Launch prep: startup-data-kit hook, one-line installs, sample-data demo

.claude-plugin/marketplace.json

@@ -0,0 +1,71 @@
+{
+  "name": "bruin-agent-resource-kit",
+  "owner": {
+    "name": "Bruin",
+    "url": "https://github.com/bruin-data"
+  },
+  "description": "Example agent skills for working with a startup's data through Bruin. Sample code, provided as-is; not a supported Bruin product.",
+  "plugins": [
+    {
+      "name": "bruin-agent",
+      "description": "Install the Bruin CLI, create a project, choose a template, connect a source and register the Bruin MCP server.",
+      "source": "./",
+      "strict": false,
+      "skills": [
+        "./resources/bruin-agent"
+      ]
+    },
+    {
+      "name": "startup-data-kit",
+      "description": "Set up a Bruin template for a startup's data and answer questions from it: revenue, marketing, product, web, AI search, finance, support, AI coding spend, sheets and Notion.",
+      "source": "./",
+      "strict": false,
+      "skills": [
+        "./resources/startup-data-kit/ai-coding-spend",
+        "./resources/startup-data-kit/ai-search-visibility",
+        "./resources/startup-data-kit/finance-accounting",
+        "./resources/startup-data-kit/marketing-analytics",
+        "./resources/startup-data-kit/product-analytics",
+        "./resources/startup-data-kit/revenue-analytics",
+        "./resources/startup-data-kit/sheets-and-notion",
+        "./resources/startup-data-kit/support-analytics",
+        "./resources/startup-data-kit/web-analytics"
+      ]
+    },
+    {
+      "name": "data-engineering-skills",
+      "description": "Keep Bruin pipelines healthy: triage failures, investigate anomalies, backfill safely, open maintenance pull requests, report what happened.",
+      "source": "./",
+      "strict": false,
+      "skills": [
+        "./resources/data-engineering-skills/anomaly-investigate",
+        "./resources/data-engineering-skills/bruin-builtin-skills",
+        "./resources/data-engineering-skills/maintenance-pr",
+        "./resources/data-engineering-skills/pipeline-backfill",
+        "./resources/data-engineering-skills/pipeline-report",
+        "./resources/data-engineering-skills/pipeline-triage"
+      ]
+    },
+    {
+      "name": "coding-agent-workflows",
+      "description": "Repeatable work on a codebase: customer-facing changelogs and recorded terminal demos.",
+      "source": "./",
+      "strict": false,
+      "skills": [
+        "./resources/coding-agent-workflows/generate-changelog",
+        "./resources/coding-agent-workflows/record-terminal-demo"
+      ]
+    },
+    {
+      "name": "tools",
+      "description": "Sample scripts an agent can shell out to, run with your own credentials: Attio CRM, blog cover images, LinkedIn enrichment via Apify.",
+      "source": "./",
+      "strict": false,
+      "skills": [
+        "./resources/tools/attio-cli",
+        "./resources/tools/blog-image-generator",
+        "./resources/tools/linkedin-enrichment"
+      ]
+    }
+  ]
+}

.gitattributes

@@ -0,0 +1,15 @@
+# GitHub language statistics (Linguist). This repository is a catalogue of
+# Markdown skills; the Python is sample code and tooling around them.
+# https://github.com/github-linguist/linguist/blob/main/docs/overrides.md
+
+# Markdown is a prose language, which Linguist leaves out of the stats unless
+# it is marked detectable.
+*.md linguist-detectable
+
+# Sample scripts and the checker: excluded from the stats.
+resources/**/scripts/** linguist-vendored
+*.py linguist-vendored
+*.py -linguist-detectable
+
+# Generated by sample-data/generate.py: collapsed in diffs, excluded from stats.
+resources/**/sample-data/*.csv linguist-generated

.github/PULL_REQUEST_TEMPLATE.md

@@ -16,7 +16,7 @@ One or two sentences.
 
