DSIU — Dynamic Systems Intelligence Unit

Observe. Map. Upgrade. Repeat.

DSIU is a research, analysis, and operational-design lens: point it at any system — a codebase, an AI pipeline, an app, a workflow, an organization, a machine, or a behavior pattern — and it exposes the hidden Intake → Processing → Output architecture, separates the layers, locates the control points and failure points, and produces a structured upgrade plan.

This repo turns the DSIU framework from prose into a runnable, installable Claude Code skill plus a small standard-library engine.

What's here

Path	What it is
dsiu/	The skill bundle (source of truth): SKILL.md, doctrine in references/, templates/, and the scripts/dsiu_analyze.py engine.
scripts/build_skill.py	Validates the bundle and packages it into dist/dsiu.skill.
tests/	Stdlib unittest smoke tests (no third-party deps).
examples/dsiu_on_dsiu_field.md	DSIU's lens turned on its own v1 bundle — a worked Field Template and the rationale for this hardening pass.
.github/workflows/ci.yml	Runs the tests and builds the .skill on every push/PR.

The four things this delivers

1. An executable skill — the full bundle in dsiu/, built reproducibly into an installable .skill.
2. A lens you can run on a real codebase — scan mode maps a repo onto the intake → processing → output trunk and emits a DRAFT Field Template.
3. A deterministic scorecard generator — scorecard mode emits the per-layer attribute scorecard (qualities / skills / strengths / weaknesses / control points) as JSON and Markdown, with no model call.
4. A CPK FINAL hand-off — packet mode emits a Mission-Packet-shaped JSON (inputs / hard_locks / failure_bans). DSIU is the Observe front-end; CPK scores, gates, and executes the upgrade. (CPK FINAL is external to this repo — this is the documented contract, not a live call.)

Plus the keystone the framework was missing: diff mode, DSIU's own feedback loop — compare two passes and get a movement score, so "DSIU building DSIU" becomes measurable instead of metaphorical.

Usage

# Blank Field Template for any system (code or not)
python dsiu/scripts/dsiu_analyze.py template --name "My System" > my_system.dsiu.md

# Heuristic first-pass map of a codebase (always a DRAFT — verify by hand)
python dsiu/scripts/dsiu_analyze.py scan --path ./repo --name "My Repo" \
    --out repo.dsiu.md --json repo.scorecard.json

# Add --include-docs so knowledge/config-driven systems aren't invisible
python dsiu/scripts/dsiu_analyze.py scan --path ./repo --include-docs --json repo.json

# Structured attribute scorecard for a named system (JSON + Markdown, no scan)
python dsiu/scripts/dsiu_analyze.py scorecard --name "My System" \
    --out my_system.scorecard.md --json my_system.scorecard.json

# Feedback loop: measure what moved between two passes
python dsiu/scripts/dsiu_analyze.py diff --before before.json --after after.json \
    --out delta.md --json delta.json

# Hand-off to CPK FINAL (contract shape)
python dsiu/scripts/dsiu_analyze.py packet --json repo.scorecard.json --out packet.json

Build & test

python -m unittest discover -s tests   # run the smoke tests
python scripts/build_skill.py          # -> dist/dsiu.skill (installable bundle)

Requires Python 3.8+ and nothing else (standard library only).

Law 0 — Honesty of Claim

DSIU is a lens, not a magic upgrader. scan/diff output is a heuristic first pass, always marked DRAFT, and must be verified by a human before it is acted on. DSIU maps, scores, and proposes — it never claims to have upgraded, fixed, or sealed anything. Those are real engineering actions taken after the diagnosis. Full text in dsiu/references/governance.md.