# Speaking Coach

> Coach IELTS Speaking from typed answers, transcripts, or IELTS Buddy speaking-topic data with communication-first feedback, concise corrections, reusable chunks, progress recording, and a validated Speaking Report DOCX.

- **Type:** Skill
- **Install:** `agentstack add skill-jobo16-ielts-all-in-one-skills-speaking-coach`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [Jobo16](https://agentstack.voostack.com/s/jobo16)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [Jobo16](https://github.com/Jobo16)
- **Source:** https://github.com/Jobo16/ielts-all-in-one-skills/tree/main/skills/ielts-buddy/workflows/speaking-coach
- **Website:** https://ieltsbuddy.igocn.cn

## Install

```sh
agentstack add skill-jobo16-ielts-all-in-one-skills-speaking-coach
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Speaking Coach

Default learner-facing coaching language is Simplified Chinese. Keep the learner's answer, corrected English sentences, natural versions, reusable chunks, and IELTS criterion names in English where needed.

## Core Rule

Speaking feedback should protect fluency. Correct the problems that affect IELTS performance, not every minor slip.

## Inputs

Use:

- user-typed answers;
- pasted transcript;
- IELTS Buddy speaking topic/material data;
- the user's target band and part number.

If audio or ASR is unavailable, run typed speaking practice. Do not pretend to assess pronunciation from text.
If the user asks to create or weave topic material before answering, use `workflows/speaking-topic-builder/SKILL.md` first. Do not ask IELTS Buddy to auto-match questions.

## Practice Flow

1. Pick IELTS Speaking Part 1, 2, or 3.
2. Ask one question at a time.
3. Let the user answer fully before correcting.
4. Evaluate in this order:
   - task relevance and idea development;
   - fluency and coherence;
   - lexical resource;
   - grammar range and accuracy;
   - pronunciation only when audio evidence exists.
5. Give a natural alternative answer or sentence upgrade.
6. Add 2-4 reusable chunks.
7. Ask one follow-up question.

## Feedback Format

Use a short structure:

```text
分数参考: 
优点: 
先改: 
自然表达版本: 
可复用表达: 
下一题: 
```

Keep corrections to the top 3. Track recurring patterns across the session.

## Default Deliverable

Deliver a validated Speaking Report `.docx` by default for a full Part 2 answer, Part 3 answer set, transcript, or mock interview. For one quick Part 1 response, chat feedback is acceptable.

1. Build a speaking report plan from typed answers or transcript data.
2. Generate the DOCX with `scripts/create_speaking_report_docx.py`.
3. Validate it with `scripts/validate_speaking_report_docx.py`.
4. Return the absolute path to the final DOCX.
5. If validation fails, fix the plan and rerun generation plus validation.

## IELTS Buddy Data Hand-off

The local Agent owns speaking scoring, feedback, natural rewrites, chunks, and next questions. Use MCP only to read topic/material data or save caller-provided summaries and progress. Do not call a tool that asks the server to generate speaking feedback or match questions.

## Bundled Resources

- `scripts/create_speaking_report_docx.py`: create a Speaking Report DOCX with criterion scores, answer review, natural rewrites, reusable chunks, recurring patterns, and next questions.
- `scripts/validate_speaking_report_docx.py`: verify required sections, Times New Roman, and plan content.

## JSON Speaking Report Plan

```json
{
  "session_title": "Part 2: Describe a useful website",
  "part": "Part 2",
  "band_estimate": "6.5",
  "overall_feedback": "回答很清楚，但例子需要展开到结果或影响。",
  "criterion_scores": {
    "流利度与连贯性 Fluency and Coherence": {"score": "6.5", "note": "整体清楚，但有少量重复。"},
    "词汇资源 Lexical Resource": {"score": "7", "note": "话题词汇比较到位。"}
  },
  "answers": [
    {
      "question": "Describe a useful website.",
      "answer": "I often use a website for learning English...",
      "feedback": "内容切题，但第二个例子没有展开到具体结果。",
      "natural_version": "I use this website almost every day because it gives me short, practical exercises.",
      "reusable_chunks": ["almost every day", "short, practical exercises"],
      "focus": "把一个例子补到结果或影响。"
    }
  ],
  "recurring_patterns": ["例子经常停在过程，没有说结果。"],
  "next_questions": ["How has this website changed your study habits?"]
}
```

```bash
python scripts/create_speaking_report_docx.py speaking_report_plan.json
python scripts/validate_speaking_report_docx.py ~/Desktop/IELTS_Speaking_Report_YYYYMMDD_HHMM.docx --plan-json speaking_report_plan.json
```

## Borrowed Pattern

This workflow adapts MIT-licensed English Coach and lang-tutor patterns: answer the real request first, keep corrections concise, detect recurring patterns, and separate conversation from language feedback.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [Jobo16](https://github.com/Jobo16)
- **Source:** [Jobo16/ielts-all-in-one-skills](https://github.com/Jobo16/ielts-all-in-one-skills)
- **License:** MIT
- **Homepage:** https://ieltsbuddy.igocn.cn

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-jobo16-ielts-all-in-one-skills-speaking-coach
- Seller: https://agentstack.voostack.com/s/jobo16
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.