zai-distillation

name	zai-distillation
description	Distill books, documents, folders, or URLs into compact, practical Agent Skills. Use when source material should become either one reusable skill package or a split set of focused skills, each with a concise SKILL.md plus detailed references and examples.
argument-hint	<path|url> [path|url...] [--name <skill-name>] [--path <directory>] [--update] [--redact-source-map] [--split|--split-by <strategy>]
allowed-tools	Read Write Edit Glob Grep Bash(mkdir *) Bash(ls *) Bash(find *) Bash(wc *) Bash(python3 --version) Bash(python --version) Bash(py -3 --version) Bash(py --version) Bash(python3 *material-prep.py*) Bash(python *material-prep.py*) Bash(py -3 *material-prep.py*) Bash(py *material-prep.py*) WebFetch WebSearch AskUserQuestion
disable-model-invocation	false
metadata	{"author":"ai-factory","version":"1.0","category":"knowledge-management"}

Language and Coding Standards

• Communication: Always talk in Thai when interacting with users.
• Code & Technical Assets: All code, comments, documentation, and technical definitions must be in English.

Distillation

Turn source material into a useful skill. The output is not a summary dump: it is an operational skill that captures the best practices, decision rules, workflows, checks, and examples from the material.

Step 0: Load Config and Skill Context

FIRST: Read .ai-factory/config.yaml if it exists to resolve:

• language.ui for prompts, questions, progress updates, and final summaries
• language.artifacts for generated skill package content (SKILL.md, references/, examples/)
• language.technical_terms for translated artifacts; default to keep when absent

If config.yaml doesn't exist, use defaults:

• language.ui: en
• language.artifacts: same as language.ui
• language.technical_terms: keep

Read .ai-factory/skill-context/zai-distillation/SKILL.md - MANDATORY if the file exists.

Treat skill-context rules as project-level overrides for this skill. They apply to all generated skill files, references, examples, source-map policy, and final reports.

Inputs

Accept $ARGUMENTS as one or more:

• local files
• local directories
• URLs
• optional --name <skill-name>
• optional --path <directory> to save generated skill package directories under a custom output root instead of {{skills_dir}}
• optional --update to improve an existing skill instead of creating a duplicate
• optional --redact-source-map to skip generated source-map files and sections entirely, so exact source titles, URLs, local paths, repository paths, and link reference definitions are not written to output
• optional --split to create several focused skills from one material set
• optional --split-by <strategy> to choose the split strategy:
  • auto (default): infer skill boundaries from user goals, triggers, workflows, source topics, and use cases
  • goal: split by user goals or jobs-to-be-done, regardless of domain
  • topic: split by major source topics or chapters
  • workflow: split by recurring actions an agent performs
  • audience: split by distinct user roles or implementation contexts

If the target skill name is missing, derive a concise, general, lowercase-hyphenated name from the material topic or user goal, such as clean-code-style, api-design-rules, decision-making, writing-feedback, or meeting-facilitation.

Before any write, validate the final target skill name:

• It must match ^[a-z0-9]+(?:-[a-z0-9]+)*$.
• Reject empty names, overlong names, ., .., dots, path separators (/ or \), absolute paths, Windows drive paths, and hidden names.
• Reject reserved aif-* names unless the user explicitly says they are developing AI Factory itself.
• Resolve the output root: --path <directory> when present, otherwise {{skills_dir}}.
• Treat relative --path values as relative to the current working directory. Create the output root if it does not exist; reject it if it resolves to an existing file.
• Resolve the final destination path and confirm it is inside the resolved output root before creating or updating files.

Default destination for single-skill mode: <output-root>/<skill-name>/, where <output-root> is {{skills_dir}} unless --path is present.

Default destination for split mode: <output-root>/<prefix>-<child-scope>/ for each generated child skill. Every split child name must share one namespace prefix to prevent collisions with existing skills. Use --name as the preferred prefix when present; otherwise derive a concise prefix from the book title or primary material title. If --redact-source-map is present and the exact source title should not be exposed, use --name as the public namespace or derive a neutral topic prefix.

