof ("original", "copied from"), no census ("all variants"), no quality-by-frequency, no security judgment. Agent interpretations may vary; the semantic validation to date is internal with a same-model caveat. See docs/limitations.md.

Validation

Validated on three high-copy skill families, five known adaptation anchors, and 243 human-audited mutation groups. All five anchors were found and correctly classified; the consolidation guardrail reduced over-merge from 26% to 0% with 100% two-run stability. Methodology: research/validation-summary.md, research/agent-benchmark/v1/.

Contributing

False-positive and misclassification reports are the most valuable contributions — see CONTRIBUTING.md.

Credits / research inspiration

Inspired by public Agent Skills ecosystems including obra/superpowers and anthropics/skills.

Third-party test fixture redistribution was reviewed separately. No anthropics/skills Skill text is bundled because no repository license was found at audit time. See research/fixture-audit.md.

License

Apache-2.0