agentleFS
link7373link7373/agentic-bi-team/.claude/skills/analyze/SKILL.md

analyze

Skill4 starsChanged 6 days agoMITView on GitHub

Full analysis workflow for a business problem — scoping the decision, profiling data, cross-database joins, segmentation/cohort/funnel/trend analysis, pressure-testing, and a written findings report. Use for any substantive business question.

71 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use constitution, revise and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • ArvorCoSkill

    constitution

    Create or update the project constitution - personalized governance and coding principles. Use when starting a project or updating project standards. Triggers on: create constitution, update principles, set project rules.

    Also in 4 projects that use this

  • stormlightlabsSkill

    revise

    Address review comments on an open pull request and push the fixes. Use for /edit, /revise, or when asked to respond to review feedback on a PR number.

    Also in 3 projects that use this

  • DockerSkill

    write

    Write or edit reader-facing technical prose for immediate comprehension. Use for documentation, PR titles and descriptions, release notes, design documents, user-facing explanations, and substantive comments. Also use when asked to make writing clearer, more natural, less AI-generated, or easier to read. For documentation fixes, handles edits, formatting, self-review, and commits after research identifies what to change. Do not use for code-only tasks with no prose deliverable.

    Also in 6 projects that use this

  • ArvorCoSkill

    specify

    Create feature specification from natural language description. Use when starting a new feature or creating a spec. Triggers on: create spec, specify feature, new feature spec.

    Also in 5 projects that use this

  • H-mmerSkill

    submit

    Draft and submit a vulnerability report to the bug bounty platform. Reads scope.yaml for platform/program, uses brain + findings for content. Always drafts first for review.

    Also in 3 projects that use this

  • GoogleSkill

    quality

    Evaluates whether a GitHub issue is spam, empty, needs more information, or is OK to proceed.

    Also in 3 projects that use this

  • +1 more from ArvorCo

The file

Files
SKILL.md

Analyze — Business Problem Investigation

Primary owner: bi-analyst. Args carry the question, e.g. /analyze why is enterprise churn rising.

Procedure

  1. Frame the decision. Restate as: "We are deciding ___; this analysis informs it by ___; the answer would change the decision if ___." If the user's question is exploratory ("tell me about our customers"), propose 3 sharp sub-questions and confirm direction.
  2. Plan. Create analyses/YYYY-MM-DD-<slug>/ with a PLAN.md: hypothesis tree, data needed, join strategy, expected outputs. Make each hypothesis falsifiable and rank them by expected impact × ease of checking — test the cheapest decisive one first rather than boiling the ocean. For multi-day-scale work, share the plan before executing.
  3. Find the source — by tier, in order. (1) Does the question map to a metric in knowledge/metrics-catalog.md? Then get its SQL with python scripts/run_metric.py "<name>" (add --segment <name> for a defined population) and build on that — do not retype it. (2) If not, read the domain doc in knowledge/domains/ for the governed mart, its grain, its standard hygiene filter, and its gotchas. (3) Only if neither covers it, go to staging or raw tables — and record that you did, because it sets the provenance tier. Don't bail to a lower tier early: "it needs a custom date range" or "it needs a join" are reasons to adapt the catalog SQL, not to abandon it. Check knowledge/governance-policy.md before selecting anything about people: aggregate in the warehouse, and never pull a restricted column row-level.
  4. Gather & profile. Pull from sources per knowledge/data-sources.md. Profile every dataset before use: row counts, date coverage, key uniqueness, null rates. For cross-database joins: identify join keys, measure and report match rates, characterise the unmatched population (bias check). Materialise intermediates (local DuckDB/sqlite/parquet is fine) instead of repeated cross-system pulls.
  5. Analyse. Apply the appropriate toolkit — trend with seasonality control, cohort, funnel, segmentation (cut by the dimensions that matter to {{COMPANY_NAME}}), contribution/mix-shift, benchmark vs target/prior/peer. Use canonical metric definitions from knowledge/metrics-catalog.md only. If the question is "did X cause Y" or "what was the impact of", a before/after comparison is not an answer — hand it to /causal-analysis, or label the finding as correlational (analytics.md Part 4).
  6. Pressure-test before reporting. Different date window, different segment cut, denominator artifacts, Simpson's paradox on ratio comparisons, outlier sensitivity. Run the trap checklist in analytics.md Part 1 (Simpson's, base rate, inspection/collider sampling bias, regression to the mean) and the reporting hygiene checklist in standards/reporting-standards.md — including median-not-mean for skewed data and causal-vs-correlational labelling. Note which checks were run.
  7. Write FINDINGS.md (per standards/reporting-standards.md):
    • Answer (1–3 sentences, decision-relevant)
    • Key findings (each: claim + quantified evidence)
    • So what / recommendation
    • Methodology & queries (everything reproducible from the folder)
    • Caveats & open questions
    • Flagged side-findings (anything important spotted off-topic)
    • Provenance footer — source tier (the least governed one used), data-through date and when freshness was checked, owner, review status
  8. Independent review. Hand the folder to analysis-reviewer when the finding is leadership-bound or external, when it is surprising, when it rests on raw tables, or when money moves on it. Fix what it blocks, then update the footer's Review: field with the verdict and date. For a quick internal answer, self-check and say so in the footer (Review: self-checked) — never claim a review that did not happen.
  9. Hand off as needed. Stakeholder-grade packaging → insights-communicator (/make-deliverable). Recurring need → propose a dashboard or summary table. Modelling opportunity → note it for data-scientist.
  10. Update memory. New data quirks → knowledge/data-sources.md and the domain doc in knowledge/domains/. Methodological choices → knowledge/decision-log.md. If a stakeholder corrected the answer, that is an eval: /eval add.

Created by Colin Beck — https://www.linkedin.com/in/beckcolin/

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.