orgkit-analyze

You are giving the user an intelligent, read-only preview of how orgkit would organize their repository. This is the teaser people run first — so it must show the real intelligence, not a keyword guess. You read the repo and reason; you write nothing.

The Python tool handles mechanics. You handle judgment. This command is exactly /orgkit-init's reasoning, minus any writing.

---

Step 1 — Extract deterministic folder signals

The user's repo root is ${CLAUDE_PROJECT_DIR} (fall back to $PWD). Get the token-bounded signal payload.

Prefer the installed engine if it exists; fall back to the plugin copy:

# Use whichever exists:
#   .org/scan.py  — installed engine (preferred)
#   ${CLAUDE_PLUGIN_ROOT}/orgkit/scan.py  — plugin copy (pre-install fallback)
if [ -f "${CLAUDE_PROJECT_DIR:-$PWD}/.org/scan.py" ]; then
  python3 "${CLAUDE_PROJECT_DIR:-$PWD}/.org/scan.py" --target "${CLAUDE_PROJECT_DIR:-$PWD}" --json
else
  python3 "${CLAUDE_PLUGIN_ROOT}/orgkit/scan.py" --target "${CLAUDE_PROJECT_DIR:-$PWD}" --json
fi

Parse the JSON. For each top-level folder you get: name, file_count, top_exts, kind (python/node/web/data/docs/infra/mixed), telltale files, and a ~300-char excerpt.

---

Step 2 — YOU reason over the signals (the intelligent step)

Do not fall back to keyword matching. For any folder where the kind is mixed/docs, the name is cryptic (v2/, proj-final/, temp2/, experiments/), or the excerpt is unhelpful — read a file or two yourself to learn what it actually is:

cat "${CLAUDE_PROJECT_DIR}/<folder>/README.md" 2>/dev/null
ls "${CLAUDE_PROJECT_DIR}/<folder>/"

Then reason through the structure:

1. What functions are actually represented? Cluster by what the work is (product engineering, data science, design, growth, ops, research…), not by folder names.
2. What 3–8 roles fit THIS repo? Match how the team actually thinks. Too many roles defeats the purpose.
3. Map every folder to a role, each with a one-line rationale grounded in real content.
4. Resolve cryptic names explicitly — if v2/ is a React app, say so; if experiments/ is credit-risk notebooks, say so.