AI Shifu Learning Report

AI-Shifu Learning Report

Turn observed data from one course into a decision-ready report for teaching managers and teachers. Keep collection, interpretation, and presentation separate so every conclusion can be traced to a defined metric without exposing learner data.

Required References

Read these files completely, in order, for every report:

1. references/data-collection-and-privacy.md
2. references/analysis-guidelines.md
3. references/report-structure.md

Resolve every ## Required References declaration in those files transitively before acting.

Scope Router

Workflow

1. Resolve the request. Identify exactly one course and any requested time range. Default to zh-CN and cumulative-to-date data. Use en-US only when the user explicitly asks for English; schema keys, enum values, commands, and file names stay unchanged.
2. Collect or validate. Follow data-collection-and-privacy.md. Live collection delegates authentication, course resolution, outline resolution, analytics syntax, and platform privacy controls to the current ai-shifu-course-creator; never recreate those mechanisms here. Resolve the current published outline and remove hidden, unpublished, and container nodes before normalizing any lesson-scoped signal.
3. Normalize. Create a schema_version: "1.0" report object. Keep unavailable data as null with an explicit quality explanation instead of guessing or converting it to zero.
4. Analyze. Follow analysis-guidelines.md. Separate observations from interpretations, preserve conflicting signals, and write 3–5 evidence-linked recommendations.
5. Write the data artifact. Save the privacy-safe object as course-learning-report.json. This file is the single source for the rendered report.
6. Validate and render. Follow report-structure.md, validate the JSON, then run the bundled renderer to create course-learning-report.html from that exact JSON.
7. Run the release gate. Confirm that both files describe one course, use the requested language, contain no raw learner text or identifiers, label metric definitions and time scopes, show missing-data states honestly, and contain no external runtime assets.
8. Deliver both files. Summarize the reporting window, major data limitations, and whether follow-up text was sampled. Do not paste private source rows into the handoff.

Non-Negotiable Boundaries

• Use the course creator skill's current CLI for live data. Never read a token, inspect its environment file, compose authentication headers, or call platform HTTP endpoints directly.
• Do not copy or freeze the analytics query language in this skill. The course creator skill owns query syntax, table semantics, codes, and recipes.
• Never place raw follow-up text, answers, phone numbers, emails, names, nicknames, learner labels, or any raw *_bid value in either final artifact.
• Build every lesson-scoped analysis from the current published outline. Include only published, visible teaching leaf lessons, and use that same eligible set for the course entrant denominator, completion numerator, learning path, lesson health, follow-up attribution, and recommendations. If publication or visibility cannot be resolved reliably, mark the affected metrics unavailable instead of falling back to a draft outline.
• Calculate 课程完成率 from one consistent learner cohort: the denominator is distinct learners whose first progress on an eligible lesson falls inside the metric's time scope, and the numerator is the subset of those learners who complete every required lesson or one valid required branch path by the report cutoff. Count each learner once. Do not impose a fixed 30-day or other maturation window, and do not remove late starters to improve the rate. Put the exact cohort dates, cutoff, numerator, denominator, eligible lesson scope, and completion rule in the metric definition and source notes. Do not append 代理 or 近似 to reader-facing Chinese content. Treat 进行中 / In progress as a recorded state, never proof that learners are stuck.
• Keep orders, revenue, payment channels, and AI-Shifu credit consumption out of the teaching report unless the user explicitly requests an operations appendix.
• The JSON is the factual contract and the HTML is its presentation. Do not add claims to HTML that are absent from JSON.
• Present the HTML in the Swiss International Style defined by report-structure.md; preserve its modular grid, typographic hierarchy, flat square geometry, and restrained color system when applying brand overrides.

Completion Checklist

• course-learning-report.json passes the bundled validator for schema version 1.0.
• course-learning-report.html is self-contained, responsive, accessible, printable, generated from the validated JSON, and rendered with the required Swiss International Style system.
• Every metric includes key, label, value, unit, definition, time_scope, data_quality, is_approximate, and source_notes.
• The report contains 3–5 recommendations with cited evidence, confidence, an action, and a validation method.
• Follow-up analysis discloses its recent-sample size and collection status; an opt-out produces an explicit not-collected state, not an empty-data inference.
• Privacy scan finds no raw source text, identity data, internal IDs, or sensitive learner profile values.