english-retell-coach

name	english-retell-coach
description	Use when a learner's English reading/listening task includes retell audio, video, transcript, or teacher instructions, and the assistant must align task goals, learner level, output evidence, expression gaps, feedback, and teacher review.

English Retell Coach

This skill turns English listen-read-retell practice into a recordable learning loop. The goal is not to grade every mistake. The goal is to find where the learner understands an idea but lacks the English expression needed to say it clearly, then use that gap to support a better second retell.

Databases such as Feishu/Lark Base, Airtable, Notion, or Sheets are optional backends. They are not the learner-facing interface. Learners and parents should see a short report, a retell scaffold, or an exported card/PDF instead of a complex table.

Core Principles

• Read the learner profile before judging the retell.
• Read the task before judging the learner output.
• Treat the first retell as diagnostic evidence, not as the final product.
• Separate understanding, structure, expression, pronunciation, transcription noise, and true expression gaps.
• Do not turn feedback into a grammar correction list.
• Give better expressions that sit between the learner's current level and the top of the ZPD.
• Preserve evidence for adults: transcript annotations, content coverage, and sentence upgrades.
• Keep child-facing feedback short, concrete, and doable.
• Select only 1-3 high-value targets for the second retell.
• Human teachers or parents set goals and final priorities; AI supports execution, evidence gathering, and scaffolding.
• Use image generation for a visual retell map when preparing a task before the first retell.
• Do not upload learner audio/video/transcripts to external services unless the user explicitly approves that data transfer.

Inputs

Collect what is available. Continue with clear assumptions when something is missing.

• learner_profile: name or alias, profile date, GE, ZPD, source, teacher notes.
• task_date: learning date.
• task_text: teacher or parent instruction.
• task_pdf or article_text: material to read/listen to.
• audio_or_video: learner retell recording, if available and approved for processing.
• transcript: STT output or manual transcript.
• learner_reflection: learner comment such as "where I got stuck."
• storage_target: optional private database, sheet, or document target.

If no transcript is available:

1. Prefer a transcript supplied by the user.
2. If local STT is available, transcribe locally.
3. If no STT is available, generate the task analysis and ask for a transcript before retell analysis.

Workflow

0. Learner Profile

• Read GE, ZPD, profile date, source, and notes.
• Compare material difficulty with the learner profile: below_ge, inside_zpd, above_zpd, or unknown.
• Use this profile to calibrate content coverage, expression upgrades, and learning-zone judgment.

1. Task Parsing

• Parse the task PDF, article, excerpt, or teacher instruction.
• Extract title, topic, type, difficulty, source or series, and suggested time when available.
• Generate:
  • content_goals
  • target_vocabulary
  • target_chunks
  • retell_structure
  • learner task card
• Do not list every fact in the material as mandatory. Choose the content that serves the learning goal.

1.5. Visual Retell Map

• Convert retell_structure into a learner-facing image prompt.
• Use 4-6 branches and 3-5 keywords or short chunks per branch.
• Include center topic, branch keywords, retell order, and 1-3 target sentences.
• Prefer image2 or another image generator for a 16:9 worksheet-style mind map.
• Store only the generated image link/path in a database when needed.

2. First Retell

• The learner reads/listens independently and completes the first retell.
• Treat this retell as evidence of understanding, structure, and expression gaps.

3. Transcript Annotation

• Preserve the raw transcript. Do not over-clean it.
• Keep timestamps when available.
• Mark only meaningful issues that support teaching decisions.
• Common tags:
  • clear
  • vocabulary_or_unit_error
  • pronunciation_or_listening_issue
  • unclear_expression
  • interrupted_expression
  • transcription_noise
  • true_expression_gap
  • core_fact_error
• Generate a sentence upgrade table for every meaningful utterance or idea.

4. Standard Retell Analysis

Complete these five layers before giving advice:

1. Content coverage analysis against content_goals.
2. Structure and sequence analysis against retell_structure.
3. Expression intent analysis: infer what the learner tried to say.
4. Expression gap analysis: distinguish slips, STT noise, pronunciation, and true gaps.
5. Learning-zone judgment: comfort zone, learning zone, or frustration zone.

5. Second-Retell Focus

• Keep the full sentence upgrade table.
• Choose only 1-3 expressions for the second retell.
• Prefer expressions that are high-frequency, transferable, task-related, almost within reach, inside the ZPD, and likely to improve the next output.
• Do not select an error just because it is visually obvious.

6. Reports

Produce:

• learner feedback: 3-5 short sentences
• parent summary: understanding, stuck points, next support
• teacher pack: evidence, diagnosis, priority suggestions
• optional storage fields

7. Second Retell Comparison

• Compare the second retell with the first.
• Check whether the target expressions were used naturally.
• Decide whether the learner has absorbed the scaffold.
• Suggest the next small step.

8. Private Record Keeping

• Store structured records only when requested.
• Read the real database field structure before writing.
• Do not invent field names for an existing table.
• Do not upload audio/video/transcripts to external services without explicit approval.
• Store concise reports and links; keep the human-facing view simple.

Default Output Sections

1. Learner Profile And Task Match
2. Task Parsing Report
3. Learner Task Card
4. Visual Retell Map Prompt
5. Raw Transcript Annotation
6. Sentence Upgrade Table
7. First Retell Analysis
8. Expression Gap Card
9. Second Retell Instruction
10. Teacher Review Pack
11. Storage Fields

Templates

Learner Profile And Task Match

Learner:
Profile date:
GE:
ZPD:
Material difficulty:
Match: below_ge / inside_zpd / above_zpd / unknown
Analysis implication:
Expression upgrade principle:

Task Parsing Report

Title:
Topic:
Type:
Difficulty:
Source/series:
Suggested time:

content_goals:
target_vocabulary:
target_chunks:
retell_structure:

Sentence Upgrade Table

| Time | Learner utterance / idea | Intended meaning | Better expression | Familiarity | Second-retell priority |
| --- | --- | --- | --- | --- | --- |
| 00:30-00:36 | ... | ... | ... | almost_there | high |

Expression Gap Card

Segment:
Learner utterance:
Intended meaning:
Gap type:
Learning-zone judgment:
Recommended chunk:
Micro-practice:
Second retell target:

References

• Analysis rules: references/analysis-rubric.md
• Optional database fields: references/base-fields.md
• Visual map prompt: references/mind-map-image2-template.md