new-co-trader-worktree

Files (1)

• SKILL.md

SKILL.md (first lines)

# new-co-trader-worktree

Creates a new git worktree for auto-co-trader work (optimization, regression,
backtesting, brainstorming, or any other purpose) and optionally seeds it with
a named starting strategy.

---

## Step 0: Branch Context Check

Before doing anything else, check which branch you are on:

```bash
git branch --show-current
```

### Case A — On a non-master branch

If the current branch is **not** `master`, stop and tell the user:

> You're on branch `<current-branch>`, not `master`. Worktree creation should be
> done from the `master` branch so each worktree starts from a clean baseline.
>
> Would you like to switch to master first, or proceed anyway using this branch
> as the base?

Do not proceed further until the user confirms.

### Case B — On master

This is the normal flow. Continue to Step 1 below.

---

## Step 1: Identify Starting Strategy

Check whether the user named a starting strategy in their message:

- **Named** (e.g. "start from energy_momentum_v1", "use financials_mar20"):
  - Locate `strategies/<name>.py` in the current repo.
  - Its functions will be injected into train.py in the worktree (see Step 4).
- **Not named (default)**: Use train.py as-is — skip Step 4.

---

## Step 2: Determine Branch Name

The branch name needs a descriptive prefix so the worktree folder is self-explanatory
(e.g. `energy-mar21`, `regression-apr05`, `backtest-may08`). Derive it from, in order:

1. **The user's message** — if it mentions a sector, theme, purpose, or set of tickers
   (e.g. "energy stocks", "semiconductors", "mag7", "regression run", "backtesting"),
   use that as the prefix.
2. **The named starting strategy** — if the strategy name implies a sector
   (e.g. `energy_momentum_v1` → `energy`, `semis_mar20` → `semis`), use that.
3. **Neither** — if there is no usable context, ask the user before proceeding:
   > To name the worktree, can you briefly describe what it's for?
   > (e.g. "energy stocks", "regression run", "backtesting mag7") — no need to list
   > exact tickers.

Once you have a prefix, the tag is `<prefix>-<month><day>` (e.g. `energy-mar21`).

Check for collision:
```bash
git branch -a | grep "autoresearch/<tag>"
```
If it exists, append a suffix: `energy-mar21-b`, etc.

Full branch name: `autoresearch/<tag>`.

---

## Step 3: Create Worktree

Get the current branch:
```bash
git branch --show-current
```

Use the `create-worktree` skill:
- **Branch name**: `autoresearch/<tag>`
- **Base branch**: the current branch (as above)

The worktree lands at `../<tag>`.

> All subsequent file edits go into the new worktree, not the current repo.

---

## Step 3.5: Copy Cached Data

After the worktree is created, copy any cached regression data so the new worktree
can use it without re-downloading or re-computing.

> **Note:** The main `data/*.parquet` bar files are NOT copied. They live in the
> shared global path (`~/projects/auto-co-trader/global/...`) and all worktrees
> read from there — copying them per-worktree is no longer necessary.

### Copy data/regression/<date> folders

```bash
if [ -d "data/regression" ]; then
  mkdir -p "../<tag>/data/regression"
  for dir in data/regression/*/; do
    cp -r "$dir" "../<tag>/$dir"
  done
  echo "Copied data/regression date folders"
fi
```

If the directory does not exist, skip silently — the worktree will fetch or
generate data as needed.

---

## Step 4: Inject Starting Strategy (only if user named one)

*Skip if using the default train.py.*

From giladresisi/ai-dev-env, MIT.

More from giladresisi/ai-dev-env