chore: enforce main-only workflow and adopt AGENTS.md

This commit is contained in:
lucaronin
2026-04-07 18:42:41 +02:00
parent 578f369f4c
commit b384cbf6ee
12 changed files with 427 additions and 513 deletions

275
.github/HOOKS.md vendored
View File

@@ -1,257 +1,64 @@
# Git Hooks
## Pre-Commit Hook: CodeScene Check
This repo uses Husky hooks from `.husky/`. Those files are the source of truth.
Il repository ha un pre-commit hook che verifica la qualità del codice prima di ogni commit.
## Installation
## Post-Commit Hook: Auto-Implement Design Changes
`pnpm install` runs the `prepare` script and installs the hooks into `.git/hooks`.
Quando committi modifiche a `ui-design.pen`, il post-commit hook:
1. Analizza automaticamente le modifiche (colori, typography, spacing, layout)
2. Spawna Claude Code in background per implementare le modifiche
3. Ti notifica quando l'implementazione è completa
---
## Pre-Commit Hook Details
### Cosa Fa
1. **Analizza file staged** — controlla solo TypeScript/Rust modificati
2. **Confronta con base branch**`origin/main` per branch, `HEAD~1` per main
3. **Avvisa per file grandi** — >500 linee modificate
4. **Suggerisce review** — con Claude Code + CodeScene MCP per analisi dettagliata
### Bypass Hook
Se sai cosa stai facendo:
If you need to reinstall them manually:
```bash
# Skip hook per questo commit
git commit --no-verify -m "your message"
# O includi nel commit message
git commit -m "your message [skip codescene]"
pnpm exec husky
```
### Installazione (già fatto per questa repo)
The hooks expect `node` and `pnpm` to be available. If they are installed via `nvm`, the hooks will try to load `~/.nvm/nvm.sh` automatically.
L'hook è già installato in `.git/hooks/pre-commit`.
## Policy
Se cloni la repo altrove, copia l'hook:
```bash
cp .github/hooks/pre-commit .git/hooks/pre-commit
chmod +x .git/hooks/pre-commit
```
- Commit on `main` only.
- Push from `main` to `origin/main` only.
- Never use `--no-verify`.
- `.codescene-thresholds` is a ratchet. It can only move up.
### Esempio Output
## Pre-commit
#### ✅ Commit Normale
```
🔍 Running CodeScene Code Health check...
Comparing against: origin/main
Analyzing code changes...
✅ CodeScene check passed
+42 -18 lines
`.husky/pre-commit` blocks commits unless all of the following are true:
💡 For detailed code health analysis, run:
claude 'Check code health of this commit with CodeScene MCP'
```
- `HEAD` is attached to `main`
- staged TypeScript files pass `pnpm lint --quiet`
- TypeScript passes `npx tsc --noEmit`
- frontend tests pass via `pnpm test --run --silent`
- current CodeScene Hotspot and Average health are both at or above `.codescene-thresholds`
#### ⚠️ File Grandi
```
🔍 Running CodeScene Code Health check...
Comparing against: origin/main
Analyzing code changes...
⚠️ Large file changes detected (>500 lines):
- src/components/Editor.tsx
- src-tauri/src/vault.rs
If `CODESCENE_PAT` or `CODESCENE_PROJECT_ID` is missing, the CodeScene portion is skipped, but the rest of the hook still runs.
Consider:
- Breaking into smaller commits
- Reviewing with Claude Code + CodeScene MCP
- Running: claude 'Review code health of staged changes'
## Pre-push
Continue anyway? (y/N)
```
`.husky/pre-push` blocks pushes unless all of the following are true:
### CodeScene MCP Integration
- the current branch is `main`
- every pushed branch ref is `refs/heads/main -> refs/heads/main`
- TypeScript and the Vite build pass
- frontend coverage passes
- Rust lint and Rust coverage pass when `src-tauri/` changed
- Playwright smoke tests pass when `tests/smoke/*.spec.ts` exists
- current CodeScene Hotspot and Average health are both at or above `.codescene-thresholds`
Per analisi dettagliata del code health, usa Claude Code:
If the remote CodeScene scores are better than the current thresholds, the hook updates `.codescene-thresholds`, stages it, and stops the push. Commit that file normally, then push again. The hook does not auto-commit or bypass itself.
## Legacy Files
The legacy `pre-commit` and `post-commit` files under `.github/hooks/` are archival only. Do not copy them into `.git/hooks`; use Husky and `.husky/` instead. `install-hooks.sh` remains as a reinstall helper that runs Husky.
## Troubleshooting
If a hook cannot find `node` or `pnpm`:
```bash
# Analizza staged changes
claude 'Check code health of staged changes with CodeScene MCP'
# Analizza file specifico
claude 'What is the code health score of src/components/Editor.tsx?'
