discussion-diagnosis-aggregator

Discussion Diagnosis Aggregator

Role

This skill does not run the 6 single-skill diagnoses itself. It assumes their outputs (structured text or JSON) are available, and performs:

1. Merge — combine 6 skill outputs into a unified report
2. Deduplicate — collapse multi-skill issues that point to the same sentence (e.g., a verb-tense problem flagged by both Grammar and Vocabulary should appear once)
3. Severity-tag — classify each issue as critical / major / minor
4. Prioritise — produce top-3 fix list (severity × frequency × position in text)
5. Anchor with examples — for each high-severity issue, point to a positive example in references/examples/
6. Score — compute overall score (weighted) + per-dimension score + narrative-coherence meta-score
7. Render — output a single Markdown report directly in the chat as the assistant's reply; see Output Mode below for the chat-first, language-adaptive delivery rules

Output Mode (chat-first, language-adaptive)

This skill's final report must be delivered inside the assistant's chat reply, not as a file artifact. Concrete rules:

• Channel: assistant message body (Markdown rendered in the conversation)
• No write / edit for the report itself — do NOT create *_diagnosis.md / *_discussion_report.md style files for the final report
• Language policy — match the conversation language at the moment of diagnosis:
  • User wrote in Chinese (中文对话) → report in Chinese (中文报告)
  • User wrote in English → report in English
  • Mixed / ambiguous → follow the dominant language in the user's most recent turn
  • Section headers, severity tags, bullet labels are translated consistently with the chosen report language
• Internal scaffolding is fine on disk — intermediate artifacts (PDF text dumps via pymupdf/pdfplumber, _extract_*.py scripts, scratch text) may still be written to disk for processing; only the final report is chat-bound
• Opt-in file fallback — if the user explicitly asks "save the report to a file" / "生成报告文件" / "export the diagnosis", fall back to writing a .md file AND still match the conversation language
• Self-containment — the report inside the chat must be complete and readable without opening any external file (no "see attached" stubs)

Required inputs (from the 6 single skills)

Each single skill is expected to emit a structured block including:

• dimension (one of: structure / cohesion / grammar / vocabulary / logic / conventions)
• score (0–20 or 0–100, depending on the skill's rubric)
• issues (list of {sentence_ref, severity, description, fix_suggestion, example_ref, cross_dimension_refs?})

Aggregator output structure

# Discussion Diagnostic Report

## Overall score: X / 100
(Weighted: structure 20% / cohesion 20% / grammar 10% / vocabulary 10% / logic 20% / conventions 20%)

## Per-dimension scores
- Structure: X / 20
- Cohesion (incl. narrative thread): X / 20
- Grammar: X / 20
- Vocabulary: X / 20
- Logic: X / 20
- Conventions: X / 20

## Narrative-coherence meta-score
(From Cohesion Level 2 — global thread + take-home persistence)
Score: X / 10 — [strong / adequate / weak]

## Severity-tagged issue list
### Critical (must fix)
1. [sent X] — [brief description] — flagged by: [dimensions]
2. ...

### Major (should fix)
1. ...

### Minor (nice to fix)
1. ...

## Top-3 priority fixes
1. **[Critical] [Sentence X]** — issue: ...; fix: ...; positive example: `references/examples/good_XX.md`
2. ...
3. ...

## Cross-dimension deduplication notes
- Issue Y was flagged by both Grammar and Vocabulary; consolidated here.
- Issue Z was flagged by both Conventions and Cohesion; consolidated here.

## Strengths (also surfaced)
- ...

Severity rubric

References (to be filled in Phase 3)

• references/output-template.md — full Markdown template
• references/severity-rubric.md — full severity criteria
• references/cross-dimension-map.md — which issues are expected to be flagged by multiple skills (and how to consolidate)
• references/examples/ — annotated good/bad Discussion examples