Clawboard

Status

• ClawHub: not available yet.
• Install path today: scripted bootstrap or manual repo install.
• If OpenClaw is missing and user wants Chutes provider, start with add_chutes.sh after creating a https://chutes.ai account.

Current Architecture Snapshot

• web (Next.js): http://localhost:3010
• api (FastAPI + SQLite): http://localhost:8010
• classifier (async stage-2 worker): topic/task classification loop
• qdrant (vector index): dense retrieval backend on internal Docker network
• clawboard-logger plugin: stage-1 firehose logging + response-time context extension

Retrieval/search stack:

• Dense vectors + BM25 + lexical matching
• Reciprocal rank fusion + reranking
• Qdrant primary, SQLite embeddings mirror/fallback

Goal

Get a user to a working Clawboard install where:

1. Clawboard web/api/classifier are running.
2. clawboard-logger plugin is installed and enabled.
3. Token flow is configured correctly (required for writes + non-localhost reads).
4. OpenClaw gateway is restarted and logging into Clawboard.

Hard Rules (Repo vs Installed Skill)

• Repo copy (version controlled): $CLAWBOARD_DIR/skills/clawboard
• Installed skill path (what OpenClaw reads): $HOME/.openclaw/skills/clawboard
• Default install mode is symlink (~/.openclaw/skills/clawboard -> $CLAWBOARD_DIR/skills/clawboard).
• Detect mode deterministically at runtime:

if [ -L "$HOME/.openclaw/skills/clawboard" ]; then
  echo "symlink"
else
  echo "copy"
fi

• In symlink mode (default), repo edits are immediately visible to OpenClaw.
• In copy mode, repo edits do not update OpenClaw until you sync/copy again. After skill file changes (SKILL.md, agents/, references/, scripts/), sync into OpenClaw:

cd "$CLAWBOARD_DIR"
bash scripts/sync_openclaw_skill.sh --to-openclaw --apply --force

• If you changed files under ~/.openclaw/skills/clawboard while in copy mode, sync them back before committing:

cd "$CLAWBOARD_DIR"
bash scripts/sync_openclaw_skill.sh --to-repo --apply

Workspace + Runtime Assumptions (for coding tasks)