# Pre-commit safeguard
claude 'Run pre_commit_code_health_safeguard on staged changes'
export NVM_DIR="$HOME/.nvm"
. "$NVM_DIR/nvm.sh"
nvm use node
```
### Troubleshooting
**Hook non si attiva:**
- Verifica che `.git/hooks/pre-commit` esista ed sia eseguibile
- `ls -la .git/hooks/pre-commit` — dovrebbe mostrare `-rwxr-xr-x`
**Vuoi disabilitare temporaneamente:**
```bash
mv .git/hooks/pre-commit .git/hooks/pre-commit.disabled
```
**Vuoi riabilitare:**
```bash
mv .git/hooks/pre-commit.disabled .git/hooks/pre-commit
```
### Future Improvements
Possibili miglioramenti:
- [ ] Integrazione diretta API CodeScene per score numerico
- [ ] Fail automatico se code health < soglia
- [ ] Cache dei risultati per evitare re-analisi
- [ ] Hook pre-push più pesante per analisi completa
---
## Post-Commit Hook: Auto-Implement Design Changes
### Cosa Fa
Quando committi modifiche a `ui-design.pen`, il hook:
1. **Analizza il diff** — usa `scripts/design-diff-analyzer.js`
2. **Identifica modifiche significative:**
- 🎨 Colori (fill, backgroundColor)
- 📝 Typography (fontSize, fontFamily)
- 📏 Spacing (padding, margin, gap)
- 🔲 Layout (nuovi componenti, riorganizzazioni)
3. **Spawna Claude Code** — in background via `openclaw sessions spawn`
4. **Auto-notifica** — quando l'implementazione è completa
### Cosa Implementa
| Tipo Modifica | Azione |
|--------------|--------|
| Colori | Aggiorna `src/theme.json` o CSS variables |
| Typography | Aggiorna `src/theme.json` typography |
| Spacing | Aggiorna `src/theme.json` spacing |
| Layout | Modifica/crea componenti React |
| Testi mockup | Nessuna azione (solo design) |
### Esempio Output
```
🎨 Design file changed - analyzing...
📋 Implementation tasks:
1. [HIGH] Update color palette
- fill: $--muted-foreground → #666666
- backgroundColor: #FFFFFF → #F5F5F5
Update src/theme.json or CSS variables to match the design.
2. [MEDIUM] Update typography
- fontSize: 14px → 16px
Update src/theme.json typography settings.
🚀 Spawning Claude Code to implement changes...
✅ Claude Code spawned - you'll be notified when implementation is complete
```
### Workflow Completo
```
You Post-Commit Hook Claude Code Brian (AI)
│ │ │ │
├─ Modify ui-design.pen │ │ │
├─ git add ui-design.pen │ │ │
├─ git commit │ │ │
│ │ │ │
│ ├─ Analyze diff │ │
│ ├─ Generate tasks │ │
│ ├─ Spawn Claude Code ────────> │
│ │ │ │
│ │ ├─ Implement changes │
│ │ ├─ Test visually │
│ │ ├─ Run tests │
│ │ ├─ Commit │
│ │ ├─ openclaw system event ────>
│ │ │ │
│ │ │ ├─ Notify Telegram
│ <─────────────────────────────────────────────────────────────────────────────┘
│ "✅ Design changes implemented and tested"
```
### Design Diff Analyzer
Lo script `scripts/design-diff-analyzer.js` rileva:
- **Color changes** — `"fill": "#OLD" → "#NEW"`
- **Font changes** — `"fontSize": 14 → 16`
- **Spacing** — `"padding": 8 → 12`
- **Content** — testi mockup (no implementation)
Uso:
```bash
# Analizza ultimo commit
node scripts/design-diff-analyzer.js
# Output JSON per automation
node scripts/design-diff-analyzer.js --json
```
### Disabilitare Temporaneamente
Se vuoi committare il design senza auto-implementazione:
```bash
# Disabilita hook
mv .git/hooks/post-commit .git/hooks/post-commit.disabled
# Commit
git commit -m "design: update mockup"
# Riabilita hook
mv .git/hooks/post-commit.disabled .git/hooks/post-commit
```
### Monitorare Claude Code
Mentre Claude Code lavora in background:
```bash
# Lista sub-agent attivi
openclaw sessions list --kinds isolated
# Vedi log di un sub-agent
openclaw sessions history --session-key <key>
# Ferma sub-agent (se necessario)
openclaw subagents kill --target design-auto-implement
```
### Troubleshooting
**Hook non parte:**
- Verifica che `ui-design.pen` sia effettivamente cambiato: `git diff HEAD~1 ui-design.pen`
- Verifica che lo script analyzer esista: `ls -la scripts/design-diff-analyzer.js`
**Nessuna notifica:**
- Claude Code potrebbe essere ancora in esecuzione — controlla `openclaw sessions list`
- Verifica che il prompt includa `openclaw system event` al termine
**Modifiche non implementate:**
- Controlla i log di Claude Code: `openclaw sessions history --session-key <key>`
- Il sub-agent viene auto-eliminato dopo completion (cleanup=delete)
### Limitazioni
- **Modifiche complesse** — layout completamente nuovi potrebbero richiedere intervento manuale
- **Timeout** — 10 minuti max (configurabile in hook)
- **Solo modifiche recenti** — analizza solo HEAD vs HEAD~1
Then retry the commit or push.

