Units-LE
io.github.nolindnaidoov1.0.0更新于 Oct 4, 2026
Extract durations, sizes, percentages and frequencies in one base unit, refusing ambiguous ones.
概览
从文档中提取时长、大小、百分比和频率,并以统一基准单位报告,对含义不明的量按名称拒绝。
- 功能
- 提供 extract_units 工具,扫描传入内容中的量(数字与单位相连),返回其原文、基准值、行、列和键路径。时长归一化为毫秒,大小归一化为字节,百分比归一化为比率,频率归一化为赫兹。对于 500m、1.5KB、1,5s 等有歧义或格式异常的量,保留其行并给出具名原因,而不是猜测数值。它不会改写文档,也不做归一化之外的算术或单位换算。
- 适用场景
- 适合在审查配置文件、values 文件、运行手册或保留策略时使用,尤其是超时、内存限制、百分比或频率以不同写法混用的场景。它有助于比较不同服务中以不同方式书写的量,并标出读者可能误读的条目,例如 500m 或 1MB。
- 运行要求
- 通过 npx units-le-mcp 以 stdio 在本地运行,需要 Node.js。无需环境变量、API 密钥或额外配置。它以参数接收内容,不读取文件,也不发起网络请求。
安装
在 SourceWeft 中
- 打开 控制台中的 Units-LE,将其添加到工作区。
- 为需要使用其工具的对话启用该服务。
Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。
其他 MCP 客户端
参照 仓库 中的启动说明。
README
Units-LE: One Quantity, One Unit
Find every quantity in a document — durations, sizes, percentages, frequencies — with its value in one base unit, and refuse the ambiguous ones by name
YAML · TOML · JSON · INI · dotenv · CSV · and anything else, scanned as text
[Install from VS Code Marketplace] [Open VSX downloads] [units-le-mcp on npm] [units-le on crates.io] [LE Tools]
Useful? A star or rating is how other developers find it — ★ GitHub · ★ Open VSX · ★ Marketplace
What it does
timeout: 30s in one service and timeout: 30000 in another are the same number of milliseconds, and nothing about the text says so. memory: 1GB and memory: 1GiB differ by 7% and look identical at a glance.
Press Ctrl+Alt+Q (Cmd+Alt+Q on Mac) and every quantity in the active document — a number welded to a unit — opens in a report beside the editor: as the document wrote it, in one base unit so two of them can be compared, with its line, column and key path. A quantity it cannot read unambiguously keeps its row and gets a reason instead of a guess. Works in VS Code and in VS Code–based editors like Cursor and VSCodium (installable from Open VSX).
- Before a migration — every timeout and memory limit in a values file, in milliseconds and bytes
- Reviewing a config change — the
500mthat means minutes to one reader and millicores to another - Against a runbook or a retention policy — the limit the document states, checked against the one the config sets
It never rewrites a document, and never guesses.
Install
A refusal is a finding
This is the whole product. Any tool can multiply 30s by a thousand;
what makes this one usable in a review is that it will not pretend.
Every one of them keeps its row, its source text, its position and a sentence a person can act on. A refusal is never a dropped row and never a guess.
si_iec_hazard annotates; it does not withhold
It is the one reason that still answers. MB is 10^6 by the standard,
so 1MB comes back as 1000000 bytes and carries
reason: "si_iec_hazard" — because a great deal of software writes
MB and means 2^20. Reporting nothing would be less useful than
reporting the standard answer with the flag attached, and picking 2^20
would be the guess this tool exists not to make.
So it is not counted as refused. A row with si_iec_hazard is the
only row that carries both a base and a reason; every other reason
means the base is null.
A bare number with no unit is not a finding at all. timeout: 30
yields nothing here — that is
numbers-le's question, and
the boundary is what keeps the two tools distinct.
Four dimensions
Case is part of the symbol. MB is a megabyte and Mb is a
megabit; folding them would silently multiply by eight, so Mb is
refused rather than treated as a synonym.
1h30m is a grammar and 1h + 30m is arithmetic. The first is one
quantity written in two parts. Inside a compound the parts disambiguate
each other — m alone is refused, and m between h and s can only
be minutes, because a compound is written largest-first. That ordering
is required rather than assumed, so 30s1h is not a quantity at all.
Physical units are a non-goal, not a gap. Length, mass, temperature
belong to a different question, and reading them would mean unit algebra
(m/s²), which is what uom is for. See
crate/SPEC.md.
Formats
JSON, YAML, CSV, TOML, INI and dotenv are parsed. Everything else is scanned as text — a Kubernetes manifest, a Terraform file, a Markdown table of limits, a log — so the files where quantities actually live yield them rather than nothing.
Unlike numbers-le's text scan, this one has a shape to look for, so
v1.2.3 yields nothing rather than two numbers. Every format with a
shape carries a key path — cache.ttl, limits[0], server.timeout,
TIMEOUT — and every finding carries a 1-based line and column, the
column in UTF-16 units, which is the number your editor shows.
Its false-positive class is measured rather than imagined. A run
inside an opaque blob still reads as a quantity: 001d in a UUID is one
day, and a base64 hash ending /2w== is two weeks. The boundary
characters that let those through are the same ones that let -30s and
ttl=30s through. Over crate/fixtures/documents/opaque.txt — 280
lines of lockfile hashes, container digests, git object names, UUIDs and
signing material, none of which is a quantity — the scan reports 5
false findings, 1.8%, and a test prints that number on every run.
