Proof Orchestrator
Runs multi-session proof projects with local-first attempts, manual GPT Pro handoff packages, and optional DeepSeek review.
- Stars
- 18,482
- Licence
- MIT
- Files
- 8 files
What it does
Once installed, the agent organizes a mathematical proof as a numbered, resumable run folder. It first tries the proof locally, checks it, and tidies the result. If it gets stuck, it writes a self-contained prompt you can paste into GPT Pro in your browser, and it keeps track of what has been proved, what is still conjectural, and what sources were used, so a later session can pick up where the last one stopped. It does not operate a browser or spend API credit by default.
Great for
- Mathematicians or students working on a long proof that spans several sessions
- Researchers who want a clean audit trail of which claims were proved locally and which came from an outside model
- People who already use GPT Pro or a similar chat tool and want a structured handoff prompt
Skip it if
- Quick, one-off proof drafting; the SKILL.md routes that to a separate /proof-writer skill
- Final verification before paper submission; this skill points to /proof-checker for that gate, so relying on it alone could miss errors
- Anyone expecting automatic verification: the output is a model's reasoning and still needs expert review
Try asking
- “Continue the proof from last week's run directory and identify the next narrow obligation”
- “Prepare a GPT Pro handoff package for the lemma that the local attempt could not close”
- “Get an independent DeepSeek second opinion on the current proof run as extra evidence”
- “Audit the notation in this proof and report the symbol scorecard”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 8 original files, the MIT licence and an attribution note.
proof-orchestrator.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip proof-orchestrator.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip proof-orchestrator.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 3
Ask for the task
No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.
Get better results
- Give the agent the exact theorem statement and any source papers up front, since the skill relies on clean source snapshots
- Use the manual handoff by default; only ask the agent to call GPT Pro or DeepSeek when you are comfortable with that data leaving your machine
- Review gpt-pro-output.md and audit.md yourself before trusting any claim marked proved
What's inside
- NOTICE.md
- references/audit-output-contract.md
- references/deepseek-routing.md
- references/dispatch-prompts.md
- references/notation-audit.md
- references/proof-audit-rubric.md
- references/stress-tests.md
- SKILL.md
Redistributed unmodified from wanshuiyin/Auto-claude-code-research-in-sleep under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.