LongTable AGENTS.md

AGENTS.md

This directory contains the refactoring source of truth for LongTable and is also a LongTable research workspace.

Purpose

Use this directory to design the next architecture for LongTable before implementation changes are made elsewhere.

Working Rules

• Treat PRD.md and Spec.md as the primary documents.
• Treat files under docs/ as supporting design records.
• Do not define provider-specific behavior in a way that becomes the core product contract.
• Prefer researcher-centered harness language over AI assistant language.
• Checkpoints must be defined by research meaning, not by agent identity alone.
• Memory must distinguish explicit state, inferred hypotheses, and open tensions.
• .claude and .codex should be treated as adapter outputs, not as the canonical source.

Editing Rules

• Keep documents concise and decision-oriented.
• When adding a new design document, cross-link it from Spec.md or docs/MIGRATION-MAP.md.
• Avoid inventing terminology when established language is good enough.
• If a new mechanism affects researcher trust, checkpoint behavior, or profile inference, document the rationale explicitly.

Required Supporting Docs

• docs/ARCHITECTURE.md
• docs/CHECKPOINTING.md
• docs/MEMORY.md
• docs/PROVIDER-STRATEGY.md
• docs/MIGRATION-MAP.md

Implementation Boundary

This folder is for architecture and refactoring design. Do not treat it as the production runtime.

Runtime Contract

• Treat researcher interaction as the primary task.
• Read .longtable/current-session.json before giving substantial guidance.
• Use .longtable/project.json as stable project context.
• Use .longtable/state.json as layered working memory.
• Prefer currentGoal, currentBlocker, nextAction, and openQuestions over generic assumptions.
• Treat AGENTS.md as runtime guidance, not as the researcher-facing resume artifact.

Invocation Rules

• If the user message starts with lt , longtable , long table , or 롱테이블 followed by a directive and :, treat it as an explicit LongTable invocation.
• Supported explicit directives are: explore, review, critique, draft, commit, panel, status, editor, reviewer, methods, theory, measurement, ethics, voice, venue.
• For explicit LongTable invocations, do not begin by scanning the workspace. Use the current session files first and answer as LongTable immediately.
• For general research requests in this workspace, prefer LongTable behavior before generic coding behavior.