Continue Cycle

# Continue Prompt: Run One Project Atlas Batch

Run exactly one batch, then stop with a concise status summary.

## Lean context rule

Do not reload the whole project history.

Read only:

- `CLAUDE.md`
- `NEXT_ASSIGNMENT.md`
- `status/WORKFLOW_STATUS.md`
- `docs/_project_index/PROJECT_FILE_INDEX.md`
- the latest non-template assignment/report/verification only if needed to confirm the previous batch passed
- the source files assigned to this batch

Do not read all previous assignments, all worker reports, all verification reports, or all architecture files during a normal continue cycle. Older reports are for repair cycles only.

If the previous batch is missing a required worker report, verification report, or mirror docs, do not start a new batch. Use `prompts/03_repair_batch_consistency.md` instead.

## Minified JavaScript / Packed File Warning

Before selecting, repairing, or processing a batch, check whether any `.js`, `.json`, `.css`, `.html`, `.mjs`, `.cjs`, or `.ts` files are minified/packed.

Do not treat a file as tiny just because it has `0`, `1`, or very few physical lines.

A file is probably minified/packed if:

- it has fewer than 20 physical lines, AND
- it is larger than 8KB, AND
- it is JavaScript, JSON, CSS, HTML, TypeScript, or similar text code/data.

For one-line/minified JavaScript command files:

- 1KB-5KB: small
- 5KB-15KB: medium
- 15KB-50KB: large
- 50KB+: very large

Batch limits for minified/packed files:

- mostly 1KB-5KB files: up to 20 files
- mostly 5KB-15KB files: 8-12 files
- mostly 15KB-50KB files: 2-5 files
- any file over 50KB: 1 file, or only with closely related tiny files

If most files are 5KB-15KB minified JS, use 8-12 files per batch.

Never assign 20-30 minified JavaScript files just because line count is `0`.

## Step 1: Resolve or create the current batch

`NEXT_ASSIGNMENT.md` is the handoff pointer, but it is not a replacement for an assignment file.

At the start of a continue cycle:

1. Read `NEXT_ASSIGNMENT.md`.

2. If `NEXT_ASSIGNMENT.md` points to or describes a batch such as `BATCH_0002`, check whether the matching assignment file exists:

   `assignments/BATCH_0002.md`

3. If that assignment file exists, process that exact batch.

4. If `NEXT_ASSIGNMENT.md` describes a batch but the matching assignment file does not exist, repair the handoff before processing:
   - create `assignments/BATCH_####.md` from the assignment table/details in `NEXT_ASSIGNMENT.md`
   - verify those files are marked `[~]` in `PROJECT_FILE_INDEX.md`
   - if they are still `[ ]`, change them to `[~]`
   - then process that batch

5. Only if `NEXT_ASSIGNMENT.md` does not contain a usable next batch, create a new `assignments/BATCH_####.md` from the next `[ ]` rows in `PROJECT_FILE_INDEX.md`.

When creating or repairing an assignment, preserve `[x]`, `[D]`, `[!]`, notes, order, and formatting in `PROJECT_FILE_INDEX.md`. Do not regenerate the whole index.

The assignment must include this table:

| Status | Source File | Lines | Bytes | Type | Effective Size | Required Mirror Doc |
|---|---|---:|---:|---|---|---|

Required mirror path format:

`src/path/file.ext` -> `docs/_file_docs/src/path/file.ext.md`

Use the batch sizing thresholds from `.claude/rules/planner_rules.md`.

Do not rely only on line count. For minified/generated/dense files, use byte size and effective size.

## Step 2: Mirror docs first

Before detailed documentation, create every required mirrored doc path from the assignment table.

Each mirror doc must start with this metadata: