auto-init

name	auto-init
description	Auto-checks project context, logs activity via scribe, and manages the activity log and artifact directories. Loaded by Optimus and Megatron on every command.
user-invocable	false

Project Context Check

Before starting any work:

1. Does .claude/transformers/context/project.md exist?

   • If NO → tell the user: "This project hasn't been initialized for Transformers yet. Run /transformers:init first for best results, or I can proceed without project context."
   • If YES → load it silently and proceed.

2. Is it stale? Check the "Generated on" date in the file.

   • If older than 24 hours → mention once: "Project context is from [date]. Run /transformers:init --update to refresh, or I'll work with what we have."
   • Don't block — just mention it.

3. Does .claude/transformers/context/skills-registry.md exist?

   • If YES → load it silently. Pass the available skills and local agents as context to the orchestrator.
   • If NO → check .claude/transformers/context/init-prompted.md for a last-prompted date.

     • If last-prompted is today's date → skip silently and proceed.

     • Otherwise → write today's date to init-prompted.md (create if missing), then tell the user:

       "Transformers hasn't been initialized for this project yet. Running /transformers:init once analyzes your project's architecture, tech stack, and git workflow — so every agent starts informed instead of exploring from scratch. It also discovers any skills and local agents you have installed. It's a one-time setup (refresh anytime with /transformers:init --update). Want me to run it now, or continue without it?"

     • If user says yes → invoke /transformers:init via the Skill tool, then continue.

     • If user says no → continue. Won't ask again until tomorrow.

Memory Check

Before starting work, load project memory:

1. Skills & agents registry — already loaded in Project Context Check above (step 3). If loaded, keep it in context — orchestrators will consult it when routing tasks.

2. Long-term memory index — read .claude/transformers/memory/long-term/index.md if it exists

   • Silently absorb relevant learnings. Don't announce them unless directly relevant to the current task.
   • If a learning is relevant, read the full category file for details.

3. User patterns — always read .claude/transformers/memory/long-term/user-patterns.md in full if it exists

   • These are behavioral preferences learned from past corrections and redirects. Apply them immediately.
   • Don't announce them — just use them.

4. Temp memory — read .claude/transformers/memory/temp.md if it exists

   • These are learnings from earlier in this conversation. Apply them.

If none exist → no memory yet. Proceed normally. Memory will grow as agents work.

Artifact Directory Check

For lifecycle commands (feature, bugfix, research):

1. Prune stale .temp/ entries — spawn Scribe to delete any subdirectory under .claude/transformers/.temp/ whose status.md has a timestamp older than 2 days. If no status.md exists in a folder, delete it too.

2. Check for in-progress work in .claude/transformers/.temp/

   • If a matching artifact directory exists with status.md showing incomplete work → "Found in-progress work for [name]. Want to resume or start fresh?"
   • If starting fresh with a name collision → append a number (features/search-2)

3. Ensure directory structure exists:

.claude/transformers/
├── context/
│   ├── project.md        ← from /transformers:init
│   └── pr-preferences.md ← from /transformers:pr-generator
├── memory/
│   ├── temp.md           ← per-conversation learnings
│   └── long-term/        ← cross-conversation learnings
│       ├── index.md      ← one-line summaries (loaded on every command)
│       └── [category].md ← detailed learnings by topic
├── reports/              ← from /transformers:report
│   └── activity.log      ← scribe appends after every command
└── .temp/                ← in-progress and completed feature/bugfix artifacts
    ├── features/
    └── bugfix/

Activity Logging

After completing any command, spawn scribe to log what was done.

Tell scribe exactly what to write — one line in this format:

YYYY-MM-DD HH:MM [command] [brief description] [files touched count] [Xk tokens, N agents]

The token/agent count is optional for standalone commands but required for lifecycle commands (feature, bugfix) and test.

Example instructions to scribe:

• "Append to .claude/transformers/reports/activity.log: 2026-03-16 14:30 feature Built dark mode toggle — 4 files changed [85k tokens, 6 agents]"
• "Append to .claude/transformers/reports/activity.log: 2026-03-16 16:00 bugfix Fixed null pointer in auth flow — 2 files changed [42k tokens, 4 agents]"
• "Append to .claude/transformers/reports/activity.log: 2026-03-16 17:00 debug Fixed typo in config — 1 file changed"

Scribe handles directory creation, appending, and pruning entries older than 7 days.