Skip to content

AGENTS.md

Latest commit

 

History

History
140 lines (105 loc) · 8.18 KB

AGENTS.md

File metadata and controls

140 lines (105 loc) · 8.18 KB
Raw

ScholarAIO - Agent Entry

This is the repository entry document for coding agents. It is intentionally short:

  • keep durable project facts and hard constraints here
  • move multi-step procedures into skills
  • move deep reference material into the indexed docs/ knowledge base

What ScholarAIO Is

ScholarAIO is an All-in-One academic harness for agents. Researchers work through a coding agent to search literature, inspect evidence, organize persistent project context, produce reviewable research outputs, and use bounded scientific-tool adapters without ScholarAIO replacing the agent's native reasoning or orchestration.

The Python package is scholaraio. Real work should usually happen through the ScholarAIO CLI and project skills, not by bypassing runtime helpers with ad hoc file edits.

How To Work In This Repo

  • Prefer a matching project skill in .claude/skills/ when the user request clearly maps to one.
  • Use the scholaraio CLI to do real work instead of only describing what should be done.
  • Load information progressively. Prefer metadata or abstracts first, then conclusions or full text only when needed.
  • Treat paper conclusions as claims, not facts. Compare evidence, point out limitations, and distinguish supported results from author interpretation.
  • Keep user-facing drafts, reports, exports, and research outputs under workspace/, not the repository root or scholaraio/.
  • Do not casually rewrite or delete runtime data. When touching runtime layout, use Config, migration helpers, and tested accessors.
  • When changing runtime layout, compatibility behavior, agent entry docs, or skill discovery, update tests and aligned docs in the same change.
  • After code or doc changes, run the relevant checks and at least one real smoke path when feasible.

Start Here

Read these in roughly this order:

  1. README.md for the product overview and top-level structure.
  2. STRATEGY.md before changing product scope or adding an external integration.
  3. docs/DESIGN.md for the repository knowledge map.
  4. docs/getting-started/agent-setup.md for repo-open vs plugin or cross-project setup.
  5. docs/guide/cli-reference.md for the current user-facing CLI surface.
  6. docs/guide/agent-reference.md for deeper agent, runtime, and skill organization details.
  7. docs/internal/PLANS.md and docs/internal/exec-plans/completed/scholaraio-upgrade-plan.md before changing runtime layout, migration, or compatibility behavior.

Skill-First Workflow

The canonical project skill source is .claude/skills/. Cross-agent discovery paths are wrappers around that same source:

  • .agents/skills/
  • .qwen/skills/
  • skills/

Representative skills to check first:

  • Core research: search, show, ingest, ingest-link, workspace, audit, translate
  • Writing: academic-writing, nature-workflow, literature-review, paper-guided-reading, paper-writing, citation-check, writing-polish, review-response, research-gap, poster, technical-report
  • Outputs and tooling: draw, document, publish, scientific-runtime, scientific-tool-onboarding

If a workflow has grown into a reusable playbook, move it into a skill instead of expanding this file.

Repo Map

  • scholaraio/core/: config, logging, and shared runtime foundations
  • scholaraio/providers/: external service clients and parsing backends
  • scholaraio/stores/: persistent library roots and storage helpers
  • scholaraio/projects/: user project and workspace behavior
  • scholaraio/services/: domain logic and orchestration
  • scholaraio/interfaces/cli/: CLI parser, startup, and command handlers

The breaking cleanup generation removed legacy root-level public facades such as scholaraio.index, scholaraio.workspace, and scholaraio.translate. New code should import canonical namespaces directly.

High-signal canonical implementation pointers:

  • scholaraio/stores/explore.py
  • scholaraio/projects/workspace.py
  • scholaraio/services/insights.py
  • scholaraio/services/translate.py
  • scholaraio/interfaces/cli/
  • scholaraio/interfaces/cli/compat.py for internal CLI wiring
  • scholaraio/cli.py as the published entrypoint only

Current Runtime Model

  • Runtime is fresh-layout-only under data/libraries/, data/spool/, and data/state/.
  • Legacy roots such as data/papers/, data/explore/, data/proceedings/, and data/inbox* are no longer normal runtime inputs. Use scholaraio migrate upgrade --migration-id <migration-id> --confirm for the one-command supported migration, or scholaraio migrate ... to inventory, verify, and run individual stores.
  • workspace/<name>/ stays a free-form user project tree.
  • Workspace paper references live in workspace/<name>/refs/papers.json.
  • workspace.yaml is additive metadata only; it does not replace refs/papers.json.
  • System-owned workspace outputs live under workspace/_system/, especially:
    • workspace/_system/translation-bundles/
    • workspace/_system/figures/
    • workspace/_system/output/
  • Runtime-layout migration control lives under .scholaraio-control/. The standardized one-command upgrade gate is scholaraio migrate upgrade --migration-id <migration-id> --confirm; finalize remains the post-store cleanup/final verification step.

Commands To Know

  • scholaraio --help
  • For repo-local validation, prefer python -m scholaraio.cli ... so you are exercising the current checkout instead of an older installed console script.
  • scholaraio setup check
  • scholaraio search --help
  • scholaraio show --help
  • scholaraio gui --help
  • scholaraio pipeline --help
  • scholaraio ws --help
  • scholaraio migrate --help
  • scholaraio migrate upgrade --help
  • scholaraio migrate finalize --help

Common verification commands in this repo:

  • python -m pytest -q -p no:cacheprovider
  • python -m ruff check scholaraio tests
  • python -m ruff format --check scholaraio tests
  • python -m mkdocs build --strict

Multi-Agent Entry Points

  • Claude Code: CLAUDE.md + .claude/skills/
  • Codex / OpenClaw: AGENTS.md + .agents/skills/
  • Qwen: .qwen/QWEN.md + .qwen/skills/
  • Cursor: .cursor/rules/scholaraio.mdc, then AGENTS.md
  • Cline: .clinerules, then AGENTS.md
  • Windsurf: .windsurfrules, then AGENTS.md
  • GitHub Copilot: .github/copilot-instructions.md, then AGENTS.md

Keep these wrappers lightweight. Do not turn every wrapper into a second full manual.

Optional webtools MCP servers are listed in .mcp.json for hosts that support project MCP JSON. Codex uses its own MCP registry; see docs/guide/webtools-integration.md for codex mcp add ... commands.

Deep Reference

Use the smallest doc that answers the question:

When in doubt, keep this file short, keep skills procedural, and keep deep detail in dedicated reference docs.