lazyline
Line-level Python profiler skill for Claude Code. Profile any package or script to find exactly which lines are slow, then optimize iterativ
Open source Repository Open in the app JSON README (API)
About
Line-level Python profiler skill for Claude Code. Profile any package or script to find exactly which lines are slow, then optimize iteratively.
Details
- Kind
- Plugins
- Topic
- No topic detected
- Publisher
- tomasvenkrbec
- Origin
- marketplace
- Category
- ferramentas
- Stars
- 2
- Last push
- 2026-04-17T07:26:10Z
- Repository state
- ativo
- Language
- Python
- License
- MIT
- Added
- 2026-08-30 01:48:58
- Updated
- 2026-08-30 01:48:58
- Origin id
tomasvenkrbec/lazyline/lazyline
README
# Lazyline: Zero-Config Line-Level Python Profiler [](https://pypi.org/project/lazyline/) [](https://pypi.org/project/lazyline/) [](https://pypi.org/project/lazyline/) [](https://github.com/TomasVenkrbec/lazyline/actions/workflows/ci.yml) [](https://codecov.io/gh/TomasVenkrbec/lazyline) [](https://github.com/TomasVenkrbec/lazyline/blob/main/LICENSE) **Zero-config, deterministic, line-level Python profiler.** Find slow Python code and profile it line by line — no `@profile` decorators, no code changes. Point it at a package or script and get exact hit counts and timing for every line. Subprocesses and multiprocessing pools profiled automatically. Find the lazy lines. ## Quick Start ```bash pip install lazyline # or: uvx lazyline # Profile a package while running its tests: lazyline run my_package -- pytest tests/ # Profile a script: lazyline run script.py -- python script.py ```  Line 14 burned 99.8% of `deduplicate` checking membership in a list on every iteration — that's your lazy line. Change `seen` to a `set` and it drops from O(n²) to O(n). > [!TIP] > Using an AI coding assistant? Install the > [lazyline plugin](#claude-code-plugin) for Claude Code, or copy > [`skills/lazyline/SKILL.md`](skills/lazyline/SKILL.md) into any > assistant that supports markdown skill files. ## Key Features Need to find performance bottlenecks in your Python code without modifying a single file? Lazyline wraps [line_profiler](https://github.com/pyutils/line_profiler) and adds everything needed to go from "I want to profile this package" to "here are the bottlenecks" in a single command: - **Zero configuration** — point at a package name, directory, or `.py` file. Every function is discovered and instrumented automatically. No `@profile` decorators, no code changes — be lazy, let the tool do the work. - **Subprocess and multiprocessing** — `ProcessPoolExecutor`, `multiprocessing.Pool`, and child Python processes (Celery workers, Airflow tasks) are profiled automatically. Results are merged into a single report. - **Deterministic precision** — exact hit counts and timing for every line, not statistical estimates. "This line ran 47,382 times and took 3.2s." When you need to distinguish O(n) from O(n²), exact counts are the difference. - **Focused scope, clean output** — you choose exactly which package to profile. Unlike tools that profile everything in your working directory, lazyline keeps output relevant and overhead contained. ## When to Use What No tool is best for everything. Pick the right one for the job: | You need... | Use | Why | | ------------- | ----- | ----- | | Exact line-level timing across a package, no code changes | **[lazyline](https://github.com/TomasVenkrbec/lazyline)** | Deterministic tracing with auto-discovery and subprocess support | | Low-overhead profiling with memory, GPU, and AI suggestions | **[Scalene](https://github.com/plasma-umass/scalene)** | Sampling (~10-20% overhead), broad feature set, web UI | | Attach to a running process in production | **[py-spy](https://github.com/benfred/py-spy)** | Out-of-process sampling, near-zero overhead, no restart needed | | "Which function is slow?" with beautiful call trees | **[Pyinstrument](https://github.com/joerick/pyinstrument)** | Statistical profiler, tree output, low overhead | ### Feature comparison | Feature | lazyline | kernprof | Scalene | py-spy | Pyinstrument | cProfile | | --------- | ---------- | ---------- | --------- | -------- | -------------- | ---------- | | Granularity | Line | Line | Line | Line | Function | Function | | Method | Deterministic | Deterministic | Sampling | Sampling | Sampling | Deterministic | | Code changes needed | None | `@profile` | None | None | None | None | | Exact hit counts | Yes | Yes | No | No | No | Yes (fn-level) | | Subprocess profiling | Automatic | No | Partial | Yes | No | No | | Multiprocessing pools | Automatic | No | Partial | Yes | No | No | | Memory profiling | Opt-in | No | Built-in | No | No | No | | GPU profiling | No | No | Yes | No | No | No | | Overhead | 1.2–7x | 1.2–7x | ~10–20% | ~0% | Low | Moderate | **Lazyline trades overhead for precision.** Deterministic tracing fires a callback on every line execution. For functions with real work (>0.1ms per call), overhead is negligible (~1.2x). For tight loops calling tiny functions millions of times, it can reach ~7x. Relative rankings are always reliable — use lazyline to find *which* code is lazy, not to measure *how fast* it runs. See [benchmarks](https://github.com/TomasVenkrbec/lazyline/blob/main/benchmarks/README.md) for detailed measurements. ## Usage ```bash # Profile a package during its test suite lazyline run my_package -- pytest tests/ # Profile while running a script lazyline run my_package -- python evaluate.py # Profile a CLI tool (hyphenated console scripts work too) lazyline run my_package -- my-tool run-all # Export results, view later lazyline run -o results.json my_package -- pytest tests/ lazyline show results.json --top 10 # Multiple scopes in one run lazyline run utils.py my_package -- python script.py ``` Requires Python 3.10+. The target package must be importable in the same environment. See the [full usage guide](https://github.com/TomasVenkrbec/lazyline/blob/main/docs/usage.md) for all CLI options, scope formats, command resolution, output details, and more examples. ## Claude Code Plugin Lazyline ships as a [Claude Code plugin](https://docs.anthropic.com/en/docs/claude-code/plugins). Install it and Claude will know how to profile your code, interpret results, and suggest optimizations: ```bash /plugin marketplace add TomasVenkrbec/lazyline /plugin install lazyline@lazyline ``` Then use `/lazyline my_package -- python main.py` or let Claude invoke it automatically when you ask about performance. The skill also works with any AI coding assistant that supports markdown skill files — copy [`skills/lazyline/SKILL.md`](skills/lazyline/SKILL.md) into your assistant's configuration. ## Documentation - **[Usage Guide][docs-usage]** — CLI reference, scope formats, output details - **[How It Works][docs-how]** — architecture, overhead, limitations - **[Benchmarks][docs-bench]** — overhead measurements and methodology - **[Contributing][docs-contrib]** — development setup, tests, code style - **[Changelog][docs-changelog]** [docs-usage]: https://github.com/TomasVenkrbec/lazyline/blob/main/docs/usage.md [docs-how]: https://github.com/TomasVenkrbec/lazyline/blob/main/docs/how-it-works.md [docs-bench]: https://github.com/TomasVenkrbec/lazyline/blob/main/benchmarks/README.md [docs-contrib]: https://github.com/TomasVenkrbec/lazyline/blob/main/CONTRIBUTING.md [docs-changelog]: https://github.com/TomasVenkrbec/lazyline/blob/main/CHANGELOG.md --- If lazyline helped you find a bottleneck, consider giving it a [star](https://github.com/TomasVenkrbec/lazyline) — it helps others discover the project. Found a problem? [Open an issue](https://github.com/TomasVenkrbec/lazyline/issues). ## License MIT — see [LICENSE](https://github.com/TomasVenkrbec/lazyline/blob/main/LICENSE).