jordanhubbard/theseus/qualify

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Run the Theseus gold-set qualification protocol (ADR 0004). Use when attempting regenerative replacement against a held-out oracle. Never mark qualified after a single generation. Kill gate: fewer than half of ≥5 attempts means characterization stays the product (ADR 0005).

Install

npx vskill@latest install jordanhubbard/theseus/qualify

SKILL.md

Qualify a gold-set family

Qualification is a research protocol, not the shipping product.

When to use

  • A family already has characterized authority + public oracle + held-out oracle + implementation
  • You are running tools/qualify.py or interpreting receipts under reports/qualification/

Hard rules

  1. Never set qualified: true unless independent_generation is true (two empty-workspace gens, different impl hashes, both pass public + held-out + isolation + held_out_guard).
  2. Do not copy factory theseus_* zero-arg wrappers. Gold-set oracles call the public API with arguments (theseus_json, theseus_*_q).
  3. Do not put held-out vectors in zspecs/ public files or in synthesis prompts.
  4. Keep gold/<family>/package.md qualification: none until dual-generation actually happens.
  5. If ≥5 families are attempted and fewer than half qualify, do not advertise replacement. Characterization remains the product.

Commands

bash
python3 tools/qualify.py <family>
python3 tools/qualify.py <family> --run1 <gen-a-pkg> --run2 <gen-b-pkg>
python3 tools/qualify.py --all
make qualify-check
python3 tools/lint_gold_wrappers.py

--run1 and --run2 are two external empty-workspace package directories (each contains the implementation). Both are required before a receipt can set qualified. --second-impl only compares a second tree with the in-repo implementation and cannot qualify. A single-family run rewrites that family's receipt and rebuilds summary.json from every receipt on disk.

Receipts

Each family writes reports/qualification/<family>.json. Summary and kill-gate bit: reports/qualification/summary.json.

Security

  • Security scanPassed, 95 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
35
Repository forks
3
7-day trend score
23

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/jordanhubbard/theseus/qualify/badge)](https://verified-skill.com/skills/jordanhubbard/theseus/qualify)

Something wrong with this skill or its page? Tell us.