From FORSVN
Polishes translated Vietnamese text to read natively in a target register (báo chí, semi-casual, bro, or pop-marketing). Fixes pronoun drift, particles, idioms, passive calques, typography, and AI translation tells.
How this skill is triggered — by the user, by Claude, or both
Slash command
/forsvn:polish-vn [vietnamese text or file path] [--register bao-chi|semi-casual|bro|pop-marketing][vietnamese text or file path] [--register bao-chi|semi-casual|bro|pop-marketing]This skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
*Communication — Horizontal. **Terminal** polish for Vietnamese-market copy. Three-agent pipeline (diagnose → polish → critic) for register **calibration**, idiom + **cultural** reference repair, and removal of **ai-translation** tells.*
agents/critic-agent.mdagents/diagnostic-agent.mdagents/polisher-agent.mdreferences/_shared/before-starting-check.mdreferences/_shared/execution-policy.mdreferences/_shared/manifest-spec.mdreferences/_shared/mode-resolver.mdreferences/_shared/pre-dispatch-protocol.mdreferences/anti-patterns.mdreferences/examples/vn-tone-walkthrough.mdreferences/format-conventions.mdreferences/playbook.mdreferences/procedures/absolute-prohibitions.mdreferences/procedures/dispatch-mechanics.mdreferences/procedures/pre-dispatch.mdreferences/translation-artifacts.mdreferences/vn-tone-corpus.mdrouting.yamlscripts/append-loop-result.tsscripts/bootstrap-experience.tsCommunication — Horizontal. Terminal polish for Vietnamese-market copy. Three-agent pipeline (diagnose → polish → critic) for register calibration, idiom + cultural reference repair, and removal of ai-translation tells.
Core Question: "Would a native VN writer of this register write this exact text — and would a native reader stumble anywhere?"
Why / philosophy / when NOT to use:
references/playbook.md[PLAYBOOK].
humanmaxxing for source tone, or any MT).bao-chi / semi-casual / bro / pop-marketing. If ambiguous, ask first.Critic 7-check rubric (zero Hard Tells from 28-pattern catalog · pronoun pair held · particle density in range (0% báo chí; 15–25% casual/pop/bro) · facts preserved · typography normalized · read-aloud clean · score ≥28/36). Full rubric: agents/critic-agent.md. Pattern catalog: references/translation-artifacts.md.
Absolute Prohibitions (any one = auto-FAIL): em dashes · quý khách/quý vị outside formal · title-case headlines · particles in báo chí · pronoun drift · dropped facts · cliché stacks · literal idiom calques. Enforcement: references/procedures/absolute-prohibitions.md [PROCEDURE].
Per references/_shared/before-starting-check.md — load brand voice + register-mapping experience.
| Artifact | Source | Required? |
|---|---|---|
research/product-context.md | research-icp | Recommended — brand voice → register inference |
docs/forsvn/artifacts/marketing/content/[slug].md | upstream | Optional — register from frontmatter |
docs/forsvn/experience/brand.md | (any) | Optional — Brand — VN target register key |
Protocol: references/_shared/pre-dispatch-protocol.md. Needed dimensions: register · dialect (north/south/neutral) · subvariant (if bro: otofun/voz). Read-order + Register Resolution priority + Warm/Cold Start + write-back + hard-blocks: references/procedures/pre-dispatch.md [PROCEDURE].
Per references/_shared/mode-resolver.md — --fast runs single-pass polisher only (skips diagnostic + critic). --fast does NOT skip Register Resolution, Critical Gates 1-4, or Absolute Prohibitions.
Session execution profile (single-vs-multi): inherit per references/_shared/execution-policy.md.
3 sub-agents in agents/: diagnostic (L1: scan, register gap, violation log) → polisher (L2: register-correct rewrite) → critic (L2 final: 3-pass, 36-point rubric).
Route A (slash-command): Pre-Dispatch → diagnostic → user checkpoint → polisher → critic. FAIL → re-dispatch polisher (max 2 cycles) → deliver.
Route B (upstream caller — brief-shortform / write-social / write-copy / write-ad when market = VN): trust caller's register; if prior polish → skip diagnostic; no standalone artifact.
Spawn mechanics, fallback, checkpoint, pipeline, rewrite loop: references/procedures/dispatch-mechanics.md [PROCEDURE].
docs/forsvn/artifacts/marketing/content/[slug].vn-tone.md · (B): none, embedded in callerpipeline — re-run renames to [slug].vn-tone.v[N].mdskill, version, date, status, target_register, subvariant, dialect, critic_scorepolish_chain_applied: vn-tone + critic_score: N/36format-conventions.mdTemplate + per-section format rules: references/format-conventions.md [PROCEDURE].
references/anti-patterns.md [ANTI-PATTERN] — 11 pipeline + 4 cross-cutting rows. Most common: pronoun drift · em dash · cliché stack · particle over-injection.
Hand polished text to caller (B) or user (A). On DONE_WITH_CONCERNS, surface concerns before publish.
Route A walkthrough + cycle-2 FAIL + --fast variant: references/examples/vn-tone-walkthrough.md [EXAMPLE].
references/{playbook, format-conventions, anti-patterns, vn-tone-corpus, translation-artifacts}.mdreferences/procedures/{pre-dispatch, dispatch-mechanics, absolute-prohibitions}.mdreferences/_shared/{before-starting-check, mode-resolver, pre-dispatch-protocol, manifest-spec}.mdagents/{diagnostic, polisher, critic}-agent.md (critic holds 36-point rubric)npx claudepluginhub hungv47/meta-skills --plugin forsvnUse this skill for any task involving English-Thai or Thai-English translation, localization, or rewording. Trigger whenever the user asks to: translate text between Thai and English, localize content for a Thai audience, render English idioms in Thai (or vice versa), pick the right pronouns and register, choose between keeping a term in English vs ทับศัพท์, or fix awkward translations. Also trigger for requests like "แปลเป็นไทย", "แปลอังกฤษเป็นไทย", "ช่วยแปลหน่อย", "localize Thai", "translate this", "Thai version", or any variation. If the task involves moving meaning between Thai and English at any level — word, sentence, document — use this skill.
Translates content between Chinese and English with three modes (quick, normal, refined) and custom glossary support. Handles URLs, files, and direct text.
Rewrites AI-generated Korean text (ChatGPT, Claude, Gemini) to sound human-written. Detects 40+ AI-typical patterns across 10 categories (translationese, passive overuse, uniform rhythm, excessive emoji/bullets) and rewrites style only, preserving content.