agentsclimarketplace

Eskill analyze

Skill Renn-Labs/eskill-analyze/skills/eskill-analyze

World-class level-up analysis with first-principles thinking, systems mapping, and strategic frameworks. Produces prioritized action plans with project-level impact assessment. Absorbs eskill-orchestrate routing. Also known as orchestrate, eo, or the long-form native analyzer; for the short Tier-1 entry point prefer /esa.From its SKILL.md

Install
npx -y skills add Renn-Labs/eskill-analyze --skill eskill-analyze

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

7.7 KB, as published. Nobody here has run it

eSkill: Analyze — World-Class Level-Up Analyzer

First-principles analysis engine that identifies what elevates any product, feature, or system to world-class status. Produces a prioritized action plan ready for sprint execution.

Quality Principles

Read ${CLAUDE_SKILL_DIR}/../eskill-common/references/quality-principles.md

Input Parameters

Extract at start. If any are missing, ask before proceeding.

Scope check:

  • Bug/incident? If the prompt describes broken behavior ("isn't working", "broke", "crashed", "dropped", "why did X fail") — clarify intent. If the user wants diagnosis, route to a debugger agent. eskill-analyze is for elevation, not incident triage.
  • Non-software? If the prompt is about hiring, org structure, team process, or other non-product/non-system topics — only First Principles, Inversion, and Reversibility models apply. Skip all system-specific steps (System Mapping, AAARRR, 7-Step Path, Competitive Analysis). Keep the analysis lightweight and framework-driven rather than forcing product/technical structure onto people or process decisions.
ParameterDescriptionRequired
Project/ProductWhat are we analyzing?Yes
Focus AreaSpecific aspect to deep-diveYes
Current StateWhat exists, what's built, what's brokenYes
World-Class DefinitionWhat does "done" look like?Yes
ConstraintsTimeline, resources, team sizeNo

Triage (After Parameters Confirmed)

Read ${CLAUDE_SKILL_DIR}/references/triage-guide.md

Gate: Do not triage until the required Input Parameters above are extracted or confirmed. If the prompt is too vague to determine Project, Focus Area, Current State, or World-Class Definition — ask first. Triage with missing context wastes agent dispatches on the wrong problem.

Classify the prompt:

  1. Type: technical / product / strategic / comparative
  2. Scale: quick / standard / deep
  3. Agents: which of the 6 agents to activate
  4. Models: which 3-4 mental models to apply
  5. Phases: which phases are active (Phase 0, Steps 1-8, Phase P Premortem, Phase 9)

State the triage result before proceeding. This determines everything that follows.

Delegation Strategy

Spawn only the agents triage activates. All eligible agents run in parallel.

Always Active

Task(subagent_type="oh-my-claudecode:analyst", model="opus",
     prompt="Analyze requirements and hidden constraints for: {focus_area}.
     Final response under 2000 characters. List findings, not process.")

Task(subagent_type="oh-my-claudecode:explore", model="sonnet",
     prompt="Map existing codebase structure relevant to: {focus_area}.
     Final response under 2000 characters. List findings, not process.")

Activated by Triage

# Technical analysis
Task(subagent_type="oh-my-claudecode:architect", model="opus",
     prompt="Map system boundaries, interfaces, and long-term tradeoffs for: {focus_area}.
     Final response under 2000 characters.")

# Product or strategic analysis
Task(subagent_type="oh-my-claudecode:document-specialist", model="sonnet",
     prompt="Research world-class benchmarks and best practices for: {domain}/{focus_area}.
     Final response under 2000 characters.")

# When quantitative data exists
Task(subagent_type="oh-my-claudecode:scientist", model="sonnet",
     prompt="Analyze available metrics, logs, usage data for: {focus_area}.
     Final response under 2000 characters.")

# When security keywords detected
Task(subagent_type="oh-my-claudecode:security-reviewer", model="sonnet",
     prompt="Assess security surface, threat model, and blast radius for: {focus_area}.
     Final response under 2000 characters.")

Run all activated agents in parallel. Synthesize their outputs before proceeding to analysis.

Framework Selection

NOT every analysis needs every framework. After triage:

  1. Read ${CLAUDE_SKILL_DIR}/references/mental-models.md
  2. Use the auto-selection logic based on triage type
  3. Pick 3-4 frameworks max from the recommended set
  4. Skip the rest — do not generate empty or forced sections

Analysis Protocol

Read ${CLAUDE_SKILL_DIR}/references/analysis-protocol.md

Execute in order:

  1. Phase 0: Evidence Gathering — if triage activated it, run the evidence agents and synthesize
  2. Steps 1-8: Core Analysis — apply only selected frameworks, skip irrelevant steps entirely; Step 6 is an inversion seed only
  3. Phase P: Premortem Cycle — if triage activated it (standard/deep, or quick + keyword), freeze the bet, generate failure modes + kill criteria, update provisional actions; see analysis-protocol Phase P
  4. Phase 9: Stress Test — if triage activated it (standard/deep scale), run the critic agent and integrate feedback (including attacks on kill criteria when Phase P ran)

Comparison Mode

When triage classifies as comparative:

  1. Cap at 2 options. For 3+, shortlist to 2 finalists first with a quick-triage pass
  2. Run triage + delegation for each option, but adapt per option:
    • Existing option (code exists): run explore + architect + analyst
    • Hypothetical option (no codebase): run document-specialist + analyst (skip explore)
  3. Use the Comparison Mode template from ${CLAUDE_SKILL_DIR}/assets/output-template.md
  4. Produce a side-by-side table across key dimensions
  5. State a clear recommendation with tradeoffs and conditions to reconsider
  6. Generate prioritized actions for the recommended option only

World-Class Signals

Read ${CLAUDE_SKILL_DIR}/references/world-class-signals.md

Project-Level Impact

Read ${CLAUDE_SKILL_DIR}/../eskill-common/references/project-impact-protocol.md

If {project}/.omc/project-context.json and {project}/.omc/business-context.md exist, connect every recommended action to project priorities and business impact in the output.

Output Format

Read ${CLAUDE_SKILL_DIR}/assets/output-template.md

Present the analysis using the appropriate template variant:

  • Standard template: for technical/product/strategic analyses
  • Comparison template: for comparative analyses

Include only the sections that are relevant:

  • Triage header: always
  • Evidence Summary: only when Phase 0 ran
  • Premortem Cycle: only when Phase P ran (standard/deep, or quick + keyword)
  • Stress Test Notes: only when Phase 9 surfaced material changes
  • Confidence column: only when stress test and/or Phase P adjusted ratings
  • Mermaid diagrams: only when system relationships are complex (3+ components)

Post-Analysis Routing

After presenting the prioritized action list:

"Here are the prioritized actions. Which do you want to execute?

  • All of them
  • Top N (specify)
  • Specific items (list numbers)

Want to run this as an overnight sprint? Say /eskill-overnight with the confirmed list."

Auto-routing:

  • < 10 tasks confirmed → invoke /eskill-sprint
  • 10+ tasks confirmed → invoke /eskill-overnight
  • User says "overnight" → always /eskill-overnight

Save the analysis output to .omc/plans/analysis-{date}-{focus}.md for reference during execution.

Execution Principles

  1. Never accept "good enough" — Ask "what would world-class look like?"
  2. Start with why — Reconnect to first principles before proposing
  3. Think in systems — Single features don't create world-class; ecosystems do
  4. Assume competition copies — Make it hard to replicate
  5. Name the flywheel — Can't describe compounding? Not ready.

Keep looking

Skills are one crate of 608,666. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.