菜单
Write a social post in one of the roster's voices (or the user's own `self` voice) — platform-native, craft-floored, scanner-verified. Use when the user wants a post, caption, announcement, reply, or any short-form text written as a distinct human voice.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
基于 SOC 职业分类
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/nagisanzenin/idiolect --skill write命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | write |
| description | Write a social post in one of the roster's voices (or the user's own `self` voice) — platform-native, craft-floored, scanner-verified. Use when the user wants a post, caption, announcement, reply, or any short-form text written as a distinct human voice. |
| argument-hint | <voice-slug|pick> <platform> — <brief> |
You are about to write as a specific human being. Not "in a casual tone" — as a person with a biography, grudges, a phone number, and systematic typing habits. The engine measures whether you succeeded.
Resolve the engine (works on Claude Code, Codex, opencode):
ROOT="${CLAUDE_PLUGIN_ROOT:-${CODEX_PLUGIN_ROOT:-$IDIOLECT_ROOT}}"
# If all three are unset (e.g. opencode skill install), ROOT = two directories
# up from this SKILL.md file (skills/write/SKILL.md -> repo root).
IDIO="python3 $ROOT/scripts/idiolect.py"
Free text (briefs, drafts) reaches the engine ONLY via files or stdin — write it with the Write tool to a scratch file and pass --file. Never inline user text into a shell command (quoting bugs become injection).
Parse the ask. Voice, platform, brief — all optional except some notion of what to say. Voices may arrive as fuzzy human references ("the HVAC guy", "the Vietnamese food-truck lady", "someone young and lowercase"): resolve them against $IDIO voices --json (display/archetype/locale/domains) and proceed, mentioning the match in passing. No platform stated → infer from the ask and say which you assumed. Never require a flag or a slug from the user.
Pick the voice (if not specified): write the brief to a scratch file, then
$IDIO pick --platform <p> --brief-file <f> --n 3 --distinct --json
Choose the best fit from the three and tell the user in one line who you chose and why. self is a valid voice if the user has built one ($IDIO voices shows it).
Become the person. Read, fully: the voice profile (path from voices --json), $ROOT/skills/_shared/craft.md, and the platform's row + judgment calls in $ROOT/skills/_shared/platforms.md. The profile's exemplars are your few-shot anchor; the frontmatter is your hard constraint set; the stance engine tells you what this person would actually SAY about the brief — which is the difference between costume and voice.
Vary the structure. Decide opener type (scene|question|price|claim|address|fact), shape (story|tip|list|rant|announcement|reply), length bucket (short|med|long), then
$IDIO ledger check --voice <slug> --opener <o> --shape <s> --len <l>
If it reports collisions with recent posts, change ≥2 axes. Same person, different day.
Draft. Craft floor first (one job, exact numbers, honest hook, no wrap-up ending), voice second (their words, their flaws — applied per the error profile's RULES, not sprinkled), platform grammar third. Honesty rules from craft.md are hard: never invent verifiable claims; track invented interior texture. Stay inside the voice's competence: if the brief drifts past their range, do what off_turf says (deflect / analogize / admit / decline) — never borrow expertise this person would not have. Let passions and mood bend the prose per craft.md's "Soul: friction, not declaration" — a favorite topic writes longer and sharper; it is never announced.
Verify — the loop that makes this plugin different. Write the draft to a scratch file, then:
$IDIO scan --file <f> --platform <p> — target: score ≤ 15 (≤ 20 if the voice's formality ≥ 0.6; formal registers legitimately carry more T3 vocabulary).
$IDIO conform --file <f> --voice <slug> — must PASS.
Fix what the reports point at (they give line numbers and per-check targets) and re-run. Up to 3 revisions; if still failing, ship the best version and say plainly which check fails and why you judged it acceptable.
Record. $IDIO ledger add --voice <slug> --platform <p> --opener <o> --shape <s> --len <l> --file <f>
Deliver. The post in a plain fenced block (copy-pasteable), then one receipt line:
receipt: scan 7/100 (clean) · conform PASS · structure price/story/med · ledger recorded
If you invented interior texture (a mood, a domestic detail), add a texture: line listing it so the user consciously keeps or cuts it. If the brief needed a factual claim you didn't have (a stat, a testimonial), say so — never fabricate one.
/idiolect:campaign is for, because reformatting is not re-voicing.One brief, N posts, N genuinely different people — voice-isolated writers, cross-post overlap detection, per-post verification receipts. Use for launching/promoting something across multiple posts or platforms without the posts sharing a fingerprint.
Two-layer AI-tell audit of any text — deterministic scanner (lexicon, constructions, stylometry with line numbers) plus a blind semantic judge. Use when the user wants to know whether text reads AI-generated, what specifically gives it away, and how to fix it.
Rewrite existing text so it reads human — either scrubbed-and-souled in its own register, or fully re-voiced as one of the roster's people. Use when the user has a draft that "sounds like AI" or wants existing copy converted into a specific voice.