 - [ ] `name` matches the folder, `description` starts with `Use when`
 - [ ] `**Status:**` and `**Risk:**` line, chosen conservatively
-- [ ] Listed in the root `README.md` index
+- [ ] Listed in the root `README.md` index and in `.claude-plugin/marketplace.json`
 - [ ] Says what the tool cannot do, and what the user has to decide
 - [ ] No credentials, customer data or production output
 - [ ] Does not ask a user to paste a credential into a conversation

.github/workflows/check.yml

@@ -22,8 +22,9 @@ jobs:
         with:
           python-version: "3.12"
 
-      # Frontmatter, status labels, upstream notes and their dates, index
-      # completeness, local links, forbidden files. No dependencies.
+      # Frontmatter, status labels, upstream notes and their dates, index and
+      # plugin manifest completeness, local links, sample data, forbidden
+      # files. No dependencies.
       - name: Validate catalogue
         run: ./tests/check.py
 

.gitignore

@@ -14,5 +14,7 @@ venv/
 # Local databases from trying a template.
 *.db
 *.db.wal
+*.duckdb
+*.duckdb.wal
 
 .DS_Store

AGENTS.md

@@ -158,12 +158,13 @@ confirm. Never write an unchecked claim as though you had checked it.
 
 Add a `**Status:**` and `**Risk:**` line under the heading. Be conservative:
 `experimental` unless someone other than the author has used it. Add the
-resource to the index in [`README.md`](README.md); the checker fails if you do
-not.
+resource to the index in [`README.md`](README.md) and to its group's plugin in
+[`.claude-plugin/marketplace.json`](.claude-plugin/marketplace.json); the
+checker fails if you do not.
 
 Run `./tests/check.py` before finishing. No dependencies. It verifies
-frontmatter, labels, upstream notes and their dates, index completeness, local
-links, and forbidden files.
+frontmatter, labels, upstream notes and their dates, index and plugin manifest
+completeness, local links, sample data, and forbidden files.
 
 Never commit a credential, customer record or production output. `.bruin.yml`,
 `.env` and similar are gitignored, and both the checker and gitleaks will fail

CONTRIBUTING.md

@@ -57,7 +57,11 @@ The rules the checker enforces:
   carries a date. A file that drives no tool gets a one-line precedence note
   instead, saying what wins where it disagrees; the checker does not enforce
   that one.
-- Every resource is listed in the root `README.md` index.
+- Every resource is listed in the root `README.md` index, and its folder in one
+  plugin's `skills` list in `.claude-plugin/marketplace.json`. Both one-line
+  installs read that manifest, so a skill missing from it cannot be installed.
+- Sample data lives in a `sample-data/` folder, is invented, and uses only
+  `example.com`, `example.org` or `example.net` email addresses.
 - Local links resolve, and no internal identifiers (Slack channel IDs, project
   ULIDs, service accounts) or forbidden files (`.bruin.yml`, `.env`) appear.
 - Files shared across a group of skills stay identical. Every
@@ -142,7 +146,8 @@ prevent.
 ```
 
 No dependencies. It checks frontmatter, status labels, upstream notes and their
-dates, index completeness, local links, and forbidden files.
+dates, index and plugin manifest completeness, local links, sample data, and
+forbidden files.
 
 Then say, in the pull request: what the resource does in one sentence, what you
 ran to verify it, which status and risk you chose and why, and anything you

README.md

@@ -1,5 +1,45 @@
 # agent-resource-kit
 
+**Give Claude Code, Cursor and Codex your startup's data.**
+[`startup-data-kit`](resources/startup-data-kit/) is a set of agent skills that