View File

@@ -1,25 +1,26 @@
#!/bin/bash
# Install git hooks for laputa-app
set -e
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
HOOKS_DIR="$(git rev-parse --git-dir)/hooks"
if ! command -v pnpm >/dev/null 2>&1 || ! command -v node >/dev/null 2>&1; then
export NVM_DIR="${NVM_DIR:-$HOME/.nvm}"
if [ -s "$NVM_DIR/nvm.sh" ]; then
# shellcheck disable=SC1090
. "$NVM_DIR/nvm.sh" --no-use
nvm use --silent node >/dev/null 2>&1 || true
fi
fi
echo "Installing git hooks..."
# Copy pre-commit hook
cp "$SCRIPT_DIR/pre-commit" "$HOOKS_DIR/pre-commit"
chmod +x "$HOOKS_DIR/pre-commit"
echo "✅ Installed pre-commit hook"
# Copy post-commit hook
cp "$SCRIPT_DIR/post-commit" "$HOOKS_DIR/post-commit"
chmod +x "$HOOKS_DIR/post-commit"
echo "✅ Installed post-commit hook"
if ! command -v pnpm >/dev/null 2>&1 || ! command -v node >/dev/null 2>&1; then
echo "❌ node and pnpm must be available to install Husky hooks"
exit 1
fi
echo "Installing Husky hooks from .husky/ ..."
pnpm exec husky
echo "✅ Husky hooks installed"
echo ""
echo "Hooks installed:"
echo " - pre-commit: CodeScene code health check"
echo " - post-commit: Auto-implement design changes via Claude Code"
echo "Source of truth:"
echo " - .husky/pre-commit"
echo " - .husky/pre-push"
echo ""
echo "To bypass pre-commit, use: git commit --no-verify"
echo "Or include [skip codescene] in your commit message"
echo "Never use --no-verify in this repo."

