TL;DR
- Re-pasting conventions every session is wasted work and it drifts.
- Anthropic's own docs say context is the constraint — and that a
CLAUDE.mdplus on-demand skills is how you give it.- A
coding-standardsskill encodes your rules once; pair it withtdd-workflowso new code arrives tested, not just styled.
The tax you pay every morning#
Open Claude Code, start a task, and watch what you type before any real work happens: "use our import ordering, prefer named exports, no default exports, tests go next to the file, 2-space indent, we use Vitest not Jest." Every session. Every project. Every teammate, slightly differently.
That is setup tax. It is slow, it is inconsistent across the team, and it silently drifts — the conventions you pasted last month are not quite the ones you paste today, and Claude has no way to know which is current.
This is a context problem, and Anthropic says so#
It is tempting to think the fix is a smarter model. It isn't. On SWE-bench Verified — a benchmark of real GitHub issues — Claude Opus 4.5 scores 80.9%, the first model past 80%, with Claude Sonnet 4.5 at 77.2%. The model can already resolve four out of five real-world issues. Raw ability is not your bottleneck. Whether it knows your conventions is.
Anthropic's best-practices guide for Claude Code is blunt about where that knowledge belongs:
"CLAUDE.md is a special file that Claude reads at the start of every conversation. … This gives Claude persistent context it can't infer from code alone."
— Anthropic, Best practices for Claude Code
So the rules go in a file the agent reads automatically — not in a paragraph you retype. But there's a catch the same guide is equally blunt about:
"Bloated CLAUDE.md files cause Claude to ignore your actual instructions!"
— Anthropic, Best practices for Claude Code
A CLAUDE.md that tries to hold every convention, framework note, and edge case becomes noise, and the agent starts dropping the rules that matter. The guide's own resolution is to split the load: keep CLAUDE.md lean, and move the sometimes-relevant detail into skills that load only when needed.
"For domain knowledge or workflows that are only relevant sometimes, use skills instead. Claude loads them on demand without bloating every conversation."
— Anthropic, Best practices for Claude Code
Encode it once, as a skill#
A skill is a folder of instructions Claude discovers and loads when the task calls for it. When Anthropic shipped Agent Skills in October 2025, Barry Zhang, Keith Lazuka, and Mahesh Murag described them as "organized folders of instructions, scripts, and resources that agents can discover and load dynamically to perform better at specific tasks." The design uses progressive disclosure — the summary costs a few tokens, and the full detail loads only when relevant.
That is exactly the shape your coding conventions want. The coding-standards skill in cyberg7-skills is your import ordering, export style, file layout, and test runner — written down once, loaded automatically, and not clogging every conversation's context when you're doing something unrelated.
Instead of pasting a paragraph of rules, you install the skill and Claude follows them without being told again.
Make it stick with tests#
Style is only half the problem. The other half is that "looks right" is not "works right." Pair coding-standards with tdd-workflow: now Claude writes the test first, watches it fail, then writes the minimal code to pass — in your conventions.
The combination is the point. One skill makes the code look like your team wrote it; the other makes sure it behaves before it ships. Neither requires you to re-explain anything tomorrow. (If you want the deterministic version of "this check must always run," that's a job for hooks — see The Stop-hook loop that ran for six hours.)
How to start#
- Install the free starter pack —
coding-standardsis in it. - New to the library? Start with Your first 20 minutes with cyberg7-skills.
- Add
tdd-workflowfrom the full library when you want tested-by-default output — see pricing. - Run
csm syncwhenever we ship updates; your conventions travel with you across every machine.
More on repeatable ways of working in the Workflows pillar.