OpenClaw Optimization

Appearance

TOOLS.md — Coder ​

Infrastructure references, account details, and tool-specific patterns for this workspace.

Workspace ​

  • Root: ~/.openclaw/workspaces/coder/
  • Temp files: /tmp/coder-* — always clean up when done

GitHub CLI (gh) ​

  • Account:
  • Auth: pre-configured via gh auth login before first use
  • Scope: /*

PR Creation (Mandatory Pattern) ​

Never pass multiline markdown with --body "..." — shell escaping breaks it. Always use --body-file:

bash
cat <<'EOF' > /tmp/pr_body.md
## Summary
- What changed and why

## Testing
- How it was verified
EOF

gh pr create --title "fix: ..." --body-file /tmp/pr_body.md

Git ​

  • Branch naming: fix/issue-<num>-<short-desc> or feat/<short-desc>
  • Commits: conventional commits — feat:, fix:, chore:, docs:, test:, etc.
  • Never commit directly to main

⚠️ No Per-Task Clones ​

Never run git clone or gh repo clone to create a new directory for each task. Each repo gets one canonical clone; all tasks use worktrees off it.

The initial clone (when the canonical doesn't exist yet) is fine and expected — see the worktree pattern below.

Worktree Pattern (Mandatory) ​

Each repo has a single canonical clone in the workspace root. All tasks use worktrees off that clone — they share the git object store, support full parallelism, and avoid checkout conflicts.

bash
REPO_NAME="my-repo"
CANONICAL="$HOME/.openclaw/workspaces/coder/$REPO_NAME"
TASK_ID="issue-42"
BRANCH="fix/$TASK_ID-short-desc"
WORKTREE="/tmp/${REPO_NAME}-wt-${TASK_ID}"

# One-time: clone the canonical if it doesn't exist yet
# Use --filter=blob:none for large repos (lazily fetches blobs, keeps full commit history)
if [ ! -d "$CANONICAL" ]; then
  gh repo clone {{GH_ORG}}/$REPO_NAME "$CANONICAL"
fi

# Per-task: fetch latest, create isolated worktree on a new branch
git -C "$CANONICAL" fetch origin
git worktree add "$WORKTREE" -b "$BRANCH" origin/main

# Do all work in $WORKTREE
cd "$WORKTREE"
# ... implement, test, commit ...

# Push and open PR
git push -u origin "$BRANCH"
gh pr create --title "..." --body-file /tmp/pr_body.md

# Cleanup after PR is open
git worktree remove "$WORKTREE"

cgc (CodeGraphContext) ​

cgc is a code graph analysis tool that indexes a codebase into a queryable graph. Use it during Phase 1 planning to understand blast radius before changing shared code. Assume it is already installed globally.

bash
# Index the repo (run once from repo root, or after major changes)
cgc index .

# Find everything that depends on a module
cgc analyze deps "my-module"           # or "@myorg/shared-package" for scoped packages

# Find all callers of a function
cgc analyze callers myFunction

# Find complex functions worth reviewing before changing
cgc analyze complexity --threshold 15 --limit 10

When to use:

  • During Phase 1 planning — understand what depends on the code you're about to change
  • Before modifying shared packages in a monorepo — check blast radius
  • When reviewing PRs that touch shared code

session_status ​

Used to switch models between workflow phases:

session_status(model="codex")   # Phase 2 — implementation
session_status(model="opus")    # Phase 1/3 — planning/review

Built with OpenClaw 🤖