Lopper

Lopper is a local-first CLI/TUI for measuring dependency surface area in source repos. It compares imported dependencies to actual usage and reports waste, risk cues, and recommendations.

Install

macOS/Linux (Homebrew tap, stable):

brew tap ben-ranford/tap
brew install lopper

macOS/Linux (Homebrew tap, rolling):

brew install ben-ranford/tap/lopper-rolling

lopper-rolling tracks main and is not a stable semver release.

Windows (GitHub Releases):

# Open latest release page and download the Windows asset for your platform.
gh release view --repo ben-ranford/lopper --web

Generate a local manpage:

make manpage

The generated manpage is available as docs/man/lopper.1 and is installed automatically by the Homebrew formulas.

Run without local install:

docker run --rm ghcr.io/ben-ranford/lopper:latest --help

GitHub Action

Use the first-party Scan with Lopper action in CI:

- uses: ben-ranford/lopper@v1
  with:
    version: action
    repo: .
    language: all
    top: '20'

For reproducible CI, pin both the action ref and version to a concrete release such as v1.7.0. See docs/ci-usage.md for PR comment and SARIF workflows.

Terminal demos

Demo	What it demonstrates	GIF preview
Quick start ranking	End-to-end --top workflow with license, policy, and candidate-score context for fast triage.
Single dependency deep dive	Focused analysis of one dependency with current usage, license, and scoring context.
Baseline gate in CI flow	Baseline comparison with threshold policy and delta summary to catch regression risk in automated checks.

Quick Start

Analyze one dependency:

lopper analyse lodash --repo . --language js-ts

Rank top dependencies by waste:

lopper analyse --top 20 --repo . --language all --format table

Emit JSON:

lopper analyse --top 20 --repo . --language all --format json

Emit CSV for spreadsheet or pipeline ingestion:

lopper analyse --top 20 --repo . --language all --format csv > lopper.csv

Emit SARIF for code scanning:

lopper analyse --top 20 --repo . --language all --format sarif > lopper.sarif

Emit a preview CycloneDX JSON SBOM for direct dependency rows:

lopper analyse --top 20 --repo . --language all \
  --format cyclonedx-json \
  --enable-feature sbom-attestation-exports-preview > lopper.cdx.json

The CycloneDX export preserves Lopper-specific usage, reachability, runtime, license, provenance, waste, removal-candidate, and baseline context as component properties. It does not infer missing versions or package URLs, and it is not a full transitive dependency inventory. SPDX JSON, dashboard-wide combined SBOMs, and signed attestations are deferred from the first preview.

Launch the interactive TUI:

lopper tui --repo . --language all

Run the local MCP server for agent workflows:

lopper mcp

See docs/mcp.md for tool names, input schemas, client configuration, and safety behavior.

Tune thresholds and score weights:

lopper analyse --top 20 \
  --repo . \
  --language all \
  --threshold-fail-on-increase 2 \
  --threshold-low-confidence-warning 35 \
  --threshold-min-usage-percent 45 \
  --enable-feature reachability-vulnerability-prioritization-preview \
  --advisory-source security/lopper-advisories.yml \
  --threshold-reachable-vuln-priority high \
  --score-weight-usage 0.50 \
  --score-weight-impact 0.30 \
  --score-weight-confidence 0.20

Save an immutable baseline snapshot keyed by commit:

lopper analyse --top 20 \
  --repo . \
  --language all \
  --format json \
  --baseline-store .artifacts/lopper-baselines \
  --save-baseline

Save using a human label key:

lopper analyse --top 20 \
  --repo . \
  --language all \
  --format json \
  --baseline-store .artifacts/lopper-baselines \
  --save-baseline \
  --baseline-label release-candidate

Compare against a stored baseline key and gate CI:

lopper analyse --top 20 \
  --repo . \
  --language all \
  --format json \
  --baseline-store .artifacts/lopper-baselines \
  --baseline-key commit:abc123 \
  --threshold-fail-on-increase 2

List the newest saved snapshots, or inspect one without printing its full report payload. Baseline discovery is stable in v1.8.1:

lopper baseline list \
  --store .artifacts/lopper-baselines \
  --limit 20

lopper baseline show label:release-candidate \
  --store .artifacts/lopper-baselines \
  --format json

Attach local vulnerability advisories for reachability-weighted security triage:

lopper analyse --top 20 \
  --repo . \
  --language all \
  --enable-feature reachability-vulnerability-prioritization-preview \
  --advisory-source security/lopper-advisories.yml \
  --threshold-reachable-vuln-priority high

Advisory ingestion is preview-gated and local-only. Lopper does not fetch a proprietary or network vulnerability database. The current preview matches package name and ecosystem but does not evaluate installed versions against OSV affected ranges, so use a curated local advisory snapshot and treat fixedVersion as informational. The priority score ranks triage using advisory severity plus reachability, runtime, and static import evidence; it is not an exploitability claim.

