livef12rocks/AGENTS.md

# AGENTS.md — Operational Instructions

How the coding agent works with Fränky's projects.

Also read: `INFRASTRUCTURE.md`, `STANDARDS.md`, `BOUNDARIES.md`, project `SOUL.md`, project `MOOD.md`.  
Persona & preferences: `USER.md` (Tier 2).

---

## Rule Priority

Highest wins. One line per source:

| Priority | Source |
|----------|--------|
| 1 | Explicit chat instruction from Fränky |
| 2 | Project `SOUL.md` / project-local `BOUNDARIES.md` / project `AGENTS` snippets |
| 3 | `BOUNDARIES.md` |
| 4 | `AGENTS.md` (this file) |
| 5 | `STANDARDS.md` |
| 6 | `INFRASTRUCTURE.md` |
| 7 | `MOOD.md` — **tone only**, never overrides safety or ops rules |
| 8 | `USER.md` — persona & preferences, not operational overrides |

**UI/UX:** project `SOUL.md` overrides global `STANDARDS.md` when they conflict.

---

## Startup (Tier 1 / Tier 2)

Cursor does not auto-load context. Use two tiers:

### Tier 1 — always (light)

At the start of any substantive task, without waiting for ack:

1. Project `SOUL.md` — **Agent Quick Start** section (or full file if no Quick Start)
2. `BOUNDARIES.md` — **Never Ever** section

### Tier 2 — full load

On `@AGENTS.md ack`, first chat in a project, or when Fränky says context was lost:

1. `AGENTS.md`, `BOUNDARIES.md`, `STANDARDS.md`, `INFRASTRUCTURE.md`
2. `USER.md` — persona, work-style table, communication prefs
3. Project `SOUL.md` (full), project `MOOD.md` if present
4. Skim project structure

### Command: `AGENTS.md ack`

Fränky schreibt `AGENTS.md ack`.

**Agent:** Tier-2 read, then reply with this **compact template** (one block, no code changes):

```
AGENTS ack ✓
· Pipeline: [1-line what this project is]
· Non-goals: [from SOUL, or "SOUL missing"]
· Commit policy: [no commit unless … / project override]
· MOOD: [active session character]
· Paths: [key workdirs / mounts from SOUL or INFRA snippet]
· Conflict: [1 sentence if SOUL vs STANDARDS disagree, else "none"]
```

---

## Work Modes

| Mode | Trigger | Behavior |
|------|---------|----------|
| **Question-only** | Question, review, "how does X work?" | No file changes, commit, or drive-by fixes |
| **Standard** | Default | Restate → align on plan → implement → verify → hand off |
| **Unattended / Away** | "wenn ich zurück bin", "mach ohne mich", "overnight", explicit away | **Skip plan alignment** — proceed with best judgment; justify in handoff |

**Plan Mode (Cursor):** If file edits are blocked, **abort and wait** — never circumvent via shell. See `BOUNDARIES.md`.

### Unattended / Away — extra rules

- **Monitoring:** Cursor background shell + polling — **not** external wrapper scripts as default
- **Handoff must include:** log paths, PIDs if relevant, how to recognize success, next command for Fränky
- Write blocked items or run status to `NOTES.md` when useful
- **Notify** via `ntfyschwenkonline` when the task finishes (see [Notifications](#notifications))

---

## Default Workflow (Standard mode)

1. **Restate** the request; turn into plan or mini-PRD
2. **Align** on the plan — resolve ambiguities before coding
3. **Implement** (see Testing below)
4. **Verify** — run tests when appropriate; do not guess
5. **Hand off** — summary, how to test locally, log paths for long jobs
6. **Deploy / prod** — Fränky handles unless project or chat says otherwise

After **long-running** work: send push notification (see [Notifications](#notifications)).

---

## Autonomy Matrix

See also `USER.md` **Work Style** table for Fränky's preference scores.

| Action | Default |
|--------|---------|
| Write / change code | ✅ OK |
| Write tests | ✅ OK when non-trivial or suite exists |
| Add dependencies | ✅ OK |
| Touch README / docs | ✅ OK |
| Change CI/CD config | ⚠️ Ask first |
| Refactor "on the side" | ⚠️ Ask first |
| Update AGENTS / SOUL / BOUNDARIES | ⚠️ Ask first (or explicit "remember this") |
| Update `MOOD.md` on disk | 🚫 Only on `persist MOOD` or explicit instruction |
| Create auxiliary `.md` (`NOTES.md`, todos) | ✅ OK |
| Git commit / push / deploy | 🚫 Unless explicit or project `SOUL.md` allows |
| `ntfyschwenkonline` after long-running task | ✅ OK |

---

## MOOD: Session vs. Persist

| | Session | Persist to `MOOD.md` |
|---|---------|----------------------|
| Trigger | `mood "XY"` in chat | `persist MOOD` or explicit "save mood to file" |
| Effect | Tone for this chat only | Updates `## Aktueller Mood` in file |
| Default | **Yes** — `mood "XY"` does **not** edit the file |

---

## Testing

Fränky's bar is **pragmatic, not TDD-by-default** (`USER.md`: tests score 2/5).

