data-readout
Raw numbers turned into a decision: dataset interpretation, statistics with the business implication of each finding, chart specifications, pattern and anomaly detection, forecasts built on stated assumptions, segmentation, A/B test significance, dashboard metric selection and report templates, output as an executive summary, findings ranked by impact and the specific actions they support. Use whenever someone brings data, a spreadsheet, a CSV, an analytics export, an A/B result or an unexpected number and asks what it means, what to chart, whether the test won or which metrics to track, even if they never say analysis. Not for writing the pandas or R code that crunches the file (use production-code), for market research outside the user's own data (use research-brief) or for the strategic roadmap that follows (use strategy-memo).
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- data-readout
- version
- 1.0.0
- description
- Raw numbers turned into a decision: dataset interpretation, statistics with the business implication of each finding, chart specifications, pattern and anomaly detection, forecasts built on stated assumptions, segmentation, A/B test significance, dashboard metric selection and report templates, output as an executive summary, findings ranked by impact and the specific actions they support. Use whenever someone brings data, a spreadsheet, a CSV, an analytics export, an A/B result or an unexpected number and asks what it means, what to chart, whether the test won or which metrics to track, even if they never say analysis. Not for writing the pandas or R code that crunches the file (use production-code), for market research outside the user's own data (use research-brief) or for the strategic roadmap that follows (use strategy-memo).
02 资源文件 3
- 参考文档references/analysis-playbook.md10.1 KB
- 参考文档references/report-format.md6.2 KB
- 参考文档references/statistical-checks.md6.5 KB
03 SKILL.md 正文
Data Readout
Overview
Turns the numbers the user already owns (a CSV, a spreadsheet, an analytics export, an A/B result, one unexpected figure) into a written readout: what the data says, how strongly it says it, and which action that supports. Nothing in the output is a guess about the file, and anything you could not compute is listed as a gap rather than filled in.
Intake
What it needs, in order of importance:
- The data itself: a path you can read, or rows pasted into the chat.
- The business question in the user's own words ("are we losing the small accounts", "did the test win").
- The decision the answer feeds and who reads it.
- Scope the file cannot tell you: which date range, which rows to exclude, what counts as one unit.
- For an A/B test: assignment unit, arms, start and end dates, the one primary metric, anything else watched.
- For a forecast: the horizon, and any known break in the series (price change, outage, launch, seasonal event).
Rules:
- Read the file before you describe it. If the data is on disk, open it and take the row count, column names, types and missingness from what you actually see, never from a guess about it. Report n.
- Ask once, in a single message, and cap it at three questions. If the user says go, proceed and print the assumptions you used.
- No file and no numbers means no invented dataset. Ship the framework with
[metric — supply this]slots, or ask for the export.
Modes
| The user asks | You deliver |
|---|---|
| "What does this data say?" (dataset interpretation) | Findings in plain English ranked by business impact, each carrying its n and its observed-or-inferred label |
| "Run the numbers" (statistical analysis) | Each statistic with test, p-value or interval, effect size, and the business implication in one line |
| "What should I chart?" (chart specification) | A written build spec per chart: axes, encoding, scale choice, sort order, annotations, and why that chart answers that question |
| "Is there a trend or a hidden pattern?" (pattern detection) | Candidate trends, correlations and outlier clusters with the statistic behind each, plus the ones that look like patterns and fail the check |
| "Where is this heading?" (forecast) | A range over the stated horizon, never a point, with the assumption list attached so any stakeholder can challenge it |
| "Split it by segment" (segmentation) | Segments with size, behaviour, the metric that differs, and the strategic implication of each |
| "Did the test win?" (A/B interpretation) | Significance, effect size, guardrail read, and one call: ship, keep running, or stop, with the reason |
| "Which metrics should I track weekly?" (dashboard selection) | A metric list, each with definition, formula, owner, review cadence, and a target marked [target — yours to set] unless the user gave one |
| "Make this repeatable" (report template) | A template mapping raw input to a formatted executive report, plus the checks every run must pass |
| "This number is weird" (anomaly explanation) | Ranked hypotheses, each with the check that would confirm or kill it and what that check needs |
Method per mode: references/analysis-playbook.md. Statistical checks and traps: references/statistical-checks.md. Report shape and the chart-spec table: references/report-format.md.
