codeclone 2.0.0b3

Structural code quality analysis for Python

Structural code quality analysis for Python

---

CodeClone provides deterministic structural code quality analysis for Python. It detects architectural duplication, computes quality metrics, and enforces CI gates — all with baseline-aware governance that separates known technical debt from new regressions. An optional MCP interface exposes the same canonical analysis pipeline to AI agents and IDEs.

Docs: orenlab.github.io/codeclone · Live sample report: orenlab.github.io/codeclone/examples/report/

[!NOTE] This README and docs site track the in-development v2.0.x line from main. For the latest stable CodeClone documentation (v1.4.4), see the v1.4.4 README and the v1.4.4 docs tree.

Features

• Clone detection — function (CFG fingerprint), block (statement windows), and segment (report-only) clones
• Structural findings — duplicated branch families, clone guard/exit divergence and clone-cohort drift (report-only)
• Quality metrics — cyclomatic complexity, coupling (CBO), cohesion (LCOM4), dependency cycles, dead code, health score
• Baseline governance — separates accepted legacy debt from new regressions and lets CI fail only on what changed
• Reports — interactive HTML, deterministic JSON/TXT plus Markdown and SARIF projections from one canonical report
• MCP server — optional read-only MCP surface for AI agents and IDEs, designed as a budget-aware guided control surface for agentic development
• CI-first — deterministic output, stable ordering, exit code contract, pre-commit support
• Fast — incremental caching, parallel processing, warm-run optimization, and reproducible benchmark coverage

Quick Start

pip install codeclone          # or: uv tool install codeclone

codeclone .                    # analyze
codeclone . --html             # HTML report
codeclone . --html --open-html-report  # open in browser
codeclone . --json --md --sarif --text # all formats
codeclone . --ci               # CI mode

More examples

# timestamped report snapshots
codeclone . --html --json --timestamped-report-paths

# changed-scope gating against git diff
codeclone . --changed-only --diff-against main

# shorthand: diff source for changed-scope review
codeclone . --paths-from-git-diff HEAD~1

Run without install

uvx codeclone@latest .

CI Integration

# 1. Generate baseline (commit to repo)
codeclone . --update-baseline

# 2. Add to CI pipeline
codeclone . --ci

What --ci enables

The --ci preset equals --fail-on-new --no-color --quiet. When a trusted metrics baseline is loaded, CI mode also enables --fail-on-new-metrics.

GitHub Action

CodeClone also ships a composite GitHub Action for PR and CI workflows:

- uses: orenlab/codeclone/.github/actions/codeclone@main
  with:
    fail-on-new: "true"
    sarif: "true"
    pr-comment: "true"

It can:

• run baseline-aware gating
• generate JSON and SARIF reports
• upload SARIF to GitHub Code Scanning
• post or update a PR summary comment

Action docs: .github/actions/codeclone/README.md

Quality Gates

# Metrics thresholds
codeclone . --fail-complexity 20 --fail-coupling 10 --fail-cohesion 4 --fail-health 60

# Structural policies
codeclone . --fail-cycles --fail-dead-code

# Regression detection vs baseline
codeclone . --fail-on-new-metrics

Pre-commit

repos:
  - repo: local
    hooks:
      - id: codeclone
        name: CodeClone
        entry: codeclone
        language: system
        pass_filenames: false
        args: [ ".", "--ci" ]
        types: [ python ]

MCP Server

CodeClone ships an optional read-only MCP server for AI agents and IDE clients.

# install the MCP extra
pip install "codeclone[mcp]"

# local agents (Claude Code, Codex, Copilot, Gemini CLI)
codeclone-mcp --transport stdio

# remote / HTTP-only clients
codeclone-mcp --transport streamable-http --port 8000

20 tools + 10 resources — deterministic, baseline-aware, and read-only. Never mutates source files, baselines, or repo state.

Payloads are optimized for LLM context: compact summaries by default, full detail on demand. The cheapest useful path is also the most obvious path: first-pass triage stays compact, and deeper detail is explicit.

Recommended agent flow: analyze_repository or analyze_changed_paths → get_run_summary or get_production_triage → list_hotspots or check_* → get_finding → get_remediation

Docs: MCP usage guide · MCP interface contract

Configuration

CodeClone can load project-level configuration from pyproject.toml:

[tool.codeclone]
min_loc = 10
min_stmt = 6
baseline = "codeclone.baseline.json"
skip_metrics = false
quiet = false
html_out = ".cache/codeclone/report.html"
json_out = ".cache/codeclone/report.json"
md_out = ".cache/codeclone/report.md"
sarif_out = ".cache/codeclone/report.sarif"
text_out = ".cache/codeclone/report.txt"
block_min_loc = 20
block_min_stmt = 8
segment_min_loc = 20
segment_min_stmt = 10

Precedence: CLI flags > pyproject.toml > built-in defaults.

Baseline Workflow

Baselines capture the current duplication state. Once committed, they become the CI reference point.

