[Feature] Separate runtime artifacts from repository directory

[kr37swarm]

Summary

AutoResearchClaw writes all runtime artifacts (artifacts/, docs/kb/, experiment outputs) into the cloned repository directory. There is no runtime directory concept. This pollutes the repo, complicates git workflows, and prevents clean separation of product vs instance.

Current Behavior

autoresearchclaw/           # git clone
├── researchclaw/           # source code
├── artifacts/              # ← runtime data in repo
│   ├── rc-20260324-.../
│   └── rc-20260320-.../
├── docs/kb/                # ← runtime knowledge base in repo
└── config.yaml             # ← user config in repo root

Proposed Behavior

autoresearchclaw/            # git clone (clean, stays pushable)
├── researchclaw/            # source code
└── config.example.yaml

~/.autoresearchclaw/         # runtime directory (XDG_DATA_HOME)
├── config.yaml              # user config
├── runs/                    # per-run artifacts
├── kb/                      # knowledge base
└── cache/

Suggested Implementation

1. Add RESEARCHCLAW_HOME env var (default ~/.autoresearchclaw/)
2. Change default --output to $RESEARCHCLAW_HOME/runs/<run_id>/
3. researchclaw init creates the runtime directory structure
4. --output flag continues to work for overrides

This follows the pattern used by Claude Code (~/.claude/), Codex (~/.codex/), Gemini (~/.gemini/), and most modern CLI tools.

[Jiaaqiliu]

Hi @kr37swarm, thanks for this well-structured proposal!

You raise a valid architectural concern. Mixing runtime artifacts with the source repository does create friction for git workflows and deployment.

Current state:

We do support --output to redirect run artifacts to a custom directory, and .gitignore already excludes artifacts/ and other runtime data from git tracking. However, you are right that there is no first-class RESEARCHCLAW_HOME concept.

Our thoughts:

1. The RESEARCHCLAW_HOME env var pattern (following ~/.claude/, ~/.codex/ conventions) is a good direction. We will consider this for a future release.

2. For now, you can achieve similar separation by:

   # Set output directory for all runs
   researchclaw run --output ~/.autoresearchclaw/runs/ "your topic"

   Or configure in config.arc.yaml:

   knowledge_base:
     root: "~/.autoresearchclaw/kb"

3. The researchclaw init wizard already creates config.arc.yaml in the current directory. Extending it to support a global config location is on our roadmap.

We are tracking this as a feature request for a future version. The XDG-compliant directory structure you proposed is well-designed. Thanks for the detailed suggestion!