konjo-retrofit

name	konjo-retrofit
description	Retrofit the Konjo Quality Framework onto an existing repo that predates it. Use when asked to add konjo quality gates, improve code quality, audit an existing codebase, or run a quality sprint on any repo. Provides the step-by-step migration plan and triage protocol.
user-invocable	true

Konjo Retrofit — Existing Repo Quality Migration

The Problem With Retrofitting Blind

Installing hard quality gates on an existing codebase without measuring first causes one of two outcomes:

1. Gates fail on day 1 — blocks all work, team disables the gates in frustration
2. Gates are set too loose — they pass everything, provide no value

The Retrofit Protocol solves this by measuring before gating, then ratcheting up incrementally.

---

Step 1 — Baseline Audit (measure everything, fix nothing yet)

Run these and save the output. Do not fix violations yet — just establish the baseline.

# Coverage baseline
cargo llvm-cov nextest --workspace --json > coverage_baseline.json    # Rust
python -m pytest --cov --cov-report=json > coverage_baseline.json     # Python

# Lint baseline
cargo clippy --workspace --all-targets --message-format json > clippy_baseline.json  # Rust
ruff check --output-format json > ruff_baseline.json                                  # Python

# Dead code
RUSTFLAGS="-W dead_code" cargo check --workspace 2>&1 | grep "dead_code" > dead_code.txt  # Rust
vulture . --min-confidence 60 > vulture_baseline.txt                                        # Python

# Complexity
radon cc . -n C > complexity_baseline.txt        # Python (grade C = cyclomatic > 10)
# For Rust: cargo clippy -W clippy::cognitive_complexity 2>&1 | grep "cognitive_complexity"

# DRY
python3 .konjo/scripts/dry_check.py --json > dry_baseline.json

# File sizes
find . -name "*.rs" -o -name "*.py" | grep -v target | \
  xargs wc -l | sort -n | tail -20 > large_files.txt

# Mutation testing (sample — don't run full corpus on first audit)
cargo mutants --lib --jobs 2 2>&1 | tail -50 > mutation_sample.txt   # Rust
mutmut run --paths-to-mutate src/ 2>&1 | tail -50 > mutation_sample.txt  # Python

---

Step 2 — Triage

Classify all violations into three categories:

Priority	Category	Definition	Handle
P0	CRITICAL	Security issues, data corruption bugs, race conditions	Fix immediately, before framework install
P1	DEBT	Coverage < 60%, unwrap() in production paths, undocumented public APIs	Fix in first 2 sprints
P2	STYLE	Length violations, moderate duplication, complexity > 20	Fix incrementally, 1-2 per sprint

Never retrofit CRITICAL and DEBT in the same commit. Fix P0 first, measure, commit. Then P1.

---

Step 3 — Install Framework (at current baseline, not ideal baseline)

Install with warn-only mode for the first sprint:

# Copy framework files
bash .konjo/scripts/install-hooks.sh

# Install CI with warn-only gates
# Set coverage gate at: current_coverage - 2% (don't let it regress, but don't block yet)
# Set dead code gate to: warn, not block
# Set Wall 3 review to: --soft-fail (report only)

Rationale: A gate that fails on the day it's installed will be disabled. Set it just above current, then ratchet.

---

Step 4 — The Coverage Ratchet

Coverage gates increment on a fixed schedule. Never move the gate backward.

Sprint	Coverage Gate	Action
Sprint 0 (install)	current - 2%	No regressions allowed
Sprint 1	current	Hold the line
Sprint 2	current + 5%	Write missing tests
Sprint 3	current + 10%	Continue
Sprint N	80%	Hard floor (production gate)
Long-term	95%	Target

Add to PLAN.md: the current coverage gate value. Never let anyone lower it.

---

Step 5 — Complexity and Length Ratchet

For each oversized function or file:

Protocol (one function at a time):

1. Write a characterization test first — captures current behavior before touching anything
2. Make the smallest refactor that improves the metric
3. Run full test suite — all tests must pass
4. Run coverage — must not decrease
5. Commit: refactor(scope): extract X into Y [complexity: 23→11 lines: 80→35]

Never refactor + change behavior in the same commit.

---

Step 6 — DRY Cleanup

Run dry_check.py --json and sort violations by similarity descending. Address highest-similarity violations first (easiest wins).

For each violation:

1. Identify the canonical location (the version with the best name/tests)
2. Write a test for the canonical version if it doesn't have one
3. Replace all duplicates with calls to the canonical version
4. Confirm tests pass

---

Step 7 — Activate Wall 3 (adversarial review)

After Walls 1 and 2 are stable (at least 2 consecutive clean CI runs):

# Week 1: soft-fail mode (report only, does not block merge)
python3 .konjo/scripts/konjo_review.py --soft-fail

# Week 2: full blocking mode
python3 .konjo/scripts/konjo_review.py  # exits 1 on BLOCKER

Add ANTHROPIC_API_KEY to GitHub Actions secrets before enabling.

---

Repo-Type Checklists

Pure Rust (lopi)

• cargo audit clean
• cargo deny check configured and clean
• cargo-husky dev-dep added to Cargo.toml
• cargo llvm-cov gives ≥ 80% coverage
• clippy -D unwrap_used passes
• RUSTDOCFLAGS="-D missing_docs" cargo doc passes
• .konjo/deny.toml committed
• .github/workflows/konjo-gate.yml active

Pure Python (squish, kyro, miru, kairu, squash)

• ruff check clean; ruff format --check clean
• mypy --strict clean (or each # type: ignore justified)
• vulture < 5% dead code
• pytest --cov --cov-fail-under=80 passes
• radon cc -n C zero grade-C functions
• interrogate --fail-under 80 on public APIs
• pre-commit framework installed with .pre-commit-config.yaml
• .github/workflows/konjo-gate.yml adapted for Python

Rust + Python hybrid (kohaku, toki, drex, vectro)

• Both Rust and Python checklists above
• Python-only mode is always the correctness baseline
• Rust and Python implementations agree to documented tolerance
• PyO3 bindings tested from both sides

---

Documenting the Retrofit Sprint

After each retrofit sprint, add to CHANGELOG.md:

## [vX.Y.Z] — YYYY-MM-DD — Quality Retrofit Sprint N

### Improved
- Coverage: X% → Y%
- Dead code: N warnings → 0
- Functions refactored: list them
- DRY violations resolved: list them

### Quality Gates Now Active
- Coverage gate: Z%
- Wall 1 (pre-commit): active
- Wall 2 (CI): active
- Wall 3 (adversarial review): soft-fail / blocking

---

The Shipbuilder's Checklist (Final Verification)

Before declaring retrofit complete, answer these:

• The test suite runs in < 5 minutes (if not, profile and optimize)
• A PR that deletes a function without updating its callers would fail CI
• A PR that introduces an unwrap() would fail CI
• A PR that drops coverage below 80% would fail CI
• A PR is reviewed by the Konjo Critic before merge
• Running the code for 30 days without restart would not cause memory growth
• The code can be read and extended by someone who wasn't there when it was written

If any of these are false, the retrofit is not complete.