Adhoc AnalysisCAUTION

A framework for building agentic apps

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: adhoc-analysis
description: >-
  How to conduct ad-hoc analyses: gather data from multiple sources, synthesize
  findings, and save durable results as dashboard artifacts that anyone can re-run.
---

# Ad-Hoc Analysis

Ad-hoc analyses are deep-dive investigations that cross-reference multiple data sources and produce a written report with findings. The durable user-facing result is a dashboard. Answer in chat first unless the user explicitly asks to save the result, create a reusable artifact, or re-run/update an existing saved analysis.

## When to Use

Use the ad-hoc analysis workflow when:

- The user asks a complex question that requires data from multiple sources
- The investigation involves cross-referencing (e.g., CRM deals matched against call recordings)
- The user explicitly asks for an "analysis" or "deep dive"

Save a reusable dashboard artifact when:

- The user explicitly asks to save, create, publish, or re-run a saved analysis
- The user is already viewing or re-running an existing saved analysis
- The output needs a durable artifact because it includes generated chart images or a reusable refresh workflow

For one-off questions and exploratory deep dives, query the data and answer in chat. Do not create a dashboard or call the legacy `save-analysis` action just because the user said "analysis" or "deep dive".

If the user asks for an analysis output that needs a bespoke interactive
surface, custom visualization, multi-step workflow, or UI that cannot be
faithfully represented by native dashboard panels, create an extension and
immediately embed it in the dashboard as one or more `chartType: "extension"`
panels with `config.extensionId`. Never leave the extension as a standalone
Analytics result or direct the user to an Extensions page.

## Workflow

### Step 1: Understand the Question (catalog-first, clarify-first)

Orient before gathering data. Consult the preloaded `<resource scope="analytics-catalog">` references (else one `find-data`) and data-source status first to see which sources are configured and which one owns each fact, then settle scope:

- What is being analyzed? (deals, users, campaigns, errors, etc.)
- What time range?
- What data sources are relevant? (map each fact to the one source that owns it)
- What output does the user expect? (summary, ranking, comparison, trend)

If the metric definition, date range, or grain is ambiguous and a wrong guess would change the numbers, use the `ask-question` clarifying tool (multiple-choice) before gathering data. Ask at most once per turn, and skip it when the dictionary or the user already answered.

### Step 2: Gather Data from Multiple Sources

Use the available actions to pull data. Read the relevant `.agents/skills/<provider>/SKILL.md` before querying each source.

**Common data source combinations:**

| Analysis type           | Data sources                                                     |
| ----------------------- | ---------------------------------------------------------------- |
| Deal/account deep dive  | `account-deep-dive` bundle, then targeted HubSpot/Gong follow-up |
| Sales pipeline analysis | HubSpot deals + Gong calls + Slack mentions                      |
| Customer health check   | HubSpot deals + Pylon support tickets + BigQuery usage events    |
| Content performance     | BigQuery pageviews + GA4 + SEO keywords + HubSpot signups        |
| Engineering velocity    | GitHub PRs + Jira tickets + BigQuery deploy events               |
| Churn investigation     | Stripe billing + HubSpot deals + Pylon tickets + BigQuery usage  |

**Tips for data gathering:**

- Start with the primary source (e.g., HubSpot for deals), then enrich with secondary sources
- For named deal/account deep dives, call `account-deep-dive` first with the
  account, company, domain, deal, or opportunity name. It returns HubSpot deals,
  associated companies/contacts/tickets/notes/emails, Gong call detail, compact
  transcript excerpts, coverage counts, and gaps. Use targeted `hubspot-records`
  or `gong-calls` follow-ups only when that bundle leaves a specific gap.
- Structure named deal/account reports as: executive summary, company/deal
  overview, key contacts and roles, dated timeline, Gong evidence with call
  dates/titles, current state, risks/blockers, recommended next steps, and
  methodology/gaps. Do not answer from an all-deals dump or Gong metadata alone.
- Use action filters such as `query`, `properties`, `objectType`, `company`, and
  `limit` to narrow results before cross-referencing
- For HubSpot deal cohorts, use `hubspot-deals` structured filters (`product`,
  `pipeline`, `closedStatus`, `closedDateFrom`, `closedDateTo`) for the cohort
  definition. Do not use `query` when the user names a specific HubSpot field
  such as `products`.
- When any first-class provider action is too narrow, use
  `provider-api-catalog` / `provider-api-docs` and then `provider-api-request`
  against the provider's real HTTP API. Do not weaken the analysis just because
  the convenience action is missing an argument.
- When stitching identities across sources, follow `cross-source-analysis`: match on BOTH a stable id AND email (ids can be reassigned), de-duplicate, and record match quality. Email/company-name/domain matches alone are low-confidence — flag them as caveats, not headline numbers.
- If a data source is not configured, mention what's missing and work with what's available — never invent rows to fill a gap.

### Step 3: Analyze and Synthesize

Don't just dump raw data. Synthesize findings:

- Identify patterns, trends, and outliers
- Calculate key metrics (totals, averages, rates, distributions)
- Rank or categorize items when useful
- Call out surprises or actionable insights
- Compare against benchmarks or prior periods when possible
- Only report figures you actually retrieved from a source — never present a number you did not query. Attribute each figure to its source and time window.
- Make the eviden

CAUTIONgrade B · trust 89/100 Install with care. The audit found things worth knowing before you trust its output.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (5)

.agents/skills/an

.agents/skills/content-product-development

.agents/skills/design-exploration

.agents/skills/turn-into-app

.agents/skills/visual-edit

Gates applied: no_behavioural_pass.

[![ModelsAtlas audit: Adhoc Analysis](https://modelsatlas.com/skills/builderio/adhoc-analysis/badge.svg)](https://modelsatlas.com/skills/builderio/adhoc-analysis/)

Not your listing, or something out of date? Claim or update your listing.