+set up a maintained Bruin pipeline for revenue, marketing, product, finance or
+AI coding spend, then answer questions from it without the confident wrong number.
+
+```bash
+npx skills add bruin-data/agent-resource-kit
+```
+
+![The ai-coding-usage template building on fake Claude Code and Cursor data in DuckDB, then a weekly cost query](demo/ai-coding-spend.gif)
+
+### Try it in 2 minutes
+
+No credentials and no warehouse: the real `ai-coding-usage` template, fake
+Claude Code and Cursor usage, DuckDB. Needs the
+[Bruin CLI](https://getbruin.com/docs/bruin/getting-started/introduction/installation.html)
+and git; run it in an empty folder outside any git repository.
+
+```bash
+# 1. Get the template and the sample data
+git clone --depth 1 https://github.com/bruin-data/agent-resource-kit
+bruin init ai-coding-usage ai-coding-demo
+
+# 2. Swap the Anthropic and Cursor API assets for the fake CSVs
+cp agent-resource-kit/resources/startup-data-kit/ai-coding-spend/sample-data/*.{asset.yml,csv} \
+   bruin/ai-coding-demo/assets/raw/
+
+# 3. Build every table on DuckDB, then query one
+cd bruin
+bruin run ai-coding-demo --full-refresh --workers 1 --start-date 2026-09-14 --end-date 2026-09-27
+bruin query -c duckdb-default -q "SELECT strftime(date_trunc('week', usage_date), '%Y-%m-%d') AS week, platform, CAST(SUM(total_tokens) AS BIGINT) AS tokens, ROUND(SUM(estimated_cost_usd), 2) AS est_cost_usd FROM marts.ai_coding_usage_by_user_day GROUP BY ALL ORDER BY ALL"
+```
+
+Then ask your agent, with the skill installed, why estimated Claude Code cost
+rose in the second week. [`sample-data/`](resources/startup-data-kit/ai-coding-spend/sample-data/)
+says what the data is and why each flag is there; `--full-refresh` is only safe
+because the demo database is new.
+
+## What is here
+
 **Open-source skills for AI agents.** Drop a folder into your project and your
 agent knows how to do the thing.
 
@@ -15,13 +55,6 @@ to ask questions, work through a setup with someone, and share what you have
 built with these. If a skill sent your agent somewhere unhelpful, that is worth
 saying there as much as in an issue.
 
-> **Example code, provided as-is.** This repository is not part of the Bruin
-> platform, not a supported Bruin product, and not covered by any support
-> agreement or SLA. It is documentation and sample scripts: it stores no customer
-> data, holds no credentials, and has no access to any production system. Scripts
-> run on your own machine, under your own credentials, configured by you. See
-> [LICENSE](LICENSE) for the warranty disclaimer.
-
 ## Resources
 
 **Setup**
@@ -85,7 +118,26 @@ and do not guarantee.
 
 ## Use it
 
-**Copy the folder.** Your agent picks it up with no further setup.
+**With [skills.sh](https://skills.sh)**, for Claude Code, Cursor, Codex and
+the other agents it supports. It lists every skill here and asks which to
+install and for which agents.
+
+```bash
+npx skills add bruin-data/agent-resource-kit
+npx skills add bruin-data/agent-resource-kit --skill ai-coding-spend   # just one
+npx skills add bruin-data/agent-resource-kit --list                    # look first
+```
+
+**As a Claude Code plugin.** Each group is a plugin: `bruin-agent`,
+`startup-data-kit`, `data-engineering-skills`, `coding-agent-workflows` and
+`tools`. Plugin skills are namespaced, such as `/startup-data-kit:ai-coding-spend`.
+
+```text
+/plugin marketplace add bruin-data/agent-resource-kit
+/plugin install startup-data-kit@bruin-agent-resource-kit
+```
+
+**Or copy the folder.** Your agent picks it up with no further setup.
 
