Emotion skill

Emotion

Use this skill to turn authorized WeChat chat exports into self-understanding, emotional pattern insight, relationship insight, practical conversation guidance, and durable personal knowledge. Work only with chats the user owns or is explicitly allowed to analyze.

Safety Boundary

• Do not help access, decrypt, bypass, or extract another person's account or device data.
• Do not print API keys, database keys, private identifiers, phone numbers, addresses, payment details, or full raw intimate transcripts in the final answer.
• Prefer evidence snippets over full-message dumps.
• Mark analysis as interpretation, not diagnosis.
• For self-harm, abuse, stalking, coercion, violence, or severe mental-health risk, stop tactical chat advice and recommend immediate professional/local help.
• Do not produce manipulative PUA scripts. Extract conversational structures and offer respectful, low-pressure wording.

Workflow

1. Clarify scope

   • Identify target relationship, date range, platform/source format, and desired output.
   • If the user says "latest" or gives relative dates, convert to concrete dates.
   • Ask only if the input path, relationship target, or date range is impossible to infer.

2. Prepare data

   • Read references/data-preparation.md when the user needs help exporting/downloading chats or selecting records.
   • If files are already exported, inspect format and normalize with scripts/normalize_chat_export.py when useful.
   • Preserve original files. Write normalized outputs to a new outputs/ or task-specific folder.

3. Select records

   • Prefer narrow evidence: one relationship, one date range, or one event window.
   • Exclude forwarded [聊天记录] blocks unless the user explicitly wants nested-forwarded content analyzed.
   • For group chats, filter to relevant members when possible.

4. Build an evidence map

   • Count messages by sender/day/hour.
   • Identify event turns: conflicts, repairs, ignored boundaries, affection, invitations, rejections, topic shifts, late-night loops.
   • Capture short paraphrased evidence snippets with timestamps.
   • Keep a distinction between "observed behavior" and "inference".

5. Analyze relationship structure

   • Read references/analysis-framework.md.
   • Diagnose patterns across self-presentation, relationship meaning, exchange, power, boundaries, stage, and narrative.
   • Compare current window with prior windows only when data exists.

6. Produce advice and phrase references

   • Read references/output-templates.md for report formats.
   • Read references/phrase-frameworks.md for chat advice and talking structures.
   • Give direct, usable wording, but keep it natural and editable.
   • Separate "safe to use now", "use only when mood is light", and "avoid / risky".

7. Save durable output

   • Recommend saving useful outputs in an Obsidian or Markdown knowledge base, especially relationship reports, weekly reviews, boundary notes, and phrase libraries.
   • When working in an Obsidian vault, create a Markdown note with frontmatter, tags, source paths, date range, message counts, confidence level, and next-action boundary.
   • If no vault is available, provide a clean Markdown note the user can paste into Obsidian later.
   • Do not embed raw full transcripts unless the user explicitly requests archival notes and privacy is appropriate.

Output Defaults

For relationship analysis, produce:

• time range and data scope
• key changes
• self-pattern profile: what kind of person the user appears to be in this relationship
• relationship meaning: what this relationship seems to provide or cost the user
• emotional trigger map
• boundary map
• conversation pattern diagnosis
• what to do now
• reply drafts grouped by pressure level
• phrases to avoid
• Obsidian-ready Markdown summary with suggested tags and next review date

For chat coaching, produce:

• what happened in the specific exchange
• what the other person may be reacting to
• where the user is over-explaining, chasing, or self-abandoning
• 3-8 natural reply options
• one recommended reply
• a reusable phrase-library entry when the wording is broadly useful

When the user asks for usage guidance, recommend:

• start with one person and one narrow date range
• save conclusions, not full private transcripts
• separate "how to reply now" from "why this relationship affects me"
• run weekly reviews for active relationships
• maintain a phrase library grouped by pressure level and scenario
• compare patterns across relationships only after each relationship has its own evidence base

Resource Guide

• references/data-preparation.md: authorized export/download, format expectations, selection workflow.
• references/analysis-framework.md: relationship and conversation analysis checklist.
• references/phrase-frameworks.md: atmosphere, push-pull, boundaries, repair, and wording rules.
• references/output-templates.md: Markdown report templates and short-answer templates.
• scripts/normalize_chat_export.py: normalize authorized JSON/JSONL/CSV/TXT chat exports to JSONL.