skillZs
★ LIVE SKILL TAGS ★
>>> LIVE SKILLS INDEX <<<
* OPEN SOURCE *
NO LOGIN, NO TRACKING
※ REAL INSTALL DATA ※
← back to all skills
chakrit/kien-thai94 installs

kode-thai

Run an iterative audit-and-fix loop on Thai prose using the kien-thai skill — repeats audit/fix passes until a full pass produces zero new edits. TRIGGER when user invokes `/kode-thai`, asks for "audit loop" / "วน audit" / "ตรวจวนๆ" / "ขัดภาษาไทยให้สุด" on Thai writing, says variants of "แก้ไปเรื่อยๆ จนกว่าจะไม่เจอที่ผิด", or explicitly requests repeated review passes on Thai prose. DO NOT TRIGGER for single-pass rewrites or one-off Thai edits (use kien-thai directly), or for non-Thai content.

How do I install this agent skill?

npx skills add https://github.com/chakrit/kien-thai --skill kode-thai
view source ↗

Is this agent skill safe to install?

  • Gen Agent Trust Hubpass

    The skill defines a structured, iterative workflow for auditing and refining Thai prose using the kien-thai linguistic rule set. It focuses on ensuring high-quality language output through repeated passes until convergence and contains no malicious code or suspicious patterns.

  • Socketpass

    No alerts

  • Snykpass

    Risk: LOW · No issues

What does this agent skill do?

kode-thai

โคตรไทย — invoke kien-thai in a loop until the prose stops changing.

Why this exists

Single-pass review misses issues that only surface after earlier fixes shift sentence shape. A connective collapses, the next sentence's framing changes, a new awkward seam appears. Loop until clean.

Protocol

  1. Load kien-thai in full — SKILL.md plus all eight references (ai-tells.md, craft.md, grammar.md, style-rules.md, register.md, examples.md, exemplars.md, forbidden-phrases.md). Don't skip references. Both audit and fix passes need depth — this is a deep language-analysis job, not mechanical scanning. (The pytest harness injects a leaner audit-mode bundle instead — register-scoped, draft-workflow sections stripped; see the project CLAUDE.md "Two-tier injection". "In full" governs interactive runs.)

  2. Read the target file end-to-end before editing anything. Skim-and-fix produces shallow passes.

  3. Audit pass. Deep-read end-to-end first, then list every issue. Cite each issue with the rule's slug (e.g. f4/targhak-closure, wrong-classifier, f6/ko-resumptive) and quote the offending text inline. As a pre-check, scan forbidden-phrases.md against the prose — un-backticked occurrences only (use/mention exemption). If everything passes, output a single line CLEAN — no prose, no commentary.

  4. Fix pass. Apply the listed edits.

  5. Re-read the edited file end-to-end. Diff-level review misses paragraph-flow problems that only show on full re-read.

  6. Repeat steps 3–5 until one audit pass produces zero new issues.

  7. Stop only on a clean pass — not on "good enough", not on "running out of obvious things". If you think it's done, run one more audit pass to confirm.

Stop condition

The loop ends when an audit pass surfaces zero edits. A pass that finds one issue and fixes it doesn't end the loop — the next pass might surface a new issue caused by that fix. Only a fully clean pass terminates.

Token cost

Each pass reloads kien-thai (~tens of thousands of tokens of skill + references) plus the full target file. For long prose this compounds fast. Warn the user before starting on files over ~1000 words so they can decide whether to scope the loop to a section.

Best input: a Thai-native model draft

This loop audits any Thai prose, but its highest-value use is over a draft from a Thai-native model (Typhoon and peers), not Claude/Codex output. See kien-thai's "Best output" section and skills/kien-thai/scripts/: run thai-route.sh to produce the draft, then run this loop on it. The route script handles drafting and fallback; this loop is the audit half — the part that is language judgment, not a scriptable transform.

Relationship to kien-thai

kien-thai is the rule set. kode-thai is the loop. kode-thai doesn't add new rules — it enforces that kien-thai's rules get applied to convergence. If the audit surfaces a pattern no kien-thai rule covers, follow the iteration discipline in the project CLAUDE.md ("Iteration discipline — READ FIRST"): trace the gap before adding a rule, don't synthesize on vibes.

Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.

<a href="https://skillzs.dev/skills/chakrit/kien-thai/kode-thai">View kode-thai on skillZs</a>