View File

@@ -2,8 +2,6 @@ name: CI
on:
push:
branches: [main, experiment/*]
pull_request:
branches: [main]
jobs:
@@ -82,16 +80,14 @@ jobs:
# ── 3. Code Health (CodeScene — Hotspot + Average Code Health gates) ──
# Enforces minimum floors on BOTH hotspot and average code health.
# Hotspot: weighted avg of most-edited files (9.6 current | target 9.8)
# Average: project-wide avg across all files (9.37 current | target 9.5)
# Both gates must pass — average catches regressions in non-hotspot files.
- name: Code Health gates (Hotspot ≥9.5 + Average ≥9.0)
# Thresholds come from .codescene-thresholds so CI and local hooks match.
- name: Code Health gates
env:
CODESCENE_PAT: ${{ secrets.CODESCENE_PAT }}
CODESCENE_PROJECT_ID: ${{ secrets.CODESCENE_PROJECT_ID }}
run: |
HOTSPOT_THRESHOLD=9.5
AVERAGE_THRESHOLD=9.33
HOTSPOT_THRESHOLD=$(grep '^HOTSPOT_THRESHOLD=' .codescene-thresholds | cut -d= -f2)
AVERAGE_THRESHOLD=$(grep '^AVERAGE_THRESHOLD=' .codescene-thresholds | cut -d= -f2)
API_RESPONSE=$(curl -sf \
-H "Authorization: Bearer $CODESCENE_PAT" \
-H "Accept: application/json" \

View File

@@ -1,6 +1,43 @@
#!/bin/sh
# Pre-commit: same checks as CI. Fix here, not in CI.
# Pre-commit: fast local gate before commit. Full suite runs in pre-push/CI.
set -e
ensure_node_tooling() {
if command -v node >/dev/null 2>&1 && command -v pnpm >/dev/null 2>&1; then
return 0
fi
NVM_DIR="${NVM_DIR:-$HOME/.nvm}"
if [ -s "$NVM_DIR/nvm.sh" ]; then
# shellcheck disable=SC1090
. "$NVM_DIR/nvm.sh" --no-use
nvm use --silent node >/dev/null 2>&1 || true
fi
if ! command -v node >/dev/null 2>&1 || ! command -v pnpm >/dev/null 2>&1; then
echo "❌ node and pnpm must be available before committing"
echo " Install them or make sure your nvm setup is available to git hooks."
exit 1
fi
}
require_main_branch() {
if ! git symbolic-ref -q HEAD >/dev/null 2>&1; then
echo "❌ Commits must happen on main. Detached HEAD is not allowed."
exit 1
fi
CURRENT_BRANCH=$(git rev-parse --abbrev-ref HEAD)
if [ "$CURRENT_BRANCH" != "main" ]; then
echo "❌ Commits must happen on main. Current branch: $CURRENT_BRANCH"
echo " Merge or cherry-pick your work onto main, then commit there."
exit 1
fi
}
require_main_branch
ensure_node_tooling
echo "🔍 Pre-commit checks..."
# Lint + types (only if TS files staged)
@@ -19,7 +56,9 @@ echo "✅ Pre-commit passed"
# ── CodeScene Code Health gate ────────────────────────────────────────────
# Blocks commit if scores drop below thresholds in .codescene-thresholds.
# Thresholds are a ratchet — only go up, auto-updated after each successful push.
# Thresholds are a ratchet — only go up.
# When remote scores improve, pre-push updates .codescene-thresholds and stops
# so the new floor is committed with normal verified hooks.
# Note: remote scores lag behind local changes (update after push + re-analysis).
# If check fails: extract hooks, split components, reduce complexity.
# Never use eslint-disable, #[allow(...)], or `as any`.

