本文へ移動
cccskills
無料GitHub で公開

habit-words

Learns the words a user habitually writes to their coding agent that make its answers worse ("be 100% sure", "don't assume", "full review", "are you sure?", "all fine, right?"), from what they actually typed in their recent Claude Code sessions. Shows what each one does, what went wrong after it, and what to say instead, then writes the habit words block that maps each word to the checks it should trigger, so the checks run without the words. Use when asked to learn, read, refresh or review the user's habit words or prompts, "which of my words make you worse", from setup-first-pass's profile step, or when the start-of-session check says the habit words are due.

joetawil7/first-pass
インストール方法を見る

含まれるファイル（1）

  • SKILL.md6.5 KB

SKILL.md（原文）

インストールする前に、エージェントに与えられる指示の中身を確認できます。

habit-words

"Be 100% sure" or "make it bug free" names no place to look, so it changes how sure the answer sounds, not what gets checked. This skill finds the words this user writes, from their own prompts, and maps each to the checks it should mean. The user never has to write them again, and when they do, the words trigger the checks instead of a more certain tone.

The block this skill writes follows ${CLAUDE_SKILL_DIR}/../setup-first-pass/assets/words-block.md: read it first, and keep its markers and shape.

1. Read the sessions

Plugin: node "${CLAUDE_SKILL_DIR}/../../scripts/cli.mjs" words reads the user's last 20 Claude Code sessions that hold a prompt they typed (--sessions <n> for another number), from ~/.claude/projects (or $CLAUDE_CONFIG_DIR/projects). It writes only what they typed to a temp file and prints the file's path, what it left out, the newest prompt's time, and how many prompts hit each phrase family. Left out: tool output, pasted text, notifications, messages from other sessions, skill and system text, commands, compaction summaries and turns a script started (claude -p). Prompts queued while the agent was busy are kept: they are often the corrections. The one thing kept that the user did not type is the end of the agent's reply before each prompt tagged "possible pushback". Anything that looks like a key, token, password, email or phone number is replaced: best effort, so never quote a line that still looks like one. Each run writes one new file; note every path.

Without the plugin (another tool), ask the user to paste 30 to 50 recent prompts, and work from those.

Say the coverage before the findings: sessions, dates, prompts, what was left out. Read the whole file. Past about 800 prompts, read every prompt the families or the pushback marker tag, plus a sample of 150 of the rest, and say so.

2. Find the words

  • The families are candidates, not verdicts. Keep a phrase where it asks for certainty or completeness without naming a check ("be 100% sure", "cover all cases"); drop it where it names one ("make sure the tests pass", "check if this is still true").
  • Find the user's own phrases the families miss, by reading: stacked demands ("dont assume, be 100% confident, cover all gaps" in one breath), yes-shaped questions, absolutes ("no hacking is possible"), frustration, open grants, and the ones that work well.
  • Count every phrase you report: node "${CLAUDE_SKILL_DIR}/../../scripts/cli.mjs" words --count "<regex>" (repeat --count for several; one read), with the same --sessions as step 1, so the counts cover the prompts you read. Every number shown comes from a count, never from reading. Prompts where the user talks about the words (like a request to run this skill) are counted too: say how many.
  • The impact: a prompt tagged "possible pushback" shows the end of the reply before it. A habit word followed, in the same session, by the user correcting the result ("reread your numbers", "that's wrong", "is it really working?") is the evidence. Quote at most a few words of a prompt, never a whole one, and never names, contacts, amounts or anything secret.

3. Map each word to checks

For each word keep:

  • what the user writes, in their spelling;
  • how often (prompts, sessions), from a count;
  • what it does to an agent, in one line (why the answer gets worse, not better);
  • what happened after it, with the date, where the sessions show it;
  • the check it now triggers: the working rule that answers it, by its name (Evidence, the pre-mortem, Done means proven, Say the coverage, No yes by default, Pushback gets checked, Bugs: reproduce first, Grants stay narrow, One step at a time), or, when no rule covers it, one new instruction in the same style;
  • what to say instead: the words that name the check.

Also list the prompts that worked (they named the check, the source or the scope), so the user keeps them.

4. Show the user

A table: You wrote · How often · What it does · What happened · Say instead, the prompts that worked, then the block you will write. Ask before writing it.

5. Write the block

With a yes, write the block between its markers: the heading, one line saying the words mean the checks and never a more certain tone, and one bullet per word: "<the user's words>" (<n> prompts): <the check it triggers>. The start marker is

<!-- first-pass:words:start v<plugin version> through <newest prompt's time from step 1> from <n> sessions (written by the habit-words skill from the user's own prompts; re-run it to refresh) -->

The start-of-session check reads through to say when the words are due again: once 20 sessions the user typed in have changed since. If the user says "not now" to a refresh, set through in the existing marker to the current time and change nothing else.

