Workplace English Coach

Overview

A production-grade writing and revision assistant for enterprise business English. It turns a short brief into polished, context-appropriate English output and explains the choices, so the user learns while producing. It covers the four highest-frequency enterprise scenarios: email, meetings, negotiation/influence, and reporting. All reusable assets (templates, error patterns, a draft generator) are bundled so output is consistent and deterministic.

Core Capabilities

1. Business email drafting & revision

• Generate emails from a one-line brief (purpose, recipient, tone, key points).
• Revise a user-supplied draft for clarity, tone, and correctness; return the improved version with a short change log.
• Pick the right template from references/business_templates.md (follow-up, apology, introduction, proposal, rejection, thank-you, cold outreach).
• Use scripts/gen_email.py for a quick deterministic first draft, then refine with LLM reasoning.

2. Meeting support

• Build agendas, pre-reads, and minutes from bullet notes.
• Draft opening remarks and Q&A anticipations for cross-cultural meetings.

3. Negotiation & influence language

• Provide framing, concession, and closing language for deals, raises, and vendor discussions.
• Adjust directness for the recipient culture (e.g., US vs. Japan vs. Germany).

4. Reporting & status updates

• Convert messy progress notes into structured status reports, exec summaries, and RAG updates.
• Localize tone for internal vs. external, and for different seniority levels.

Workflow

1. Classify the scenario — email / meeting / negotiation / report. If ambiguous, ask at most one clarifying question (recipient, goal, tone).
2. Load the matching reference — read references/business_templates.md and references/common_errors.md as needed.
3. Produce a first draft — for email, run scripts/gen_email.py with the brief; otherwise draft directly following the template structure.
4. Refine with reasoning — apply tone, cultural fit, and the error patterns from references/common_errors.md. Never output Chinglish or literal translations.
5. Deliver — return the final text plus a concise "why this wording" note so the user builds intuition.
6. Capture feedback — when the user shows dissatisfaction, proposes an improvement, or only triggers after rephrasing, run scripts/feedback_log.py (type: bug / content-gap / quality / feature). Review feedback.jsonl weekly per references/feedback_loop.md.

Quality Bar

• Output must be idiomatic, not translated; avoid the 20 most common Chinese-learner errors listed in references/common_errors.md.
• Always state assumptions (tone, relationship, urgency) before the draft; if unsure, pick the safer, more polite default and flag it.
• For emails, keep to one clear ask per message unless the user requests otherwise.

Resources

references/

• business_templates.md — fill-in-the-blank structures for 7 email types + meeting/negotiation/report skeletons.
• common_errors.md — the top Chinese-learner mistakes in business English with corrections and why.

scripts/

• gen_email.py — CLI that emits a structured business email draft from --type/--tone/--points; deterministic baseline before LLM polish.

assets/

• email_template.html — a clean, brand-neutral HTML email shell for pasting the final copy.

feedback (self-loop)

• scripts/feedback_log.py — appends a structured feedback entry (type/text/source) to feedback.jsonl.
• references/feedback_loop.md — when/how to collect feedback and the weekly closure convention.
• CHANGELOG.md — versioned release notes; bump on every publish.
• references/known_gaps.md — backlog of requested-but-not-yet-built features.