Each one carries its line and column, which is what makes it a row you
discard rather than a number you trust.
It has no opinions
No "this timeout is too low". No defaults database. No conversion flag, no rewriting, no arithmetic. It reports what a document says and what that means in one base unit; which limits are right is the reviewer's call.
Use it from an AI agent
The same engine runs as an MCP server, so an agent can read quantities directly instead of converting units by hand.
It returns the report the editor renders, as data — quantities capped at 500 by default with meta.truncated. It reads no files and makes no network requests. Published as units-le-mcp on npm and as io.github.nolindnaidoo/units-le in the MCP registry. It answers exactly as the Rust CLI's server does: one corpus runs against both, and a differential test feeds both thousands of generated documents in every format — broken ones included, so each parser's error text is compared too.
Configuring it by hand — any host with an MCP config file
Or install it once with npm install -g units-le-mcp and point at units-le-mcp. It needs no environment variables, no API key and no configuration of its own. To check it:
The CLI
The same extraction runs over a tree from a terminal or a CI step: a Rust CLI in crate/, sharing one corpus with the extension — crate/fixtures/ — so the two can never read a quantity differently.
Exit codes follow grep — 0 quantities found, 1 none found, 2 the question was malformed. A refusal does not change the exit code; --strict makes it one.
Commands
Settings
Languages
Twelve languages besides English:
German · Spanish · French · Indonesian · Italian · Japanese · Korean · Portuguese (Brazil) · Russian · Ukrainian · Vietnamese · Chinese (Simplified)
Both halves are covered — the manifest (command titles, setting names and descriptions) and everything shown while the extension runs (notifications, the status bar and the report's headings). A refusal's sentence and a parser's error are the engine's English, identical to the CLI's.
Privacy & security
- No network access. The extension never sends data anywhere. The
telemetryEnabledsetting only writes events to a local Output Channel you can inspect (Units-LE). - It reads the active document and nothing else, and never writes to it.
- The MCP server holds the same line. It takes content as an argument and returns data: no filesystem access, no network calls, no telemetry.
- Error notifications redact home directories and credential-shaped fragments.
Documentation
Performance
Median of 7 runs after warmup, on Apple M5 Pro, 24 GB RAM, Node 24.3.0. Inputs are generated
by scripts/benchmark.ts rather than checked in, so the sizes above are
exactly what was measured. Reproduce with bun run benchmark.
These are machine-specific and are not asserted in CI — a benchmark that gates a build only tells you how busy the runner was.
Testing
531 test cases across 11 files, plus an integration suite that runs
in a real VS Code extension host and an end-to-end test that installs the
built .vsix into a clean profile.
Generated from a real run — coverage/coverage-summary.json and
coverage/test-results.json — by scripts/coverage-readme.js; CI fails if
this section drifts. Reproduce with bun run test:coverage, and the case
count is the one vitest prints.
More from the LE family
Sixteen single-purpose tools for the work in front of every model. Each ships a Rust CLI and an MCP server. One page: letools.dev
Get it out
- String-LE — Extract every string in a codebase, with its position, so a person can read them
- Numbers-LE — Extract every hardcoded number in a codebase, so a person can check them
- Units-LE — Extract every quantity with its unit, normalized, and refuse the ambiguous ones by name
- Dates-LE — Extract every date and timestamp, and the exact instant each one resolves to
- IDs-LE — Extract every UUID, ULID, NanoID, ObjectId and Snowflake, and decode the time inside
- IPs-LE — Extract every IP address, CIDR block and MAC, normalized and classified by scope
- URLs-LE — Extract every URL in a codebase, with its protocol and exact position
- Paths-LE — Extract every file path in a codebase, and say whether it still points at anything
- Colors-LE — Extract every color in a codebase, and say which ones are not in your palette
Check it
- Regex-LE — Find every regex in a codebase, and report which can be driven into catastrophic backtracking
- Versions-LE — Find where one dependency is constrained differently across a repository's manifests
- i18n-LE — Identify the i18n library a project uses, then audit its catalogs by that library's rules
- Scrape-LE — Check whether a page is scrapeable before the scraper is written, and say when it cannot tell
Guard it
- Secrets-LE — Find hardcoded credentials in a codebase, and never print one into the report
- EnvSync-LE — Compare the dotenv files in a tree, and say which keys are missing from which
- Unicode-LE — Find the Unicode that hides meaning — bidi controls, invisibles, homoglyphs, mixed scripts
Each stands on its own: no shared crate, no published core. Where two of them agree, it is because the same answer was right twice.
Contact — nolindnaidoo.com · GitHub · LinkedIn
Also by nolindnaidoo
Rust — pixelcoords and pixelactions are one loop: pixelcoords answers where, pixelactions acts there. Their own tools, their own voice — not part of the LE family.
- pixelcoords — Freeze your screen, mark regions, get pixel-exact coordinates and crops pixelcoords.dev · crates.io · docs.rs
- pixelactions — Consume human-verified coordinates, perform the interaction, confirm it landed pixelactions.dev · crates.io · docs.rs
License
MIT © nolindnaidoo
来源:README.md,提交 b6c320b
工具
0版本历史
1- v1.0.0最新Oct 4, 2026

