| name | analyze |
| description | Use this skill when the user asks to analyze, investigate, assess, evaluate, compare, or research a codebase, architecture, system, product, market, or research topic in a structured way — including requests that need explicit scope, framework-driven reasoning, and separated facts/inference/recommendation but do not use the word "analyze" — to run a deep analysis workflow producing traceable scope, evidence, and conclusions. Not for quick one-off lookups or implementation work — this is an investigation-and-conclusions workflow. |
| context | fork |
| argument-hint | ["topic","question","or area to analyze"] |
Analyze
Perform structured analysis with explicit scope, evidence, and conclusions.
When to use
- Analyzing, investigating, assessing, or evaluating a codebase, architecture, or system
- Comparing options or alternatives in a structured way
- Researching a product, market, or topic where evidence and conclusions must be traceable
- Any request needing explicit scope, framework-driven reasoning, and separated facts/inference/recommendation
- Not for quick one-off lookups or implementation work — this is an investigation-and-conclusions workflow
0. Gather Context
Read CLAUDE.md (or AGENTS.md) at the project root to understand the project's domain, tech stack, and business context. This shapes framework selection and analysis scope.
1. Classify the Request
Identify:
- Subject
- Core question
- Scope
- Desired depth
- Audience
- Decision the analysis should inform
2. Choose a Framework
Pick one to three frameworks from analytical-frameworks.md based on the question type.
Typical pairings:
- Current-state assessment -> gap analysis, architecture assessment
- Root cause -> 5 Whys, fishbone, causal analysis
- Comparison -> weighted scoring, MCDA
- Strategy -> SWOT, Porter's Five Forces, JTBD
- Technical assessment -> architecture review, technical debt assessment
3. Plan the Investigation
Break the work into MECE dimensions.
For each dimension, define:
- What evidence is needed
- Where it will come from
- What would confirm or contradict the current hypothesis
4. Collect Evidence
Use the minimum set of relevant sources:
- Repository files and docs
- Structured data or metrics
- Targeted commands
- Web research when current external information is required
Track evidence quality and note assumptions.
5. Analyze
Rules:
- Lead with evidence, not opinion
- Consider at least one alternative explanation
- State confidence explicitly
- Separate facts, inference, and recommendation
6. Present the Result
Use output-templates.md to structure the output.
All outputs should include:
- Executive summary
- Context
- Analysis
- Findings
- Recommendations
- Risks and limitations
Integration
- Input from: direct analysis/investigation/comparison/research request;
CLAUDE.md or AGENTS.md for project domain, tech stack, and business context (step 0)
- Drives: framework selection from
analytical-frameworks.md; output structure from output-templates.md
- Output: structured deliverable with executive summary, context, analysis, findings, recommendations, and risks/limitations — suitable for hand-off to a decision-maker or a downstream planning/implementation workflow
- Web research: used only when current external information is required (step 4)
Failure modes
- Opinion before evidence — recommendations stated without grounding violate step 5; lead with evidence, not opinion
- No alternative considered — step 5 requires at least one alternative explanation; a single-hypothesis analysis is incomplete
- Conflated facts/inference/recommendation — keep these separate in the output; mixing them hides confidence and assumptions
- Framework sprawl — picking more than one to three frameworks (step 2) dilutes focus; constrain to the question type
- Skipped context gathering — omitting step 0 leads to mis-scoped framework selection and analysis depth
- Untracked evidence quality — failing to note assumptions and source quality (step 4) makes conclusions unverifiable
Companion Resources
analytical-frameworks.md
output-templates.md