Archie Enrich

# Archie Enrich — AI-Enriched Per-Folder CLAUDE.md

Re-enrich per-folder CLAUDE.md files with AI-generated patterns, anti-patterns, debugging tips, and code examples. Use this after code changes to refresh enrichments, or if you ran `/archie-init` before this feature existed.

**Prerequisites:** Requires `.archie/scan.json` and `.archie/blueprint.json`.

**IMPORTANT: Do NOT write inline Python. Every step uses pre-installed scripts from `.archie/`.**

## Step 1: Prepare the folder DAG

```bash
python3 .archie/enrich.py prepare "$PWD"
mkdir -p .archie/enrichments

Step 2: Process in readiness waves

Maintain a list of done folder paths (starts empty).

Repeat until done:

1. Get ready folders (folders whose children are all done, or leaves with no children):

python3 .archie/enrich.py next-ready "$PWD" <done1> <done2> ...

First call with no done folders returns all leaf folders.

2. If the ready list is empty ([]), all folders are done. Proceed to Step 3.

3. Get batches for the ready folders:

python3 .archie/enrich.py suggest-batches "$PWD" <ready1> <ready2> ...

4. For each batch, generate the prompt and spawn a subagent:

python3 .archie/enrich.py prompt "$PWD" --folders <comma-separated> --child-summaries .archie/enrichments/ > /tmp/archie_enrich_prompt.txt

Read the prompt file. Spawn a Sonnet subagent (model: "sonnet") with the prompt content. The subagent must return ONLY valid JSON with folder paths as keys. Spawn ALL batches in a wave in parallel.

5. Save each subagent's JSON output:

Write .archie/enrichments/<batch_id>.json with the subagent's JSON output

6. Add all folders from completed batches to the done list. Go to (1).

Step 3: Merge enrichments into CLAUDE.md files

python3 .archie/enrich.py merge "$PWD"

Step 4: Clean up

rm -f /tmp/archie_enrich_prompt.txt

Tell the user: "Per-folder CLAUDE.md files have been enriched with AI-generated developer notes." ```