View File

@@ -28,6 +28,64 @@
# ─────────────────────────────────────────────────────────────────────────
set -e
ensure_node_tooling() {
if command -v node >/dev/null 2>&1 && command -v pnpm >/dev/null 2>&1; then
return 0
fi
NVM_DIR="${NVM_DIR:-$HOME/.nvm}"
if [ -s "$NVM_DIR/nvm.sh" ]; then
# shellcheck disable=SC1090
. "$NVM_DIR/nvm.sh" --no-use
nvm use --silent node >/dev/null 2>&1 || true
fi
if ! command -v node >/dev/null 2>&1 || ! command -v pnpm >/dev/null 2>&1; then
echo "❌ node and pnpm must be available before pushing"
echo " Install them or make sure your nvm setup is available to git hooks."
exit 1
fi
}
require_main_push() {
if ! git symbolic-ref -q HEAD >/dev/null 2>&1; then
echo "❌ Pushes must happen from main. Detached HEAD is not allowed."
exit 1
fi
CURRENT_BRANCH=$(git rev-parse --abbrev-ref HEAD)
if [ "$CURRENT_BRANCH" != "main" ]; then
echo "❌ Pushes must happen from main. Current branch: $CURRENT_BRANCH"
exit 1
fi
while IFS=' ' read -r LOCAL_REF LOCAL_SHA REMOTE_REF REMOTE_SHA; do
[ -z "$LOCAL_REF" ] && continue
case "$LOCAL_REF:$REMOTE_REF" in
refs/heads/main:refs/heads/main)
;;
refs/tags/*:refs/tags/*)
;;
*)
echo "❌ Pushes must be main -> main only."
echo " Attempted: ${LOCAL_REF:-<none>} -> ${REMOTE_REF:-<none>}"
exit 1
;;
esac
done <<EOF
$PUSH_INPUT
EOF
}
if [ -t 0 ]; then
PUSH_INPUT=""
else
PUSH_INPUT=$(cat)
fi
require_main_push
ensure_node_tooling
START_TIME=$(date +%s)
echo ""
@@ -85,7 +143,7 @@ if [ "$RUST_CHANGED" = true ]; then
cargo llvm-cov \
--manifest-path src-tauri/Cargo.toml \
$LLVM_COV_FLAGS \
--ignore-filename-regex "search\.rs|lib\.rs" \
--ignore-filename-regex "lib\.rs|main\.rs|menu\.rs" \
--fail-under-lines 85 \
-- --test-threads=1
echo " ✅ Rust coverage OK"
@@ -98,12 +156,14 @@ echo ""
SMOKE_FILES=$(find tests/smoke -name '*.spec.ts' 2>/dev/null | head -1)
if [ -n "$SMOKE_FILES" ]; then
echo "🎭 [4/5] Playwright smoke tests..."
SMOKE_OUTPUT=$(pnpm playwright:smoke 2>&1) || true
set +e
SMOKE_OUTPUT=$(pnpm playwright:smoke 2>&1)
SMOKE_STATUS=$?
set -e
echo "$SMOKE_OUTPUT" | tail -20
# Fail only on real failures (not flaky). Flaky = passed on retry.
if echo "$SMOKE_OUTPUT" | grep -qE '^\s+\d+ failed' && ! echo "$SMOKE_OUTPUT" | grep -qE '^\s+\d+ passed'; then
if [ "$SMOKE_STATUS" -ne 0 ]; then
echo " ❌ Smoke tests FAILED"
exit 1
exit "$SMOKE_STATUS"
fi
echo " ✅ Smoke tests OK"
else
@@ -112,10 +172,11 @@ fi
# ── 5. CodeScene code health gate (ratchet) ──────────────────────────────
# Thresholds live in .codescene-thresholds and only ever go UP (ratchet).
# After every successful push the file is updated if scores improved.
# If remote scores improved, the hook updates the file and stops so the new
# floor is committed with normal verified hooks before the next push.
THRESHOLDS_FILE="$(git rev-parse --show-toplevel)/.codescene-thresholds"
HOTSPOT_MIN=9.5
AVERAGE_MIN=9.31
HOTSPOT_MIN=9.45
AVERAGE_MIN=9.29
if [ -f "$THRESHOLDS_FILE" ]; then
HOTSPOT_MIN=$(grep HOTSPOT_THRESHOLD "$THRESHOLDS_FILE" | cut -d= -f2)
AVERAGE_MIN=$(grep AVERAGE_THRESHOLD "$THRESHOLDS_FILE" | cut -d= -f2)
@@ -137,8 +198,9 @@ else
else
echo " Remote Hotspot Code Health: $HOTSPOT_SCORE (threshold: $HOTSPOT_MIN)"
echo " Remote Average Code Health: $AVERAGE_SCORE (threshold: $AVERAGE_MIN)"
PYTHON_STATUS=0
python3 -c "
import sys, os
import sys
hotspot = float('$HOTSPOT_SCORE')
average = float('$AVERAGE_SCORE')
@@ -161,8 +223,6 @@ else:
if failed:
sys.exit(1)
# Ratchet: update thresholds file if scores improved
# Truncate (floor) to 2 decimals so the threshold never exceeds the actual score
import math
thresholds_file = '$THRESHOLDS_FILE'
new_hotspot = max(hotspot_min, math.floor(hotspot * 100) / 100)
@@ -171,9 +231,16 @@ if new_hotspot > hotspot_min or new_average > average_min:
with open(thresholds_file, 'w') as f:
f.write(f'HOTSPOT_THRESHOLD={new_hotspot}\nAVERAGE_THRESHOLD={new_average}\n')
print(f' 📈 Ratchet updated: Hotspot {hotspot_min} → {new_hotspot}, Average {average_min} → {new_average}')
# Stage and amend the thresholds update into the last commit would change history — instead just auto-commit it
os.system(f'git add {thresholds_file} && git commit -m \"chore: ratchet CodeScene thresholds to {new_hotspot}/{new_average}\" --no-verify 2>/dev/null || true')
" || exit 1
sys.exit(3)
" || PYTHON_STATUS=$?
if [ "$PYTHON_STATUS" -ne 0 ] && [ "$PYTHON_STATUS" -ne 3 ]; then
exit "$PYTHON_STATUS"
fi
if [ "$PYTHON_STATUS" -eq 3 ]; then
git add "$THRESHOLDS_FILE"
echo " ❌ Commit the updated .codescene-thresholds with a normal verified commit, then push again."
exit 1
fi
fi
fi

