writing-style
Apply consistent writing style to commits, PR descriptions, issue bodies, and inline code comments. Invoke before writing any user-facing text to ensure it matches the project's established voice.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-15
- Stars
- ★ 4
- Forks
- 1
01 前言 Front Matter
- name
- writing-style
- description
- Apply consistent writing style to commits, PR descriptions, issue bodies, and inline code comments. Invoke before writing any user-facing text to ensure it matches the project's established voice.
03 SKILL.md 正文
<!-- This skill lives in bundles/base/skills/writing-style/ so it's available in every session (base bundle fires for user-alice always).
SKILL.md files are injected as agent context when the skill is invoked.
The frontmatter name and description fields control how llmenv
materializes the skill as a plugin slash-command.
This particular skill encodes a personal writing voice so that Claude produces commits and PR descriptions that sound like the author wrote them, not like a language model summarizing a diff. Adapt the sections below to your own style.
HOW IT CONNECTS TO THE REST OF THE SYSTEM:
- Invoked explicitly: user types /writing-style or Claude invokes it
when writing commit messages or PR descriptions.
- The skill is loaded from this SKILL.md into the agent's context.
- It doesn't require any special hooks or MCP servers — it's pure
context injection.
-->
Writing Style Guide
Use this guide whenever writing commits, PR titles/bodies, issue descriptions, or inline comments. The goal is prose that reads as if a senior engineer wrote it in a Slack message: direct, specific, no ceremony.
Core Principles
- Direct: lead with the fact, not a preamble.
- Specific: name the thing, not "the functionality" or "the component".
- Brief: one sentence is better than three when one works.
- Technical but not jargon-heavy: write for a peer who knows the system, not for a documentation audience.
Commit messages
Format: imperative mood, ≤72 characters, present tense.
Add rate limiting to the ingest endpoint
Fix off-by-one in page cursor calculation
Drop unused `legacy_auth` middleware
What NOT to do:
# Too long, passive voice, vague
Added some changes to fix the issue with the rate limiting that was
causing problems in the endpoint
# Marketing speak
Implement robust, enterprise-grade rate limiting solution
The subject line is a command to the codebase: "Do X." If the subject needs more than ~72 chars, the change is probably too big for one commit.
PR descriptions
Lead with what the code does now, not what you changed or why you changed it.
Good:
The ingest endpoint now rejects requests above 100 req/s per API key with a 429. Burst headroom is 20 extra requests; the window is a sliding 60s.
Bad:
This PR adds rate limiting. I decided to use a sliding window because token bucket was too complex. I also considered leaky bucket but rejected it.
Skip: "In this PR", "This change", "As part of this work". Start with the system behavior.
Issue descriptions
State the problem, not the investigation. Assume the reader is a peer who will act on it.
Good:
cargo teston a fresh checkout fails witherror[E0432]: unresolved importoncrate::auth::Token. Introduced in #314. Bisect points to the rename insrc/auth/mod.rs:47.
Bad:
I was trying to run the tests and encountered an error. After some investigation I believe there might be an issue with the imports.
Inline code comments
Only write a comment when the WHY is non-obvious. Never explain WHAT.
// ponytail: global lock; per-account locks if throughput matters
let _guard = GLOBAL_LOCK.lock().await;
Not:
// Lock the global lock to ensure thread safety
let _guard = GLOBAL_LOCK.lock().await;
Tone markers
- OK: "This breaks X", "Avoids the O(n²) scan", "Skipped Y — only matters if Z"
- Not OK: "This enhancement improves", "Leverages", "Robust", "Seamless"
- Hedging: only when genuine uncertainty exists. "Probably" and "might" are fine when you mean them; drop them when you're certain.
04 同仓库更多 Skills
clippy
Rust clippy linter skill
hello
Hello world skill
setup-llmenv
Interactive setup wizard for new llmenv users. Evaluates existing Claude Code configuration, helps sort into bundles, and writes llmenv-native equivalents.
llmenv
How to use llmenv's built-in features (task tracker, memory, context-mode, codebase-memory) effectively. Load this first; it points to a reference file per enabled feature rather than dumping all of them into context.
fyi
Use when the user asks for an "fyi", "morning todo", "morning briefing", "what's on my plate", "what do I have open", "what's in flight", "where did I leave off", "what needs my attention today", or "my open work across projects". Surveys the user's currently-open, in-flight, and pending work across GitHub, Linear, Slack, Pylon, and Notion, then produces a ranked, GitHub-markdown briefing for the day. Forward-looking sibling of daily-update (which reports what already happened).
daily-update
Use when the user asks for a "standup", "scrum update", "daily update", "status update", "what did I work on", or "what did I do yesterday". Scans llmenv session logs, Linear, Slack, and GitHub for the user's own activity on a given day and produces a short scrum-style update, copied to the clipboard as Slack markdown.