import-dialog

Import Dialog — Update Advisor from .md Files

Import exported markdown dialog files and use them to create or update an advisor profile through the conversation-dump and soul-extraction pipeline.

Phase 1 — Specify Inputs

Ask the user for:

1. File path(s): One or more .md dialog files to import. Accept glob patterns (e.g., docs/*.md).
2. Target advisor: An existing advisor slug (from advisors/index.md) or a new advisor name. If new, also gather background info (field, themes, skills) for the profile header.

Phase 2 — Parse and Store

Run the bundled parser to convert .md files into standard JSON turns:

python3 <skill-base-dir>/../conversation-dump/parse_md_dialog.py parse <file.md>

For multiple files:

python3 <skill-base-dir>/../conversation-dump/parse_md_dialog.py batch <directory> --outdir docs/dialog/md-import/raw/

Save JSON outputs to docs/dialog/md-import/raw/. Verify each file parsed correctly (non-zero turns).

Phase 3 — Classify and Analyze

Follow conversation-dump Phases 2–3:

1. Classify each session into a topic using the standard taxonomy: skill-design, debugging, documentation, refactoring, feature-implementation, paper-review, research-brainstorming, data-analysis, system-design, testing, code-review, literature-survey, writing, automated, other.

2. Move classified sessions to docs/dialog/md-import/<topic>/.

3. Present topic counts to the user. Ask which topics to analyze in depth.

4. Deep analysis: For each selected topic, tag every user message across the six dimensions (bloom, depth, probe, presup, discourse, mechanism). Save enriched JSON to docs/dialog/md-import/<topic>/.

Phase 4 — Soul Extraction

For each selected topic, follow soul-extraction Phases 2–4:

1. Extract patterns: Identify trigger→reaction pairs. Cluster similar turns (3-of-4 dimension match). Record patterns with frequency and examples.

2. Detect logic jumps: Find user messages that are not direct responses to the assistant's prior turn. Curate the 5–12 most valuable. Present each candidate to the user for confirmation with causality chain options.

3. Output: Write thinking-pattern.md and master-thinking.md to docs/dialog/md-import/<topic>/.

Phase 5 — Update Advisor Profile

Read the target advisor's existing advisors/<slug>/profile.md.

If advisor exists:

• Preserve the Background section (unless user provides updates).
• For each topic with sufficient data (2+ patterns), generate the five thinking-style subsections: Cognitive Style, Attention Patterns, Reasoning Strengths, Conversation Dynamics, Potential Blind Spots.
• Add new topic sections or replace existing ones that were re-analyzed.
• Keep existing topic sections that were not re-analyzed.

If advisor is new:

• Create advisors/<slug>/profile.md with Background + topic sections.
• Create advisors/<slug>/ directory.

Update advisors/index.md — add or update the row for this advisor.

Present the updated profile to the user for review:

Your advisor profile has been updated at advisors/<slug>/profile.md. The new analysis added/updated the following topic sections: [list]. Please review — raw dialog data stays in docs/dialog/md-import/ and is not included in the profile.

Supported .md Formats

The parser (parse_md_dialog.py) auto-detects these role marker patterns:

Format	Human marker	Assistant marker
Claude.ai export	## **Human**	## **Claude**
Bold variant	## **User**	## **Assistant**
Plain heading	## Human	## Claude or ## Assistant
Colon format	**Human:**	**Claude:** or **Assistant:**

Messages are separated by --- lines (ignored during parsing). Nested markdown headings within assistant responses are preserved as content.