Output format
One artifact, in this order (fill-in rules in references/report-format.md):
- Executive summary: three to five lines, each with a number and the decision it changes.
- The data, briefly: source, n, date range, what you excluded and why, missingness.
- Findings ranked by impact, each labelled
Observed(the file shows it) orInferred(a plausible reading that still needs a check). - Actions: the finding each one rests on, the owner, the metric it moves.
- Chart specs for the findings that need a picture to be believed.
- Assumptions and caveats, then Gaps listing every open slot.
Worked fragment, shown with slots rather than invented results:
Finding 2 of 4 — Observed
Refunds on the [tier] tier rose from [x]% to [y]% between [month] and [month]
(n = [rows] orders; [test name], p = [value]; effect size [value], 95% CI [lo]–[hi]).
Business implication: [one line tying the number to revenue, retention or cost].
Action: [owner] verifies [specific check] by [date]; watch [metric] weekly.
Confidence: [high|medium|low] because [n, missingness, the one confounder].
Quality bar
- A pattern you can see is not a finding until it has a number and a test behind it. Never infer significance from appearance.
- Never write "statistically significant" without the test, the p-value or interval, and the effect size. A big sample makes tiny differences significant; say which you have.
- Observed and inferred stay visibly separated. A plausible cause must not read like a result.
- Forecasts ship as ranges with their assumptions listed. If an assumption breaks, the forecast is void, and you say so.
- Chart specs are build instructions for someone else. Never describe a screenshot you have not seen and never draw an image of data you did not compute.
- Benchmarks and industry averages appear only with a source and a date, otherwise the slot stays empty and marked
[unverified]. Targets are the user's to set. - Rank findings by impact on the metric the user named, not by how interesting the calculation was.
- Say what you did not do: columns you did not inspect, joins you could not verify, tests the toolchain would not let you run.
- An anomaly earns ranked hypotheses with confirming checks, never one confident cause.
Pre-handoff check, run against your own output:
- [ ] Every number traces to the file or to a calculation you ran on it, and n is stated.
- [ ] No statistic stands without its test, interval and effect size.
- [ ] Each finding carries
ObservedorInferred. - [ ] Every forecast is a range and carries its assumptions.
- [ ] Every chart spec names axes, encoding, scale and why that chart.
- [ ] Every metric target is the user's, or sourced and dated, or an open slot.
- [ ] No promise of an outcome ("this will lift conversion"); mechanisms and evidence only.
- [ ] The Gaps list is complete and matches every slot in the body.
- [ ] Missing data, exclusions and known confounders are named, not buried.
Handoffs
| Signal | Sibling | What you send |
|---|---|---|
| The user wants the pandas, R or SQL that produces these numbers as reusable code | production-code |
The calculation spec: input file, columns, filters, formula per metric, expected output shape, the checks you used to verify |
| A finding needs a figure from outside the user's data (market size, competitor behaviour, an industry rate) | research-brief |
The exact question, the granularity and period needed, and how the number will be used |
| The readout supports a decision the user now has to make and sequence | strategy-memo |
The ranked findings, the actions already implied, the constraints the data exposed |
| The user has already decided and wants the readout used to attack that decision | crash-test |
The findings with their confidence labels, and the checks that were never run |
| The user only wants the prose of the readout tightened | wordsmith |
The readout text with the numbers untouched |
Handoff block, in this exact shape:
HANDOFF — data-readout -> <target skill>
Goal: <one line>
Brief: <subject/dataset/role>
Evidence: <what the user supplied>
Assumptions: <what you assumed to continue>
Gaps: <unverified figures, missing inputs>
Next: <one line offering the target skill>
If the sibling is not installed, output the block anyway and say which skill normally owns it.
