ajaygunalan

Mark a human-verified milestone — summarizes work since last checkpoint, creates an empty checkpoint commit, pushes

Finds and synthesizes papers using seed-then-snowball strategy. S2 for seeds + citation chasing, OpenAlex for semantic search + trust signals. Three modes: landscape (map a field), validate (test a claim), critique (stress-test a document). AI reads papers and teaches through chat. User may also read directly.

One-time setup — create main + N numbered worktree slots for any repo

Search past Claude Code conversation history. Use when asked to recall, find, or search for anything from previous conversations - including content discussed, links shared, problems solved, topics covered, things posted, or work done together. Triggers include "what did we do today", "summary of our work", "what did we work on", "from our conversations", "what did we discuss", "which X was about

Create D2 diagrams. Triggers on "create a diagram," "visualize," "draw architecture," "show relationships."

Build the documentation index for a codebase from scratch — README, diagrams, markdown files, AGENTS.md

Sync the documentation index against current codebase state — README, diagrams, markdown files, AGENTS.md

Push to Google Drive — uses current directory to decide scope. From gdrive root pushes everything, from a subfolder pushes only that folder.

End-of-session distillation — scans the conversation and existing issues, filters signal from noise through adaptive questioning, then creates, updates, or closes issues. Use when the user says "harvest", wants to wrap up a session, capture decisions before closing, or bridge work to a future conversation.

Squash-merge a worktree branch into main — rebase, test, merge, free slot

Generate a D2 diagram that explains how a codebase feature, flow, or architecture works. Designed for fast onboarding — a visual mental model readable in under 2 minutes. Use when asked to walkthrough, explain a flow, trace a code path, show how something works, or explain the architecture.

Autonomous investigation — spawn subagents, trace code, find answers, report back

Check plans or modified files against conversation intent. Fix what's clear, ask about what's ambiguous.