Do not save distilled skills into the package skills/ directory unless the user is explicitly developing AI Factory itself.

ZeaZ Platform & apps/* Monorepo Rules

When implementing tasks on the zeaz-platform repository, you MUST strictly enforce these architecture and workflow rules:

1. Monorepo Architecture (apps/*): The platform is a unified monorepo. ALL applications, microservices, frontends, and AI toolings (e.g., zLinebot, zwallet, zdash) reside inside the apps/ directory. Do not create top-level directories for apps. When refactoring or adding features, always scope your work to the specific apps/<app-name>/ folder.
2. Environment Variables: Avoid scattering .env files. Consolidate environment variables into a central .env.example inside the respective app folder. Canonical Cloudflare variables (e.g. CLOUDFLARE_API_TOKEN, CLOUDFLARE_ZONE_ID) MUST be used instead of legacy CF_ variants.
3. Commit Workflow: NEVER use git commit or git push directly. ALWAYS stage your intended files with git add and commit using make gpg-finalize COMMIT_MSG="..." from the repository root to ensure all GitOps and DevSecOps checks pass.
4. Security: NEVER commit or generate real secrets. Unsafe placeholders like test-secret-value-value-value, test-secret-value-value-value, test-secret-value-value-value are FORBIDDEN.
5. Language: Code, documentation, and technical definitions MUST be in English.

Workflow

1. Prepare sources.

   • For normal text, markdown, JSON, YAML, HTML, or code files, read directly.
   • For large folders or PDFs, use {{skills_dir}}/zai-distillation/scripts/material-prep.py only when a working Python 3 interpreter is available. Detect it with python3 --version, python --version, py -3 --version, then py --version; use the first command that exits successfully and reports Python major version 3.
   • When invoking the helper, expand the selected interpreter to the concrete command shape, such as python3 ...material-prep.py or py -3 ...material-prep.py. Do not run arbitrary Python payloads; the pre-approved tool contract only covers version probes and material-prep.py execution.
   • If Python 3 is not available, do not invoke the helper. Continue with direct Read/Glob/Grep/find/wc sampling for accessible text files, ask the user for a text/markdown export for PDFs or very large sources, and clearly report any reduced coverage.
   • For URLs, fetch the source and any critical linked pages needed to understand the topic.

2. Distill, do not copy.

   • Extract transferable principles, workflows, heuristics, checklists, terminology, and failure modes.
   • Inventory examples from the source, especially code snippets, before deciding the output structure.
   • Group source examples by topic so coverage can be checked later.
   • Preserve only short source excerpts when essential. Prefer paraphrase and cite sources.
   • Convert narrative advice into agent-operable instructions and source examples into original, reusable examples.

3. Choose single-skill or split-skill design.

   • Default to single-skill mode unless --split or --split-by is present.
   • In split mode, resolve one shared namespace prefix before writing the boundary map. Use --name when present; otherwise use a normalized book/material title. Every proposed child name must start with <prefix>-.
   • In split mode, create a skill boundary map before writing: proposed prefixed skill name, user-facing job, trigger description, owned source topics, references/examples needed, and overlap risks.
   • Prefer split mode when the material contains independent goals that should trigger separately, such as reviewing, planning, diagnosing, rewriting, teaching, deciding, testing, facilitating, auditing, or troubleshooting.

Required Supporting Guidance

Read these before generating or updating a distilled skill:

• references/DISTILLATION-PROTOCOL.md
• references/OUTPUT-STRUCTURE.md
• references/LARGE-MATERIALS.md

Use examples/REQUESTS.md for invocation patterns.

Artifact Ownership

• Primary ownership: generated or updated skill packages under <output-root>/<skill-name>/, or multiple direct child skill packages under <output-root>/<prefix>-<child-scope>/ in split mode. <output-root> is {{skills_dir}} unless the user passes --path <directory>.
• Read-only context: .ai-factory/config.yaml, existing AI Factory context artifacts, and existing skill files except the selected target skill in update mode.
• Config policy: config-aware for language.ui, language.artifacts, and language.technical_terms only. Do not write config.yaml.