Add refactoring-cadence agentic workflow

[Copilot]

Creates a daily weekday workflow that tracks repository code health over time and auto-opens a GitHub issue when quality degrades — a forcing function for continuous refactoring.

Workflow: refactoring-cadence

Trigger: Daily on weekdays (scattered cron). Skips if an open [refactoring-cadence] issue already exists.

4 health metrics → 0–100 score:

Metric	Weight	Tools
Avg file length	25 pts	wc -l
Avg cyclomatic complexity	25 pts	gocyclo (Go, primary), branch-count grep (JavaScript, primary), radon (Python, optional), cargo clippy (Rust, optional)
Files grown >20% vs baseline	25 pts	diff against cached baseline
TODO/FIXME/HACK net churn	25 pts	grep

Primary languages: Go and JavaScript/CommonJS — all file scans target *.go, *.js, and *.cjs by default. Python and Rust analysis runs automatically if those files are present, but are not required.

When score drops below threshold (default: 70): opens issue [refactoring-cadence] 🔧 Refactoring Required: Code Health Alert labeled refactoring + ai-generated, containing per-file degradation details, an AI-generated prioritized refactoring plan, and an actionable checklist.

Baseline persistence: metrics stored as JSON in cache-memory between runs. First run is baseline-only — no issue created.

Configurable via .refactoring-cadence.yml at repo root (threshold, max avg file length, max avg complexity, growth %, TODO churn limit). Default extensions: go,js,cjs.

[pelikhan]

@copilot this code base is Go and JavaScript

[Copilot]

@copilot this code base is Go and JavaScript

Updated in 0a4ecfc. Go (gocyclo) and JavaScript/CommonJS (branch-count grep) are now the primary analysis targets — Python (radon) and Rust (cargo-clippy) are demoted to optional secondary sections that only run when those files are present. Default file extensions changed from go,py,rs,js,ts,cjs to go,js,cjs.

[Copilot]

Pull request overview

Adds a new scheduled agentic workflow (“Refactoring Cadence”) that tracks code-health metrics over time and opens a refactoring issue when the score degrades, plus accompanying compiled workflow artifacts.

Changes:

• Introduces the refactoring-cadence workflow definition and its compiled lockfile.
• Implements baseline persistence via cache-memory and issue creation via Safe Outputs when health drops below threshold.
• Updates architecture-guardian.lock.yml to run Copilot via copilot_driver.cjs.

Show a summary per file

File	Description
.github/workflows/refactoring-cadence.md	New agent workflow spec to compute health metrics, compare to baseline, and (conditionally) open an issue.
.github/workflows/refactoring-cadence.lock.yml	Generated compiled workflow implementing the new agent spec with caching, safe outputs, and scheduling.
.github/workflows/architecture-guardian.lock.yml	Regenerated compiled workflow to invoke Copilot through the standard driver script.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comments suppressed due to low confidence (1)

.github/workflows/refactoring-cadence.md:179

• The file-growth loop uses while read f; do ... which will mangle paths containing backslashes and can trim leading/trailing whitespace. Use while IFS= read -r f (and ideally NUL-delimit via -print0) to keep file paths intact.

find . -type f \( -name "*.go" -o -name "*.js" -o -name "*.cjs" \) \
  ! -path "*/vendor/*" ! -path "*/node_modules/*" ! -path "*/.git/*" \
  ! -name "*_test.go" ! -name "*.lock.yml" ! -name "*.pb.go" \
  | sort | while read f; do
      lines=$(wc -l < "$f" 2>/dev/null || echo 0)
      echo "$f $lines"
    done

• Files reviewed: 3/3 changed files
• Comments generated: 5

[Copilot]

Metric 1 uses find ... | xargs wc -l, which will mis-handle file paths containing whitespace/newlines and can silently skew avg_file_length/file_count. Prefer find ... -print0 | xargs -0 wc -l or find ... -exec wc -l {} + to make the metric path-safe.

This issue also appears on line 173 of the same file.

Suggested change

	| xargs wc -l 2>/dev/null | awk '$2 != "total" {sum+=$1; count++} END {if(count>0) print sum/count; else print 0}'
	-exec wc -l {} + 2>/dev/null | awk '$2 != "total" {sum+=$1; count++} END {if(count>0) print sum/count; else print 0}'

[Copilot]

GO_FILES is filtered to exclude vendor/ and *_test.go, but the gocyclo invocation runs on . (which may include vendor/tests/generated code). This makes the complexity metric inconsistent with the documented exclusions; pass only the filtered file set (or run gocyclo on filtered packages/paths) so avg_complexity reflects the intended scope.

Suggested change

	GO_AVG=$(gocyclo -avg . 2>/dev/null | grep "^Average" | awk '{print $NF}')
	GO_AVG=$(echo "$GO_FILES" | xargs gocyclo -avg 2>/dev/null | grep "^Average" | awk '{print $NF}')

[Copilot]

The JS complexity comment says it counts branch points including &&, ||, and ??, but the regex only counts if/for/while/switch statements. Either update the regex to match the described operators or adjust the comment so the metric definition matches the implementation.

Suggested change

	# Count branch points: if, for, while, switch, &&, ||, ??
	# Count branch statements: if, for, while, switch

[Copilot]

The Python path installs radon with pip install (no venv) and then unconditionally runs radon via xargs. On GitHub-hosted runners, pip install often fails due to an externally-managed environment, and the workflow doesn't verify radon is available before use. Consider using a venv (as done in .github/workflows/shared/python-dataviz.md) and/or guarding the radon call with command -v radon, plus adding the appropriate network.allowed entry if this workflow is expected to fetch packages.

Suggested change

	pip install radon --quiet 2>/dev/null || true
	echo "$PY_FILES" | xargs radon cc -s -a 2>/dev/null \
	| grep "Average complexity:" | awk '{sum+=$NF; count++} END {if(count>0) print "py:" sum/count; else print "py:0"}'
	if python3 -m venv .radon-venv 2>/dev/null; then
	. .radon-venv/bin/activate
	if python -m pip install --quiet radon 2>/dev/null && command -v radon >/dev/null 2>&1; then
	echo "$PY_FILES" | xargs radon cc -s -a 2>/dev/null \
	| grep "Average complexity:" | awk '{sum+=$NF; count++} END {if(count>0) print "py:" sum/count; else print "py:0"}'
	else
	echo "py:0"
	fi
	else
	echo "py:0"
	fi

[Copilot]

Metric 4 claims the workflow excludes *_test.go and generated files, but this find command does not exclude *_test.go (or *.pb.go). That makes TODO churn inconsistent with the other metrics and can cause churn to change due to test/generated updates. Align the exclusions here with Metrics 1/3.

Suggested change

	! -name "*.lock.yml" \
	! -name "*_test.go" ! -name "*.lock.yml" ! -name "*.pb.go" \