149
AGENTS.md Normal file
View File

@@ -0,0 +1,149 @@
# AGENTS.md — Laputa App
> Quick links: [Project Spec](docs/PROJECT-SPEC.md) · [Architecture](docs/ARCHITECTURE.md) · [Abstractions](docs/ABSTRACTIONS.md) · [Wireframes](ui-design.pen)
---
## 1. Task Workflow
### 1a. Pick up a task
Run `/laputa-next-task` — fetches next task (To Rework first, then Open), moves to In Progress, returns full description.
**Before writing a single line of code:** run `mcp__codescene__code_health_score` to check the current codebase health against `.codescene-thresholds`. If the score is already below the threshold, **stop and refactor first** — find the worst files with the MCP, improve them, commit, then start the task. Never start feature work on a codebase that is already below the gate.
- Read task description and all comments fully
- For To Rework: the ❌ QA failed comment tells you exactly what to fix
- Check `docs/adr/` for relevant architecture decisions before structural choices
- Add a comment: `🚀 Starting work on this task. [Brief description of approach]`
### 1b. Implement
- Work on `main` branch — **no branches, no PRs, ever**. Pre-commit and pre-push block work from any other branch.
- Commit every 2030 min: `feat:`, `fix:`, `refactor:`, `test:`, `docs:`
- **⛔ NEVER use --no-verify**
- For UI tasks: open `ui-design.pen` first, study visual language, design in light mode
### 1c. When done
**Phase 1 — Playwright (only for core user flows):**
Write smoke test in `tests/smoke/<slug>.spec.ts` only if feature touches: vault open, note create/save/delete, search, wikilink navigation, git commit/push, conflict resolution. Do NOT write Playwright tests for cosmetic changes — use Vitest instead. Suite must stay under **10 minutes**.
```bash
pnpm dev --port 5201 &
sleep 3
BASE_URL="http://localhost:5201" npx playwright test tests/smoke/<slug>.spec.ts
```
**Phase 2 — Native app QA:**
```bash
pnpm tauri dev &
sleep 10
bash ~/.openclaw/skills/laputa-qa/scripts/focus-app.sh laputa
bash ~/.openclaw/skills/laputa-qa/scripts/screenshot.sh /tmp/qa-native.png
```
Use `osascript` for keyboard interactions. Write result as Todoist comment (✅ or ❌). **⚠️ WKWebView:** `osascript keystroke` blocked inside editor — rely on Playwright for text input features.
After both phases pass, add a **completion comment** to the Todoist task before running `/laputa-done`. The comment must include:
- What was implemented (12 lines)
- QA: what was tested and how (Playwright / native screenshot / osascript)
- Refactoring: any files refactored to meet the CodeScene gate (or "none needed")
- ADRs: any new/updated ADRs (or "none")
- Docs: any updated docs (ARCHITECTURE.md, ABSTRACTIONS.md, etc.) (or "none")
- Code health: final Hotspot and Average scores after push
Then run `/laputa-done <task_id>` → moves to In Review, notifies Brian, self-dispatches next task.
---
## 2. Development Process
### Commits & pushes
- Push directly to `main` — no PRs, no branches. Pre-push blocks non-`main` pushes.
- Pre-push hook runs full check suite (build + tests + Playwright + CodeScene)
- **A task is NOT done until `git push origin main` succeeds.** If the hook blocks: read the error, fix it (clippy, tests, CodeScene, build), commit the fix, push again. **⛔ NEVER use --no-verify**
### TDD (mandatory)
Red → Green → Refactor → Commit. One cycle per commit. For bugs: write failing regression test first, then fix. Exception: pure CSS/layout changes.
**Test quality (Kent Beck's Desiderata):** Isolated · Deterministic · Fast · Behavioral · Structure-insensitive · Specific · Predictive. Fix flaky tests first. Prefer E2E over unit tests for user flows.
### Code health (mandatory)
Pre-commit and pre-push hooks enforce **Hotspot Code Health** and **Average Code Health** ≥ thresholds in `.codescene-thresholds`. Both gates block commit/push. Thresholds are a **ratchet** — only go up. When pre-push sees improved remote scores, it updates `.codescene-thresholds`, stages it, and stops so you can commit the new floor with normal verified hooks before pushing again. Never add `// eslint-disable`, `#[allow(...)]`, or `as any`.
**⛔ NEVER edit `.codescene-thresholds` to lower the values.** If the gate blocks you, improve the code — do not lower the bar.
**Before every commit:** run `mcp__codescene__code_health_review` on files you touched and verify score is higher. **Boy Scout Rule:** every file you touch must leave with a higher score.
**If CodeScene gate blocks your push:** use `mcp__codescene__code_health_score` to find the worst file, refactor it, commit, push again. Do NOT stop or wait for laputa-refactor — that is a background loop, not a substitute for fixing your own regressions.
### Check suite (runs on every push)
```bash
pnpm lint && npx tsc --noEmit && pnpm test && pnpm test:coverage # frontend ≥70%
cargo test && cargo llvm-cov --manifest-path src-tauri/Cargo.toml --no-clean --fail-under-lines 85
```
### ADRs & docs
ADRs live in `docs/adr/`. Create in the same commit as the code. Never edit existing — create a new one that supersedes. Use `/create-adr`. **When:** new dependency, storage strategy, platform target, core abstraction, cross-cutting pattern. **Not for:** bug fixes, styling, refactors.
After any Tauri command, new component/hook, data model change, or new integration: update `docs/ARCHITECTURE.md`, `docs/ABSTRACTIONS.md`, and/or `docs/GETTING-STARTED.md` in the same commit.
---
## 3. Product Rules
### User vault (`~/Laputa/`)
Default to `demo-vault-v2/`. If you must use `~/Laputa/` for testing: **never commit changes** — always run `cd ~/Laputa && git checkout -- . && git clean -fd` when done.
### UI design
Open `ui-design.pen` first (light mode). Create `design/<slug>.pen` for the task; on completion merge into `ui-design.pen` and delete it.
### UI components — mandatory rules
**Always use shadcn/ui components.** Never use raw HTML form elements (`<input>`, `<select>`, `<button>`, native `<input type="date">`, etc.) for user-facing UI. Every interactive element must use the shadcn/ui equivalent:
| Need | Use |
|---|---|
| Text input | `Input` from shadcn/ui |
| Dropdown/select | `Select` from shadcn/ui |
| Date picker | `Calendar` + `Popover` from shadcn/ui (NOT native `<input type="date">`) |
| Button | `Button` from shadcn/ui |
| Autocomplete/combobox | Reuse existing combobox components from the app (check `src/components/`) |
| Wikilink picker | Reuse the wikilink autocomplete component already used in the editor and Properties panel |
| Emoji picker | Reuse the emoji picker component already used for note/type icons |
| Color picker | Reuse the color swatch picker used for type customization |
| Toggle/switch | `Switch` or `ToggleGroup` from shadcn/ui |
| Dialog/modal | `Dialog` from shadcn/ui |
**When in doubt:** search `src/components/` for an existing component before building new. **Visual language:** all new UI must feel native to Laputa — if it looks like a browser default, it's wrong.
---
## 4. Reference
### macOS / Tauri gotchas
- `Option+N` → special chars on macOS. Use `e.code` or `Cmd+N`
- Tauri menu accelerators: `MenuItemBuilder::new(label).accelerator("CmdOrCtrl+1")`
- `app.set_menu()` replaces the ENTIRE menu bar — include all submenus
- `mock-tauri.ts` silently swallows Tauri calls — not a substitute for native testing
### QA scripts
```bash
bash ~/.openclaw/skills/laputa-qa/scripts/focus-app.sh laputa
bash ~/.openclaw/skills/laputa-qa/scripts/screenshot.sh /tmp/out.png
bash ~/.openclaw/skills/laputa-qa/scripts/shortcut.sh "command" "s"
```
### Diagrams
Prefer Mermaid (`flowchart`, `sequenceDiagram`, `classDiagram`, `stateDiagram-v2`). ASCII only for spatial wireframe layouts.