Generate an org-level dashboard across multiple repos:

lopper dashboard \
  --repos "./api,./frontend,./worker" \
  --format html \
  --output org-report.html

Use a dashboard config file:

lopper dashboard --config lopper-org.yml --format json

Remote dashboard config entries can use repoUrl with the dashboard-remote-repos feature and may pin exactly one of branch, tag, or full commit SHA. Unpinned remote entries continue to track remote HEAD; dashboard JSON, CSV, HTML, and saved dashboard baselines include the resolved commit SHA for materialized remote repos.

Languages

• Supported adapters: js-ts, python, cpp, jvm, kotlin-android, go, php, ruby, rust, dotnet, elixir, swift, dart, powershell
• js-ts merges workspace-level declarations from pnpm-workspace.yaml, package.json#workspaces, and Yarn .yarnrc.yml catalogs.
• Source of truth for adapter IDs: lopper --help
• Language modes:
  • auto: choose highest-confidence adapter
  • all: run all matching adapters and merge results
  • <id>: force one adapter

Repo-level config example (.lopper.yml):

policy:
  packs:
    - ./policies/org-defaults.yml
    - https://example.com/lopper/policy.yml#sha256=0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef
thresholds:
  fail_on_increase_percent: 2
  low_confidence_warning_percent: 35
  min_usage_percent_for_recommendations: 45
  reachable_vulnerability_priority: high
  removal_candidate_weight_usage: 0.50
  removal_candidate_weight_impact: 0.30
  removal_candidate_weight_confidence: 0.20
advisories:
  source: security/lopper-advisories.yml

Threshold defaults:

• fail_on_increase_percent: -1 (disabled)
• low_confidence_warning_percent: 40
• min_usage_percent_for_recommendations: 40
• reachable_vulnerability_priority: off
• removal_candidate_weight_usage: 0.50
• removal_candidate_weight_impact: 0.30
• removal_candidate_weight_confidence: 0.20

Threshold ranges:

• fail_on_increase_percent must be -1 or >= 0
• low_confidence_warning_percent must be between 0 and 100
• min_usage_percent_for_recommendations must be between 0 and 100
• reachable_vulnerability_priority must be off, low, medium, high, or critical
• removal candidate weights must be >= 0 and at least one must be greater than 0

Precedence is CLI > repo config > imported policy packs > defaults.

Additional guides:

• docs/threshold-tuning.md
• docs/notifications.md
• docs/feature-flags.md

Launch TUI:

lopper tui --repo . --language all

Compare the TUI summary against a stored baseline:

lopper tui \
  --repo . \
  --language all \
  --baseline-store .artifacts/lopper-baselines \
  --baseline-key commit:abc123

Inside the TUI, open a dependency to inspect codemod suggestions and apply safe suggestions without leaving the session:

open js-ts:lodash
apply-codemod --confirm

apply-codemod refuses dirty git worktrees by default; add --allow-dirty only when you intentionally want to apply into a dirty tree. The action prints applied, skipped, and failed files plus the rollback backup path. Baselines can also be saved and compared in-session:

Python safe unused-import suggestions are stable and enabled by default. They are syntactically conservative rather than proof that an import has no module side effects. Mutation still requires --apply-codemod --apply-codemod-confirm, refuses a dirty worktree by default, and writes rollback evidence. Use --disable-feature python-codemod-suggestions for an explicit rollback.

save-baseline
save-baseline release-candidate
compare-baseline label:release-candidate
compare-baseline --file baseline.json

In-session baseline commands default to .artifacts/lopper-baselines and commit:<HEAD> when no store or label/key is provided.

