nemawashi-analyze

Profile Analyzer

Analyze a person's collected profile data using psychological frameworks to identify behavioral patterns, classify psychological tendencies, and generate evidence-based communication strategies.

The work is decomposed into independent per-framework agents (one per framework, dispatched in parallel) plus a synthesis pass. Each framework owns its own file under frameworks/<slug>.md, and the top-level profile.md is a slim index that holds Core Pattern + summary table.

When to Use

• User says "analyze X" or "what kind of person is X?"
• After nemawashi-collect finishes (suggest this automatically)
• User wants to update analysis with new data

Prerequisites

A profile must exist at PROFILE_DIR/<person-name>/. If profile.md is missing or there is no facts.jsonl/facts.md, tell the user to run nemawashi-collect first.

Process

Step 1: Read shared inputs

Read these files once, in the main session (shared input for every per-framework agent):

• PROFILE_DIR/<person-name>/profile.md
• PROFILE_DIR/<person-name>/facts.jsonl (newer profiles)
• PROFILE_DIR/<person-name>/facts.md (legacy profiles)
• PROFILE_DIR/<person-name>/relationship.md (if exists)

If both facts.jsonl and facts.md exist, both are read and merged (sort by date descending) — the dual-read transition.

Also enumerate the framework definition files: every *.md under frameworks/ (relative to this skill).

Step 2: Dispatch one framework-analyzer agent per framework

For each framework definition file, dispatch the framework-analyzer agent (defined at agents/framework-analyzer.md) in parallel. Each agent runs in an isolated context, reads its framework definition + the profile inputs, and writes one PROFILE_DIR/<name>/frameworks/<slug>.md file. The agent's full contract — input parameters, apply steps, output schema, return shape — lives in agents/framework-analyzer.md; do not duplicate it here.

Dispatch prompt per agent

framework_slug:              <slug>
framework_definition_path:   <absolute path to skills/nemawashi-analyze/frameworks/<slug>.md>
profile_dir:                 PROFILE_DIR/<name>/
output_path:                 PROFILE_DIR/<name>/frameworks/<slug>.md
assertion_sh:                <absolute path to skills/nemawashi-analyze/assertion.sh>
today:                       YYYY-MM-DD

Issue all 6 dispatches in a single message (parallel). Each agent returns ONE line: <slug>: <classification text> (<Confidence>). The orchestrator collects the lines; full content lives in each frameworks/<slug>.md file.

Step 3: Wait for all agents, then synthesize

After every dispatched agent returns, the orchestrator runs the synthesis pass in the main session:

1. Read every PROFILE_DIR/<name>/frameworks/*.md just produced.
2. Write profile.md per the format in OUTPUT-FORMAT.md:
   • Preserve manual / non-analysis sections (Basic Info, Communication Patterns, Active Channels, Work Patterns) if they exist.
   • Replace the Core Pattern section with a fresh 1-3 sentence synthesis that unifies the framework files into a single behavioral explanation.
   • Replace the Framework Summary table — one row per framework, pulled from each file's frontmatter (classification, confidence) plus a relative link to the file.
   • Update the trailing <!-- analyzed: YYYY-MM-DD, facts_count: N --> comment.
3. Write contradictions.md (cross-framework inconsistencies — verbal contradictions, say-do gaps, audience-dependent behavior, temporal reversals). Cite 2+ facts per contradiction. If none are detected, write the empty-state template per OUTPUT-FORMAT.md.
4. Update relationship.md "Approach Strategy" section if it exists.

Step 4: Report

Print to the user:

• The Core Pattern.
• A summary table (framework → classification → confidence → file path).
• Top 3 rules across frameworks (pick the highest-confidence + most actionable ones).
• Contradictions detected (count + most significant).
• Data Gaps and what to collect next.

Output Format

See OUTPUT-FORMAT.md (relative to this skill) for full schemas:

• The new slim profile.md (Core Pattern + Framework Summary + preserved manual sections).
• The per-framework file at frameworks/<slug>.md.
• contradictions.md (unchanged).

Re-analysis Rules

• Check <!-- analyzed: YYYY-MM-DD, facts_count: N --> in profile.md.
• If facts.jsonl / facts.md has new entries since last analysis, re-dispatch the per-framework agents.
• Re-analysis replaces each frameworks/<slug>.md atomically (write to temp, mv on success).
• Lines tagged <!-- manual --> are user additions — preserve them across re-analysis.
• The Core Pattern and Framework Summary are always regenerated (they depend on all frameworks).

Key Principles

• One framework per file. Each frameworks/<slug>.md is independently analyzable, replaceable, and consumable. Cross-profile queries grep on a single path.
• Parallel dispatch. Wall-clock cost is bounded by the slowest framework + synthesis, not the sum.
• Evidence-based. Every classification and rule cites specific signals from facts. No speculation without evidence.
• Non-judgmental framing. Describe behaviors and patterns, not character.
• Actionable. Every Confirmed classification produces at least one situation-indexed DO/DON'T rule. If it can't, downgrade to Hypothesis or Data Gap.
• Hypothesis transparency. Low-data classifications are kept but clearly marked.
• Framework reference consistency. Always consult the definition under frameworks/ — never rely on general knowledge alone.
• Append-only assertion log. Each frameworks/<slug>.md is a derived snapshot regenerated from frameworks/<slug>.jsonl on every analyze pass. Prior assertions are preserved in the .jsonl (append-only); time-travel queries via nemawashi-show --as-of YYYY-MM-DD. See docs/specs/2026-05-19-41-append-only-temporal-model.md.

Multi-framework rules

Some rules naturally cite two frameworks (e.g. [defense: info-withholding + TKI: competing]). Place the rule under its primary framework's file and tag the secondary framework inline in the rule's bracket annotation. If the rule depends roughly equally on two, pick the one whose Signal Tags it references first. This keeps each file self-contained without duplicating rules.

---

Framework File Contract

See FRAMEWORK-CONTRACT.md (relative to this skill) for the framework definition template and required fields.