04 同仓库更多 Skills
tune-and-ship
Takes one published skill at a time from as-is to tuned, fact-checked, released and live, then moves to the next one in the catalog. Each run tunes the skill with tune-skill, checks every version number, flag, price and command it states against the primary source, re-reads its siblings' real SKILL.md files, polishes the listing with findable-polish, writes the version and release notes, opens one PR, and after the merge confirms the site page and marketplace copy are live. Use whenever the user wants to fine-tune their other skills, work through the catalog, give a published skill the full treatment, fact-check a skill's claims, or ship a tune instead of drafting it, even if they never say "ship". Not for a tune with no release (use tune-skill), drafts with nothing pushed (use tune-and-polish), a new skill (use skill-launch), regressions from the nightly health check (use caffeinity-routing's Descale brew), or a whole collection (use skill-estate).
skill-estate
Takes a whole collection of agent skill repos from scattered to published, consistent and self-maintaining, in gated phases: inventory every repo, reconcile site, marketplace and README listings, test and tune the family, close workflow gaps, wire publishing so new skill repos go live on their own, bundle a plugin with agents, add trigger tests and directory listings, and leave a maintenance loop and a map. Each phase hands off to the skill that owns it and ends in PRs. Use when someone's skills site and GitHub disagree, wants all their skill repos audited or cleaned up, wants new skills to publish themselves, or wants that cleanup repeated on another org, even if they never say estate. Not for one skill (use tune-skill), one listing (use findable), the site pipeline alone (use publish-pipeline), or weekly upkeep (use caffeinity-routing's Descale brew).
paper-trail
Drafts the policy pages a small product needs before it takes money or data (privacy policy, terms of service, refund policy, cookie notice) from a fact sheet about what the product actually collects, stores, shares and charges, then checks the drafts for placeholders, missing sections and facts nobody supplied. Use when someone needs a privacy policy, terms, refund or cookie page for a site, app, SaaS, newsletter or digital product, asks what legal pages they need before launch, or wants existing policy pages checked against what the product really does, even if they never say "legal". Not legal advice and not for contracts, NDAs or employment documents (use a lawyer), and not for the launch plan itself (use solo-income-blueprint) or the site build (use site-prompt).
course-builder
Designs a course for a group of learners (a class, cohort, workshop series, bootcamp or internal training): measurable outcomes, a module sequence with prerequisites, activities and assessments aligned to each outcome, a time budget that fits the hours available, and a rubric, then checks the plan with a script so no outcome goes untaught or unassessed. Use when someone is teaching others and asks for a curriculum, syllabus, course outline, module plan, cohort or workshop structure, or wants an existing course checked for gaps and overload, even if they never say "curriculum". Not for one person's own learning plan (use study-system), for being taught the topic (use topic-primer), or for marketing the course (use sales-copy).
wordsmith
Write, draft, rewrite, edit, proofread, polish and improve text while preserving the author's intended meaning, facts and voice. Use when the user wants clearer, shorter or better-toned writing; grammar, structure or flow fixes; emails, website or marketing copy, headlines, outlines or summaries; a different reading level; an accessibility, inclusivity or accuracy review; or content prepared for publication, even if they never say "edit". Also handles /wordsmith commands such as draft, critique, audit, tighten, tone and final. Not for resumes, CVs or LinkedIn profiles (use recruiter-lens if installed), not for code or code comments, and not for building .docx or .pdf files (write the text here, then use the file skill).
web-preflight
Use when shipping, deploying, publishing, launching, reviewing, or pre-flighting a website or API and you want a lightweight check for accidental exposure of secrets, source or configuration files, admin and debug surfaces, unsafe redirects, missing security controls, and misleading 200 responses. Trigger phrases include "before we ship", "security check", "preflight", "audit the site", "check the production URL", "is my .env exposed", and "why does /.env return 200", including when access-log or analytics evidence is supplied. Checks only sites the user owns or is authorized to assess, with bounded GET and HEAD requests and a fixed compact path set; distinguishes genuine resources from SPA catch-all and blanket WAF responses. Not for penetration testing, exploitation, credential testing, broad port scanning, or application-code review; use a dedicated authorized security-testing or code-review workflow instead.