course-processing

name	course-processing
description	Используй когда нужно обработать онлайн-курс, серию уроков/workshop, multi-file transcript corpus или источник, который должен дать claim-backed Source Note, Knowledge Notes, methods, runtime references/evals или audit packet.
verified_at	"2026-05-30T00:00:00.000Z"
covers	["sources","synthesis","process"]
drift_covers	[".agents/skills/course-processing/**","agent-harness/source-processing/course-factory/**","source-workbench/prompts/course-factory/**","source-workbench/templates/course-factory/**"]

Course Processing

Ты ведёшь Course Factory workflow для больших учебных источников: онлайн-курсов, серий уроков, workshop series, multi-file transcript corpora and skill-impacting materials.

Главная задача skill — не пересказать курс, а провести источник через проверяемый путь:

raw/clean source
-> inventory
-> course map
-> claim bank
-> lesson synthesis
-> Source Note
-> Knowledge Notes
-> method backlog
-> runtime references
-> evals
-> audit

Required Read

Перед course-processing задачей прочитай релевантный canon:

1. agent-harness/source-processing/course-factory/README.md
2. agent-harness/source-processing/course-factory/processing-profiles.md
3. agent-harness/source-processing/course-factory/pipeline.md
4. agent-harness/source-processing/course-factory/artifact-contracts.md
5. agent-harness/source-processing/course-factory/claim-bank-contract.md
6. agent-harness/source-processing/course-factory/skill-method-contract.md
7. agent-harness/source-processing/course-factory/audit-contract.md

Если задача узкая, дочитывай только нужные references этого skill.

When To Use

Используй skill, когда пользователь просит:

• обработать курс, серию уроков, workshop, лекционный цикл или raw transcript folder;
• превратить course material в Source Note, Knowledge Notes, method backlog or runtime references;
• выбрать processing profile: lite, standard, skill-impact, production-deep;
• проверить, можно ли обновлять runtime skill из курса;
• собрать или проверить claim bank, course map, Source Note replacement, method evals or audit packet.

Do Not Use

Не используй skill для:

• обычного SMM/copywriting;
• одиночной статьи или короткой заметки без course/method impact;
• VK Ads campaign execution без обработки course source;
• prompt optimization;
• чистого summary без reusable Source Note;
• live parent-vault routing без явного scope.

Operating Rules

• Raw access is mandatory. If source files are missing or unreadable, stop and ask for raw location instead of simulating processing.
• Use at most five intake questions: raw location, full/partial course, expected units, skill update requirement, destination/project.
• Never route raw transcript -> SKILL.md or batch report -> runtime method.
• Accepted claims must be specific, source-derived, raw-anchored and destination-aware.
• Source Note must be reusable enough to replace raw source for later work.
• Runtime methods require claim IDs, Source Note section, trigger, inputs, algorithm, output format, QA criteria, failure modes and eval.
• For production-deep, require external audit packet; do not copy the full raw corpus into the packet.

Reference Map

• references/course-intake.md — profile choice, intake questions and source access stop rules.
• references/course-map-contract.md — case directory, inventory, course map and source/unit IDs.
• references/claim-bank-contract.md — accepted claim fields, density and generic-claim blockers.
• references/source-note-contract.md — Source Note build and replacement check.
• references/knowledge-note-contract.md — Knowledge Note selection and routing.
• references/skill-method-contract.md — method backlog, runtime references and eval gate.
• references/audit-contract.md — semantic audit, external audit packet and final verdict.

Output Contract

For planning or review tasks, return:

1. profile and reason;
2. required artifacts;
3. blockers before processing;
4. next smallest action;
5. verification path.

For implementation tasks, follow the repo task contract, use existing templates/mirrors, and record proof. If automated course_factory_* scripts are not present yet, say that those checks are not available instead of marking them passed.