Where: never in a file teammates share, because these are one person's words. So:

  • in a main folder, right after the profile block in the root AGENTS.md;
  • in a single repo, or wherever the profile block sits in a file the repo commits, in ~/.claude/CLAUDE.md instead;
  • an existing words block is replaced where it is, if that file is not shared; otherwise it is moved to ~/.claude/CLAUDE.md, and the report says so.

Never edit outside the markers.

6. Clean up

Delete every file step 1 wrote (each words run writes one): they hold the user's prompts. node "${CLAUDE_SKILL_DIR}/../../scripts/cli.mjs" words --delete "<file>" for each; it fails if the name matches no file, so a mistyped name never reads as deleted. A file an unfinished run left is removed by the next run once it is 6 hours old. The report ends with:

Read: <n> prompts from <n> sessions, <first date> to <last date>; left out <what>
Words: <n> mapped (<n> new instructions), block <written to file | not written, because>
Temp files: <n> written, <n> deleted

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

ログインしてレビューを書く

同じリポジトリのスキル

概要と使いどころ

fix-the-class

無料

Bug-fix routine that fixes the whole class of bug, not just the reported instance. Use for any bug report, failing production behaviour, monitoring alert, audit finding or review finding that does real harm (a smaller review finding goes on the item's list, as ship-check says), and whenever the same kind of bug has been seen before. Reproduces it with a failing test, names the failure class, searches the task's scope for the same pattern (the whole codebase when the user's own prompt asks for the fix, not when it is picked from a task's list), fixes or records every hit, and adds the rule, invariant, helper or check that stops it coming back.

日本語の概要は準備中です。原文の説明を表示しています。

joetawil7/first-pass992026年10月10日 更新

hulk

無料

Lifts the task's scope for one task. The session, its reviews and fix-the-class then look across the whole codebase and every repo that shares its code, data or vendors, and every problem found anywhere is handled by the usual rules. Runs only when the user types it.

日本語の概要は準備中です。原文の説明を表示しています。

joetawil7/first-pass992026年10月10日 更新

jev

無料

Sets up, checks or changes the optional Jev judge, the TypeSafe decision model that ship-check asks about review findings (is this real harm, is a small one worth fixing now, what proof does a small fix need). Shows whether it is on for each repo, which key it uses, switches it on or off, and sends one test request. Runs only when the user types it.

日本語の概要は準備中です。原文の説明を表示しています。

joetawil7/first-pass992026年10月10日 更新

premortem

無料

Run before writing or changing code for any feature, bug fix or refactor that touches data, jobs, money, outside services or user-facing behaviour. Finds where the change will break before it is built, by walking ten named failure classes (twice, halfway, outside call, failure-is-not-empty, neighbors, endings, money, hostile user, words, scale and time) against the code, and writes the answers into the plan. Use when planning a change, when asked to "think it through", "cover the gaps", "make sure it's bug free", or before any edit to payment, publishing, deletion, auth or background-job code.

日本語の概要は準備中です。原文の説明を表示しています。

joetawil7/first-pass992026年10月10日 更新

review

無料

Reviews a change before it is merged, from little or no input. With nothing typed it reviews the branch you are on, uncommitted work included, against its base, its PR and its ticket, so a developer can check their own work before asking for review. With PR links, repo#N or a branch it reviews someone else's change the same way, several repos at once. It judges the failed CI checks and every Bugbot and reviewer comment, gets a fresh breaker review, traces every other piece of code that uses what changed, proves each finding or marks it unproven, says what the merge needs (a build, env vars, migrations, deploy order) and how to check the deploy, never posts to GitHub, and offers fixes only after the report. Runs only when the user types it.

日本語の概要は準備中です。原文の説明を表示しています。

joetawil7/first-pass992026年10月10日 更新

setup-first-pass

無料

Set up first-pass for a main folder that holds many repos, or for one repo. In a main folder it writes one set of working rules at the root that every session loads, a workspace map, and a section in each repo with that repo's real commands, test limits and invariants, and it lists each repo's own hooks so first-pass can run them from the main folder. In one repo it writes the rules and the section into that repo. Use when asked to set up, install, add, update or re-run first-pass, to "add the shipping rules", or when the start-of-session check says something drifted. Re-run to update.

日本語の概要は準備中です。原文の説明を表示しています。

joetawil7/first-pass992026年10月10日 更新

joetawil7 のスキルをすべて見る

このスキルの問題を報告する