150
CLAUDE.md
View File

@@ -1,149 +1,3 @@
# CLAUDE.md — Laputa App
@AGENTS.md
> Quick links: [Project Spec](docs/PROJECT-SPEC.md) · [Architecture](docs/ARCHITECTURE.md) · [Abstractions](docs/ABSTRACTIONS.md) · [Wireframes](ui-design.pen)
---
## 1. Task Workflow
### 1a. Pick up a task
Run `/laputa-next-task` — fetches next task (To Rework first, then Open), moves to In Progress, returns full description.
**Before writing a single line of code:** run `mcp__codescene__code_health_score` to check the current codebase health against `.codescene-thresholds`. If the score is already below the threshold, **stop and refactor first** — find the worst files with the MCP, improve them, commit, then start the task. Never start feature work on a codebase that is already below the gate.
- Read task description and all comments fully
- For To Rework: the ❌ QA failed comment tells you exactly what to fix
- Check `docs/adr/` for relevant architecture decisions before structural choices
- Add a comment: `🚀 Starting work on this task. [Brief description of approach]`
### 1b. Implement
- Work on `main` branch — **no branches, no PRs, ever**
- Commit every 2030 min: `feat:`, `fix:`, `refactor:`, `test:`, `docs:`
- **⛔ NEVER use --no-verify**
- For UI tasks: open `ui-design.pen` first, study visual language, design in light mode
### 1c. When done
**Phase 1 — Playwright (only for core user flows):**
Write smoke test in `tests/smoke/<slug>.spec.ts` only if feature touches: vault open, note create/save/delete, search, wikilink navigation, git commit/push, conflict resolution. Do NOT write Playwright tests for cosmetic changes — use Vitest instead. Suite must stay under **10 minutes**.
```bash
pnpm dev --port 5201 &
sleep 3
BASE_URL="http://localhost:5201" npx playwright test tests/smoke/<slug>.spec.ts
```
**Phase 2 — Native app QA:**
```bash
pnpm tauri dev &
sleep 10
bash ~/.openclaw/skills/laputa-qa/scripts/focus-app.sh laputa
bash ~/.openclaw/skills/laputa-qa/scripts/screenshot.sh /tmp/qa-native.png
```
Use `osascript` for keyboard interactions. Write result as Todoist comment (✅ or ❌). **⚠️ WKWebView:** `osascript keystroke` blocked inside editor — rely on Playwright for text input features.
After both phases pass, add a **completion comment** to the Todoist task before running `/laputa-done`. The comment must include:
- What was implemented (12 lines)
- QA: what was tested and how (Playwright / native screenshot / osascript)
- Refactoring: any files refactored to meet the CodeScene gate (or "none needed")
- ADRs: any new/updated ADRs (or "none")
- Docs: any updated docs (ARCHITECTURE.md, ABSTRACTIONS.md, etc.) (or "none")
- Code health: final Hotspot and Average scores after push
Then run `/laputa-done <task_id>` → moves to In Review, notifies Brian, self-dispatches next task.
---
## 2. Development Process
### Commits & pushes
- Push directly to `main` — no PRs, no branches
- Pre-push hook runs full check suite (build + tests + Playwright + CodeScene)
- **A task is NOT done until `git push origin main` succeeds.** If the hook blocks: read the error, fix it (clippy, tests, CodeScene, build), commit the fix, push again. **⛔ NEVER use --no-verify**
### TDD (mandatory)
Red → Green → Refactor → Commit. One cycle per commit. For bugs: write failing regression test first, then fix. Exception: pure CSS/layout changes.
**Test quality (Kent Beck's Desiderata):** Isolated · Deterministic · Fast · Behavioral · Structure-insensitive · Specific · Predictive. Fix flaky tests first. Prefer E2E over unit tests for user flows.
### Code health (mandatory)
Pre-commit and pre-push hooks enforce **Hotspot Code Health** and **Average Code Health** ≥ thresholds in `.codescene-thresholds`. Both gates block commit/push. Thresholds are a **ratchet** — only go up, auto-updated after each successful push. Never add `// eslint-disable`, `#[allow(...)]`, or `as any`.
**⛔ NEVER edit `.codescene-thresholds` to lower the values.** If the gate blocks you, improve the code — do not lower the bar.
**Before every commit:** run `mcp__codescene__code_health_review` on files you touched and verify score is higher. **Boy Scout Rule:** every file you touch must leave with a higher score.
**If CodeScene gate blocks your push:** use `mcp__codescene__code_health_score` to find the worst file, refactor it, commit, push again. Do NOT stop or wait for laputa-refactor — that is a background loop, not a substitute for fixing your own regressions.
### Check suite (runs on every push)
```bash
pnpm lint && npx tsc --noEmit && pnpm test && pnpm test:coverage # frontend ≥70%
cargo test && cargo llvm-cov --manifest-path src-tauri/Cargo.toml --no-clean --fail-under-lines 85
```
### ADRs & docs
ADRs live in `docs/adr/`. Create in the same commit as the code. Never edit existing — create a new one that supersedes. Use `/create-adr`. **When:** new dependency, storage strategy, platform target, core abstraction, cross-cutting pattern. **Not for:** bug fixes, styling, refactors.
After any Tauri command, new component/hook, data model change, or new integration: update `docs/ARCHITECTURE.md`, `docs/ABSTRACTIONS.md`, and/or `docs/GETTING-STARTED.md` in the same commit.
---
## 3. Product Rules
### User vault (`~/Laputa/`)
Default to `demo-vault-v2/`. If you must use `~/Laputa/` for testing: **never commit changes** — always run `cd ~/Laputa && git checkout -- . && git clean -fd` when done.
### UI design
Open `ui-design.pen` first (light mode). Create `design/<slug>.pen` for the task; on completion merge into `ui-design.pen` and delete it.
### UI components — mandatory rules
**Always use shadcn/ui components.** Never use raw HTML form elements (`<input>`, `<select>`, `<button>`, native `<input type="date">`, etc.) for user-facing UI. Every interactive element must use the shadcn/ui equivalent:
| Need | Use |
|---|---|
| Text input | `Input` from shadcn/ui |
| Dropdown/select | `Select` from shadcn/ui |
| Date picker | `Calendar` + `Popover` from shadcn/ui (NOT native `<input type="date">`) |
| Button | `Button` from shadcn/ui |
| Autocomplete/combobox | Reuse existing combobox components from the app (check `src/components/`) |
| Wikilink picker | Reuse the wikilink autocomplete component already used in the editor and Properties panel |
| Emoji picker | Reuse the emoji picker component already used for note/type icons |
| Color picker | Reuse the color swatch picker used for type customization |
| Toggle/switch | `Switch` or `ToggleGroup` from shadcn/ui |
| Dialog/modal | `Dialog` from shadcn/ui |
**When in doubt:** search `src/components/` for an existing component before building new. **Visual language:** all new UI must feel native to Laputa — if it looks like a browser default, it's wrong.
---
## 4. Reference
### macOS / Tauri gotchas
- `Option+N` → special chars on macOS. Use `e.code` or `Cmd+N`
- Tauri menu accelerators: `MenuItemBuilder::new(label).accelerator("CmdOrCtrl+1")`
- `app.set_menu()` replaces the ENTIRE menu bar — include all submenus
- `mock-tauri.ts` silently swallows Tauri calls — not a substitute for native testing
### QA scripts
```bash
bash ~/.openclaw/skills/laputa-qa/scripts/focus-app.sh laputa
bash ~/.openclaw/skills/laputa-qa/scripts/screenshot.sh /tmp/out.png
bash ~/.openclaw/skills/laputa-qa/scripts/shortcut.sh "command" "s"
```
### Diagrams
Prefer Mermaid (`flowchart`, `sequenceDiagram`, `classDiagram`, `stateDiagram-v2`). ASCII only for spatial wireframe layouts.
This file is a Claude Code compatibility shim. Keep shared agent instructions in `AGENTS.md`.

View File

@@ -24,9 +24,6 @@ Personal knowledge and life management desktop app built with Tauri v2 + React +
# Install dependencies
pnpm install
# Install git hooks (optional but recommended)
.github/hooks/install-hooks.sh
# Run dev server
pnpm dev
@@ -69,7 +66,7 @@ claude 'Check code health with CodeScene MCP'
## Development Workflow
See [CLAUDE.md](CLAUDE.md) for coding guidelines and workflow.
See [AGENTS.md](AGENTS.md) for coding guidelines and workflow. [CLAUDE.md](CLAUDE.md) remains as a compatibility shim for Claude Code.
**Key principles:**
- Small, atomic commits
@@ -79,7 +76,7 @@ See [CLAUDE.md](CLAUDE.md) for coding guidelines and workflow.
## CI/CD
GitHub Actions runs on every push/PR:
GitHub Actions runs on every push to `main`:
- ✅ Tests (frontend + Rust)
- 📊 Coverage (70% threshold)
- 🎨 Lint & format
@@ -89,7 +86,7 @@ See [.github/SETUP.md](.github/SETUP.md) for CI/CD configuration.
## Git Hooks
Pre-commit hook checks code health before every commit. See [.github/HOOKS.md](.github/HOOKS.md) for details.
Husky installs the git hooks from `.husky/` during `pnpm install`. The repo enforces `main`-only commits and pushes; see [.github/HOOKS.md](.github/HOOKS.md) for details.
## License

