com-b-diagnostic（组合诊断）

COM-B Diagnostic Skill

Purpose

Run a structured behavioral diagnosis using COM-B, the Behavior Change Wheel, and BCT Taxonomy v1. Given a description of a stuck or problematic behavior, produce a traceable chain: structured behavior definition → dimensional lens analysis across C, O, M → cross-lens assessment → intervention functions → named techniques → actionable recommendations.

When to activate

The user describes a behavior that is stuck, inconsistent, failing to adopt, regressing, or contested — in any domain (product, team, organizational, individual). Activation phrases include:

• "Why aren't people doing X?"
• "How do we get people to start/stop/change X?"
• "Diagnose this behavior"
• "Run a COM-B analysis"
• "What's blocking adoption of X?"

Pipeline

Follow references/flow.md steps 1–5 in order. The full walkthrough with file pointers is in references/guide.md.

Step 1 — Define the behavior

Before any analysis, establish a precise behavior definition using assets/behavior-canvas.md. Extract from the user's description and artifacts, or coach them through it with follow-up questions. The canvas lists every field, coaching prompts for ambiguous cases, and optional pattern vocabulary for describing the behavior's current state.

If critical fields are missing (especially "will do what" and "to what extent"), ask before proceeding. A vague behavior definition produces a vague diagnosis.

The canvas includes a "Prior attempts" field for what has already been tried and why it didn't work. This is critical context -- prior attempts that failed often reveal which COM-B branches are not the real bottleneck.

Once the behavior canvas is complete, offer to save it as a markdown file. The canvas anchors the entire analysis and is worth preserving — especially if the user refines it during the feedback loop.

Step 2 — Research capability, opportunity, and motivation

Apply three diagnostic lenses by reading the lens files directly:

• Capability (C-lens): references/lenses/capability-lenses.md — PC, PHC dimensions.
• Opportunity (O-lens): references/lenses/physical-opportunity-lenses.md and references/lenses/social-opportunity-lenses.md — PO, SO dimensions.
• Motivation (M-lens): references/lenses/motivation-lenses.md — RM, AM dimensions.

For each relevant dimension, note: relevance, position on the scale, and evidence. Skip dimensions that don't apply — most situations activate a subset.

This is a good place to ask follow-up questions. The lens dimensions often surface distinctions the user hasn't considered — e.g. whether a capability gap is about mental models or procedural fluency, whether motivation is about identity or habit, whether opportunity constraints are about tooling or norms. If the initial description doesn't give enough signal to choose between competing dimensions, ask.

Step 3 — Synthesize and assess

The output of this step is a filled-out assets/practitioner-worksheet.md. Work through it section by section:

1. COM-B priority ranking — rank C, O, M by contribution to the gap
2. Per-lens synthesis — for each active branch, fill the table (sub-lens, dimensions, Pattern guidance conclusions, BCTs) and write a narrative synthesis
3. Cross-lens interactions — tensions, reinforcements, prerequisites, highest leverage
4. BCW function ranking — top 5 functions with rationale and top 3 BCTs each
5. Intervention implications — connect BCTs to concrete actions

The filled worksheet is both your internal reasoning artifact and a deliverable the user can request.

Step 4 — Convert to BCW and BCT recommendations

This step runs under the hood. The user does not need to see the taxonomy chain unless they request the in-depth report.

Use the BCW function ranking from the practitioner worksheet to prioritize intervention functions. Read references/com-b-bcw-bct/com-b-to-bcw-intervention-function-mapping.md for the base mapping and references/com-b-bcw-bct/bct-taxonomy.md for named techniques.

Step 5 — Frame recommendations (phased delivery)

Deliver in two phases:

Phase A — always deliver first:

• Summary: Plain-language "here's what's going on and what to do." No taxonomy codes. Write as a trusted advisor would speak.
• Key insights: 3–5 most important findings — tensions, surprises, or highest-leverage points.

Feedback loop — after Phase A, before Phase B:

After delivering the summary and key insights, pause and ask for the user's reaction. Prompt for: whether the diagnosis resonates, whether they've tried any of the implied approaches, and whether there's missing context. If the user provides substantive new information, update the behavior canvas (especially Prior attempts) and re-run from Step 2 with the enriched picture. If the diagnosis lands, proceed to Phase B.

Phase B — user chooses one or more:

• In-depth report: Full diagnostic with dimensional assessments, cross-lens tensions, BCW/BCT reasoning visible.
• Action plan: Concrete phased plan — week-by-week — with specific changes, owners, success signals, tool/AI recommendations.
• Working analysis: The filled-out practitioner worksheet showing COM-B ranking, per-lens synthesis, cross-lens interactions, and BCW function ranking with rationale.

Follow the output skeleton in assets/output-template.md for structure.

Saving outputs: After delivering any Phase B output (in-depth report, action plan, or working analysis), offer to save it as a markdown file. These artifacts represent significant processing and are expensive to reproduce. If the user treats Phase A as a final deliverable (e.g., they don't request Phase B), offer to save the summary and key insights as well.

Guardrails

1. Start with behavior definition (Step 1). Do not skip to analysis without a structured definition. Coach the user if needed.
2. Follow the pipeline order. Later steps depend on earlier outputs. Do not jump to interventions without researching through lenses first.
3. Use the practitioner worksheet. After lens analysis, use assets/practitioner-worksheet.md to synthesize findings, rank BCW functions, and design interventions.
4. The real diagnosis comes from the C, O, and M lenses. Pattern labels in the behavior canvas are orientation, not findings.
5. Multi-actor situations: do not stop at per-role analysis — trace handoffs and where one actor's output becomes another's input; failures often concentrate at interfaces.
6. Deliver Phase A first. Do not produce the full report or action plan until the user asks. Let them pull.
7. Trace everything. In the in-depth report, every intervention function should trace back to a dimensional position, every dimensional position to observed signals. If the chain breaks, the diagnosis is incomplete.
8. Name your sources. When citing BCTs, use BCT.n.m IDs (see references/lenses/lens-map.md). When citing lens dimensions, use the stable PC / PHC / RM / AM / PO / SO IDs.
9. Close the feedback loop. After Phase A, always prompt for the user's reaction before proceeding to Phase B. New information from the user — especially prior failed attempts — should update the canvas and trigger re-analysis, not be patched onto existing conclusions.
10. Offer to save substantial outputs. After delivering the behavior canvas, Phase A summary, or any Phase B deliverable, ask whether the user would like it saved as a markdown file. Use a descriptive filename (e.g. com-b-diagnosis-[short-topic].md) or let the user choose. Do not auto-save — always ask first.