Marathon work session. Pick up a ticket and stop at nothing to discharge it — fill every sorry, develop whatever API is needed, invoke /develop to plan more tickets whenever the path forks, and immediately work those tickets too. The session ends only when the target is fully proven or the work is genuinely outside…
Author or update the project's verso-blueprint — a Verso-based Lean module pairing each declaration with its mathematical statement and a paragraph-level proof sketch, producing an interactive HTML dep-graph + progress summary. Uses the standard leanprover/verso-blueprint tooling (the one behind verso-sphere-packing…
Profile Lean declarations, find the slow ones, trace-diagnose the root cause, and fix them until each elaborates fast — ideally under a second. Statements never change; maxHeartbeats never goes up.
Project-wide cleanup using an orchestrator-worker pattern. The main session is the orchestrator — it does NOT read files, run lean tools, or edit. It dispatches batched per-file Agent calls with tight prompts (working dir + branch + build state + sequence + target), then narrates progress in one-line scoreboards…
Plan a mathematical development project (planning-only). Searches mathlib, designs the API, then runs a binding methodical decomposition pre-work pass — for each top-level result, writes the prose proof, decomposes into ordered lemmas, writes every lemma as a := by sorry declaration in the project's Lean files (the…
Produce a self-contained mathematical briefing (no Lean, no file paths) for an external reviewer with no repo access — goals, plan, status, blockers, references, numbered questions.
Audit a lemma or definition for assumption-weakening opportunities — try mechanical weakenings, search the literature for the maximally-general form, auto-apply small safe changes, propose big changes for user approval.
Decide whether a Lean declaration belongs in mathlib. Methodical, gated workflow that combines thorough literature search (WebSearch + ChatGPT MCP + local references + nLab/Stacks/MathOverflow for big decls) with mathlib's five-method exhaustive search, then a generality analysis against the literature-standard form…
Chat-only mathematical status of a /develop project. The agent reads the project's .lean files, identifies the result the worker is currently on (or blocked on), and reports in mathematical English what's being proved, where the proof sits, how it connects to the overall goal, and how far along the whole project is.…
Run N rounds of neutral, independent review-and-implement on your changes before a PR. Each round spawns a FRESH, unbiased review agent (techniques from the built-in /review and from /check-style, specialised to four Lean dimensions — definition necessity, generalisation, automation, mathlib naming/style), reports its…
The Tau Ceti worker loop — one unit of work per round, first that applies: rebase, fix CI, fix findings, review a stalled PR, else author a new one. Authoring checks the source material first, cleans up, and opens the PR; CI reviews it. We only self-review a PR that CI has left unreviewed for an hour.
Turn a Lean declaration into mathematics. Renders the statement and a paragraph-level proof sketch as Unicode in the terminal by default (Γ, ℂ, →, ≤ — readable in chat without any math renderer). Optionally outputs Verso markup (--verso), Markdown (--md), or appends the result to the project's verso-blueprint as a…