View File

@@ -177,7 +177,8 @@ laputa-app/
├── tsconfig.json # TypeScript config
├── playwright.config.ts # E2E test config
├── ui-design.pen # Master design file
├── CLAUDE.md # Project instructions
├── AGENTS.md # Shared project instructions for coding agents
├── CLAUDE.md # Claude Code compatibility shim importing AGENTS.md
└── docs/ # This documentation
```

View File

@@ -0,0 +1,81 @@
# AGENTS.md — Laputa Vault
This is a [Laputa](https://github.com/refactoringhq/laputa-app) vault — a folder of markdown files with YAML frontmatter forming a personal knowledge graph.
## Note structure
Every note is a markdown file. The **first H1 heading in the body is the title** — there is no `title:` frontmatter field.
```yaml
---
is_a: TypeName # the note's type (must match the title of a type file in the vault)
url: https://... # example property
belongs_to: "[[other-note]]"
related_to:
- "[[note-a]]"
- "[[note-b]]"
---
# Note Title
Body content in markdown.
```
System properties are prefixed with `_` (e.g. `_organized`, `_pinned`, `_icon`) — these are app-managed, do not set or show them to users unless specifically asked.
## Types
A type is a note with `is_a: Type`. Type files live in the vault root:
```yaml
---
is_a: Type
_icon: books # Phosphor icon name in kebab-case
_color: "#8b5cf6" # hex color
---
# TypeName
```
To find what types exist: look for files with `is_a: Type` in the vault root.
## Relationships
Any frontmatter property whose value is a wikilink is a relationship. Backlinks are computed automatically.
Standard names: `belongs_to`, `related_to`, `has`. Custom names are valid.
## Wikilinks
- `[[filename]]` or `[[Note Title]]` — link by filename or title
- `[[filename|display text]]` — with custom display text
- Works in frontmatter values and markdown body
## Views
Saved filters live in `views/` as `.view.json` files:
```json
{
"title": "Active Notes",
"filters": [
{"property": "is_a", "operator": "equals", "value": "Note"},
{"property": "status", "operator": "equals", "value": "Active"}
],
"sort": {"property": "title", "direction": "asc"}
}
```
## Filenames
Use kebab-case: `my-note-title.md`. One note per file.
## What you can do
- Create/edit notes with correct frontmatter and H1 title
- Create new type files
- Add or modify relationships
- Create/edit views in `views/`
- Edit `AGENTS.md` (this file)
Do not modify app configuration files — those are local to each installation.

View File

@@ -1,81 +1,3 @@
# CLAUDE.md — Laputa Vault
@AGENTS.md
This is a [Laputa](https://github.com/refactoringhq/laputa-app) vault — a folder of markdown files with YAML frontmatter forming a personal knowledge graph.
## Note structure
Every note is a markdown file. The **first H1 heading in the body is the title** — there is no `title:` frontmatter field.
```yaml
---
is_a: TypeName # the note's type (must match the title of a type file in the vault)
url: https://... # example property
belongs_to: "[[other-note]]"
related_to:
- "[[note-a]]"
- "[[note-b]]"
---
# Note Title
Body content in markdown.
```
System properties are prefixed with `_` (e.g. `_organized`, `_pinned`, `_icon`) — these are app-managed, do not set or show them to users unless specifically asked.
## Types
A type is a note with `is_a: Type`. Type files live in the vault root:
```yaml
---
is_a: Type
_icon: books # Phosphor icon name in kebab-case
_color: "#8b5cf6" # hex color
---
# TypeName
```
To find what types exist: look for files with `is_a: Type` in the vault root.
## Relationships
Any frontmatter property whose value is a wikilink is a relationship. Backlinks are computed automatically.
Standard names: `belongs_to`, `related_to`, `has`. Custom names are valid.
## Wikilinks
- `[[filename]]` or `[[Note Title]]` — link by filename or title
- `[[filename|display text]]` — with custom display text
- Works in frontmatter values and markdown body
## Views
Saved filters live in `views/` as `.view.json` files:
```json
{
"title": "Active Notes",
"filters": [
{"property": "is_a", "operator": "equals", "value": "Note"},
{"property": "status", "operator": "equals", "value": "Active"}
],
"sort": {"property": "title", "direction": "asc"}
}
```
## Filenames
Use kebab-case: `my-note-title.md`. One note per file.
## What you can do
- Create/edit notes with correct frontmatter and H1 title
- Create new type files
- Add or modify relationships
- Create/edit views in `views/`
- Edit `CLAUDE.md` (this file)
Do not modify app configuration files — those are local to each installation.
This file is a Claude Code compatibility shim. Keep shared vault instructions in `AGENTS.md`.

View File

@@ -12,7 +12,7 @@ Laputa integrates with [Claude Code](https://docs.anthropic.com/claude-code) —
claude "Create a note for the book Zero to One by Peter Thiel, with a rating and a topic"
```
Claude understands Laputa's format (frontmatter, types, wikilinks, relationships) and creates or edits files accordingly. Your vault's `CLAUDE.md` file gives it full context.
Claude understands Laputa's format (frontmatter, types, wikilinks, relationships) and creates or edits files accordingly. Your vault's `AGENTS.md` file gives any coding agent full context, and `CLAUDE.md` imports it for Claude Code compatibility.
## Git sync