• Clones are classified as NEW (not in baseline) or KNOWN (accepted debt)
• --update-baseline writes both clone and metrics snapshots
• Trust is verified via generator, fingerprint_version, and payload_sha256
• In --ci mode, an untrusted baseline is a contract error (exit 2)

Full contract: Baseline contract

Exit Codes

Code	Meaning
0	Success
2	Contract error — untrusted baseline, invalid config, unreadable sources in CI
3	Gating failure — new clones or metric threshold exceeded
5	Internal error

Contract errors (2) take precedence over gating failures (3).

Reports

Format	Flag	Default path
HTML	--html	.cache/codeclone/report.html
JSON	--json	.cache/codeclone/report.json
Markdown	--md	.cache/codeclone/report.md
SARIF	--sarif	.cache/codeclone/report.sarif
Text	--text	.cache/codeclone/report.txt

All report formats are rendered from one canonical JSON report document.

• --open-html-report opens the generated HTML report in the default browser and requires --html.
• --timestamped-report-paths appends a UTC timestamp to default report filenames for bare report flags such as --html or --json. Explicit report paths are not rewritten.

The docs site also includes live example HTML/JSON/SARIF reports generated from the current codeclone repository.

Structural findings include:

• duplicated_branches
• clone_guard_exit_divergence
• clone_cohort_drift

Inline Suppressions

CodeClone keeps dead-code detection deterministic and static by default. When a symbol is intentionally invoked through runtime dynamics (for example framework callbacks, plugin loading, or reflection), suppress the known false positive explicitly at the declaration site:

# codeclone: ignore[dead-code]
def handle_exception(exc: Exception) -> None:
    ...


class Middleware:  # codeclone: ignore[dead-code]
    ...

Dynamic/runtime false positives are resolved via explicit inline suppressions, not via broad heuristics.

Canonical JSON report shape (v2.2)

{
  "report_schema_version": "2.2",
  "meta": {
    "codeclone_version": "2.0.0b3",
    "project_name": "...",
    "scan_root": ".",
    "report_mode": "full",
    "analysis_thresholds": {
      "design_findings": {
        "...": "..."
      }
    },
    "baseline": {
      "...": "..."
    },
    "cache": {
      "...": "..."
    },
    "metrics_baseline": {
      "...": "..."
    },
    "runtime": {
      "analysis_started_at_utc": "...",
      "report_generated_at_utc": "..."
    }
  },
  "inventory": {
    "files": {
      "...": "..."
    },
    "code": {
      "...": "..."
    },
    "file_registry": {
      "encoding": "relative_path",
      "items": []
    }
  },
  "findings": {
    "summary": {
      "...": "..."
    },
    "groups": {
      "clones": {
        "functions": [],
        "blocks": [],
        "segments": []
      },
      "structural": {
        "groups": []
      },
      "dead_code": {
        "groups": []
      },
      "design": {
        "groups": []
      }
    }
  },
  "metrics": {
    "summary": {},
    "families": {}
  },
  "derived": {
    "suggestions": [],
    "overview": {
      "families": {},
      "top_risks": [],
      "source_scope_breakdown": {},
      "health_snapshot": {},
      "directory_hotspots": {}
    },
    "hotlists": {
      "most_actionable_ids": [],
      "highest_spread_ids": [],
      "production_hotspot_ids": [],
      "test_fixture_hotspot_ids": []
    }
  },
  "integrity": {
    "canonicalization": {
      "version": "1",
      "scope": "canonical_only"
    },
    "digest": {
      "algorithm": "sha256",
      "verified": true,
      "value": "..."
    }
  }
}

Canonical contract: Report contract and Dead-code contract

How It Works

1. Parse — Python source to AST
2. Normalize — canonical structure (robust to renaming, formatting)
3. CFG — per-function control flow graph
4. Fingerprint — stable hash computation
5. Group — function, block, and segment clone groups
6. Metrics — complexity, coupling, cohesion, dependencies, dead code, health
7. Gate — baseline comparison, threshold checks

Architecture: Architecture narrative · CFG semantics: CFG semantics

Documentation

Topic	Link
Contract book (start here)	Contracts and guarantees
Exit codes	Exit codes and failure policy
Configuration	Config and defaults
Baseline contract	Baseline contract
Cache contract	Cache contract
Report contract	Report contract
Metrics & quality gates	Metrics and quality gates
Dead code	Dead-code contract
Docker benchmark contract	Benchmarking contract
Determinism	Determinism policy

Benchmarking Notes

Reproducible Docker Benchmark

./benchmarks/run_docker_benchmark.sh

The wrapper builds benchmarks/Dockerfile, runs isolated container benchmarks, and writes results to .cache/benchmarks/codeclone-benchmark.json.

Use environment overrides to pin the benchmark envelope:

CPUSET=0 CPUS=1.0 MEMORY=2g RUNS=16 WARMUPS=4 \
  ./benchmarks/run_docker_benchmark.sh

Performance claims are backed by the reproducible benchmark workflow documented in Benchmarking contract

License

• Code: MPL-2.0
• Documentation: MIT

Versions released before this change remain under their original license terms.

Links

• Issues: https://github.com/orenlab/codeclone/issues
• PyPI: https://pypi.org/project/codeclone/
• Licenses: MPL-2.0 · MIT docs