Runtime trace annotations

JS/TS

Capture a runtime trace:

export LOPPER_RUNTIME_TRACE=.artifacts/lopper-runtime.ndjson
export LOPPER_ROOT=/path/to/lopper
export NODE_OPTIONS="--require ${LOPPER_ROOT}/scripts/runtime/require-hook.cjs --loader ${LOPPER_ROOT}/scripts/runtime/loader.mjs"
npm test

Use the hook files from the lopper checkout or install tree. Relative hook paths resolve from the repo running npm test, not from lopper itself.

Or let Lopper run the test command and capture the trace automatically:

lopper analyse --top 20 --repo . --language js-ts --runtime-test-command "npm test"

Use trace in analysis:

lopper analyse --top 20 --repo . --language js-ts --runtime-trace .artifacts/lopper-runtime.ndjson

With runtime traces enabled:

• runtimeUsage.correlation marks each supported dependency as static-only, runtime-only, or overlap.
• runtimeUsage.modules includes runtime-loaded module paths.
• runtimeUsage.parentModules includes parent module paths that triggered the load.
• runtimeUsage.entrypoints includes entrypoint modules seen in the runtime trace.
• runtimeUsage.topSymbols includes best-effort runtime symbol hits.

If --runtime-trace points to a missing file, analysis continues with static results and adds a warning.

Python

First-party Python runtime capture is stable for these conservative pytest-family command forms (runner arguments may follow):

pytest
python -m pytest
python3 -m pytest

lopper analyse --top 20 --repo . --language python \
  --runtime-test-command "pytest"

The stable python-runner-profiles feature adds these direct-exec forms:

python -m unittest
python3 -m unittest
uv run pytest
uv run -- pytest
uv run python -m pytest
uv run python3 -m pytest
uv run -- python -m pytest
uv run -- python3 -m pytest
uv run python -m unittest
uv run python3 -m unittest
uv run -- python -m unittest
uv run -- python3 -m unittest

For example:

lopper analyse --top 20 --repo . --language python \
  --runtime-test-command "uv run -- python -m unittest discover -s tests"

The optional -- immediately after uv run is the only accepted uv wrapper delimiter. Arguments after the selected pytest command or python[3] -m pytest|unittest module are forwarded verbatim, including a later --; uv wrapper flags, arbitrary uv tools, Python interpreter flags, inline environment assignments, and shell operators are rejected.

Lopper resolves supported executables from trusted absolute PATH entries in order, then from its fixed system fallback directories. It rejects writable search directories, writable/non-executable tools, and executables outside the allowlist. Lopper injects its import hook only into the runtime command environment by prepending the shipped scripts/runtime/sitecustomize.py directory to PYTHONPATH; an existing project sitecustomize.py is chained exactly once. It does not install project dependencies or invoke a shell. Use --disable-feature python-runtime-capture or --disable-feature python-runner-profiles for rollback.

Explicit Python trace consumption remains compatible through --runtime-trace.

Each trace line should identify the Python adapter and the imported module:

{"language":"python","module":"requests.sessions","parent":"/repo/main.py","entrypoint":"/repo/main.py"}

When the import root differs from the package name, include dependency:

{"language":"python","dependency":"beautifulsoup4","module":"bs4"}

Use the trace in analysis:

lopper analyse --top 20 --repo . --language python --runtime-trace .artifacts/python-runtime.ndjson

Python caveats:

• Runtime module names map to the top-level import package, with common aliases such as bs4 -> beautifulsoup4.
• The first-party hook emits third-party imports resolved from site-packages or dist-packages and filters local modules and stdlib imports.
• Subprocesses and pytest workers inherit the trace environment when they inherit the parent process environment; concurrent writers append NDJSON lines to the same trace file.
• Project sitecustomize.py runs once before Lopper installs its import wrapper; imports performed only during that startup hook are not included in the trace.

Development

See CONTRIBUTING.md for development setup and commands. For watched hotspot packages, run make mem-profiles to capture alloc-space summaries before or after performance-sensitive changes.

Docs

• Report schema: docs/report-schema.json, docs/report-schema.md
• Multi-repo dashboard: docs/dashboard.md
• Memory profiling workflow: docs/memory-profiling.md
• SARIF code scanning: docs/sarif-code-scanning.md
• Threshold tuning: docs/threshold-tuning.md
• MCP server integration: docs/mcp.md
• Runtime trace annotations: scripts/runtime/
• Adapter and architecture extensibility: docs/extensibility.md
• CI and release workflow: docs/ci-usage.md
• Contribution guide: CONTRIBUTING.md