- Run tests when a suite exists **and** the change is non-trivial
- Follow project `SOUL.md` if stricter (e.g. pytest before handoff)
- Do not block small fixes on missing test infrastructure

### Verification (pragmatic)

- If the project has a runnable stack (Compose, CLI, service): prefer a **real smoke / happy-path** before handoff — not “code looks fine”
- Waiting on slow builds or pipeline jobs is OK and preferred over fake-green
- Unit/integration tests still apply when a suite exists; smoke does not replace them, and they do not replace smoke when infra is the product

---

## Commits & Issues

- **Default: no commit, no push** without explicit instruction
- **Format:** [Conventional Commits](https://www.conventionalcommits.org/) (`feat:`, `fix:`, `docs:`, …) when committing
- **Forge:** **Gitea** (`gitea.schwenk.online`) unless project says otherwise — not GitLab
- Link Gitea issues when they exist (`#123`)
- Do not mix unrelated changes (e.g. MOOD switch + feature) in one commit

---

## Collaboration Patterns

### `NOTES.md`

Scratchpad — use for blocked tasks, session park, overnight-run status, handoff crumbs.

### Subagents & context budget

**Keep main context small.** Delegate instead of dumping large outputs into parent chat:

| Task | Subagent |
|------|----------|
| Unknown codebase or many files to scan | **Explore** |
| Greenfield / large feature implementation | **generalPurpose** (parent keeps orchestrating deploy/ntfy) |
| Browser / E2E / web UI | **Playwright** |
| Broad research | **generalPurpose** or **Explore** |
| Shell / git / sync / remote compose | **shell** |

**Explore** for any project type when orientation is unclear — not only web. **Playwright** only when there is a web UI to test.

### Commits when allowed

One focused commit per completed task — easy to revert. No drive-by refactors.

---

## Project Infrastructure Snippet

Machine-specific paths (workdirs, external mounts, local service ports) belong in **project `SOUL.md`**, not global `INFRASTRUCTURE.md`.

Example SOUL section:

```markdown
## Infrastructure (project-local)
- Workdir: ~/.local/share/myapp
- Mounts: check MegaB before scan
- Local Immich: :2283 (downstream, not core)
```

See `INFRASTRUCTURE.md` → **Project-local overrides**.

---

## When Uncertain

1. Research — read code, run commands
2. Continue other independent tasks if possible
3. Park in `NOTES.md`, todo file, or Gitea issue
4. Ask after research, with options + recommended default

---

## Error Handling

- Root cause first — do not guess
- **Debug loop:** same failed hypothesis/fix tried **more than 2–3 times** → change strategy or escalate. A single deliberate course correction (wrong package → correct image, rebuild) is not a loop — note it and continue
- Do not paper over broken tool/dependency versions with hacks when the real fix is a rebuild or version pin
- Escalate on: debug loops, missing uninstallable software
- Long-running jobs: no arbitrary timeout unless Fränky or SOUL says so; name `progress.log` when applicable

### Away / long-task handoff

After Unattended work (or any task that took wall-clock minutes): in chat and/or ntfy, **one short line on where the time went** (e.g. image build, E2E wait, one root-cause rebuild) — no essay.

---

## Notifications

Fränky's machines have **`/usr/local/bin/ntfyschwenkonline`** — push to [ntfy.schwenk.online](https://ntfy.schwenk.online). Topic = short hostname (`hostname -s`); phone subscribes per machine. Auth is in the installed script — **never copy tokens into repos or chat**.

### When to notify

Send a notification when a **long-running agent task** completes or fails:

- **Unattended / Away** mode — always on finish (success or failure)
- Background shell jobs you started (scans, builds, batch ops, overnight runs)
- Any task Fränky left with an expectation of „meld dich wenn fertig"
- Rough guide: expected runtime **> ~2 minutes** or explicit away/unattended context

Do **not** notify for quick edits, short test runs, or question-only chats.

### How

```bash
ntfyschwenkonline "OK <project>: <one-line result>"
# or on failure:
ntfyschwenkonline "FAIL <project>: <one-line error>, see <log-path>"
```

- **English** message body (CLI/ops convention)
- One line, ~120 chars — project name, outcome, log path or next step if relevant
- Run after handoff summary; notification is in addition to chat handoff, not a replacement
- If `ntfyschwenkonline` is missing or fails: note in chat handoff, do not block

### Examples

```bash
ntfyschwenkonline "OK imagetool: scan done 142 dirs, log ~/.local/share/imagetool/progress.log"
ntfyschwenkonline "FAIL ytrecap: pytest 3 failed, see /tmp/test.log"
ntfyschwenkonline "OK schwenkonline: build+deploy done"
```

---

## Recording New Rules

When Fränky says *"never do X"*:

1. Propose file (`BOUNDARIES.md`, `SOUL.md`, `AGENTS.md`, or multiple)
2. Store in **Cursor Memories** for explicit rules
3. Write after confirmation — or immediately if explicit

---

## Language

- **Chat:** match Fränky's language (German or English)
- **CLI / terminal / shell:** always English (unless client project — ask)
- **Other artifacts:** first language of session

---

## Tools & MCP

1. MCP when available and relevant
2. Built-in tools (shell, grep, read)
3. Browser automation last resort

Check MCP schemas before calling.

---

## Tooling Context

Primary: **Cursor** (Composer / Agent).  
Server/deploy: **`INFRASTRUCTURE.md`**. Persona: **`USER.md`**.