 ```bash
 git clone https://github.com/bruin-data/agent-resource-kit
@@ -108,6 +160,17 @@ I need to work out why our MRR stopped growing last quarter.
 
 It reads the index, then the one skill it needs.
 
+## Tested
+
+> **TODO:** results pending. The comparison runs separately and fills in this
+> table.
+
+| Question | Without skill | With skill |
+|---|---|---|
+| TODO | TODO | TODO |
+| TODO | TODO | TODO |
+| TODO | TODO | TODO |
+
 ## Upstream documentation is the source of truth
 
 Every file here is a snapshot of how something worked the day it was written.
@@ -158,10 +221,20 @@ has the shape, the labels, and what gets rejected.
 ```
 
 No dependencies. It checks frontmatter, status labels, upstream notes and their
-dates, index completeness, local links, and files that must never be committed.
+dates, index and plugin manifest completeness, local links, sample data that
+uses only reserved example domains, and files that must never be committed.
 CI runs it on every change, scans for credentials with gitleaks, and checks
 external links weekly so a dead upstream link surfaces on its own.
 
+## Provided as-is
+
+> **Example code, provided as-is.** This repository is not part of the Bruin
+> platform, not a supported Bruin product, and not covered by any support
+> agreement or SLA. It is documentation and sample scripts: it stores no customer
+> data, holds no credentials, and has no access to any production system. Scripts
+> run on your own machine, under your own credentials, configured by you. See
+> [LICENSE](LICENSE) for the warranty disclaimer.
+
 ## Licence
 
 MIT. See [LICENSE](LICENSE).

demo/ai-coding-spend.tape

@@ -0,0 +1,51 @@
+# Try ai-coding-spend in 2 minutes: the ai-coding-usage template on fake data.
+#
+# Recorded with Bruin CLI v0.11.767 and VHS. Re-render from this folder:
+#
+#   python3 ../resources/coding-agent-workflows/record-terminal-demo/scripts/render_demo.py ai-coding-spend.tape
+#
+# Hidden setup stands in for step 1's `git clone`: it links this working tree
+# into a fresh folder as agent-resource-kit, so the recording shows the sample
+# data at this commit rather than whatever is on GitHub.
+
+Output ai-coding-spend.gif
+
+Set Shell bash
+Set FontSize 18
+Set Width 1280
+Set Height 760
+Set Margin 0
+Set Padding 28
+Set LineHeight 1.15
+Set TypingSpeed 25ms
+Set Framerate 30
+Set CursorBlink false
+Set Theme "GitHub Dark"
+
+Hide
+Type "export KIT=$(git rev-parse --show-toplevel)" Enter
+Type "rm -rf /tmp/try-ai-coding-spend && mkdir /tmp/try-ai-coding-spend && cd /tmp/try-ai-coding-spend" Enter
+Type "ln -s $KIT agent-resource-kit" Enter
+Type "unset NO_COLOR && export TERM=xterm-256color CLICOLOR=1" Enter
+# \[ \] mark the colour codes as zero-width, so readline wraps the long query
+# line correctly.
+Type `export PS1="\n\[$(tput setaf 6)\]> \[$(tput sgr0)\]"` Enter
+Type "clear" Enter
+Show
+
+Type "# 1. The real ai-coding-usage template" Enter
+Type "bruin init ai-coding-usage ai-coding-demo" Enter
+Sleep 2s
+
+Type "# 2. Fake Claude Code and Cursor CSVs instead of the API assets" Enter
+Type "cp agent-resource-kit/resources/startup-data-kit/ai-coding-spend/sample-data/*.{asset.yml,csv} bruin/ai-coding-demo/assets/raw/" Enter
+Sleep 1s
+Type "clear" Enter
+
+Type "# 3. Build every table on DuckDB, then query one" Enter
+Type "cd bruin" Enter
+Type "bruin run ai-coding-demo --full-refresh --workers 1 --start-date 2026-09-14 --end-date 2026-09-27" Enter
+Sleep 4s
+Type "clear" Enter
+Type@15ms `bruin query -c duckdb-default -q "SELECT strftime(date_trunc('week', usage_date), '%Y-%m-%d') AS week, platform, CAST(SUM(total_tokens) AS BIGINT) AS tokens, ROUND(SUM(estimated_cost_usd), 2) AS est_cost_usd FROM marts.ai_coding_usage_by_user_day GROUP BY ALL ORDER BY ALL"` Enter
+Sleep 6s

resources/startup-data-kit/README.md

@@ -25,14 +25,18 @@ domain reconciling against its source before adding a second.
 
 ## How each skill works
 
-Every folder holds exactly these three files:
+Every folder holds these three files:
 
 | File | What it is |
 |---|---|
 | `SKILL.md` | What is specific to the domain: which templates to consider, what usually needs the user's input, what to reconcile against |
 | `workflow.md` | The shared setup process, from installing Bruin to reconciling the first run |
 | `analysis.md` | How to answer questions once the pipeline exists |
 
+`ai-coding-spend` also carries a [`sample-data/`](ai-coding-spend/sample-data/)
+folder: fake Claude Code and Cursor usage that runs its template on DuckDB with
+no credentials.
+
 `workflow.md` is identical in every folder, so a skill copied alone still works.
 `tests/check.py` fails if the copies drift; edit one and copy it to the rest.
 

resources/startup-data-kit/ai-coding-spend/SKILL.md

@@ -10,7 +10,7 @@ description: Use when a startup wants AI coding tool usage and spend set up or a
 > **Upstream documentation wins.** Where this file disagrees with the [Bruin
 > docs](https://getbruin.com/docs/bruin/overview.html), `bruin --help`, or
 > Anthropic's or Cursor's API documentation, those are right and this file is
-> stale. Written against Bruin CLI `v0.11.766`, checked 2026-10-01.
+> stale. Written against Bruin CLI `v0.11.767`, checked 2026-10-02.
 
 Sets up a Claude Code and Cursor usage pipeline and answers adoption and cost
 questions from it. It exists to prevent an estimated cost being read as the
@@ -20,6 +20,10 @@ bill, and usage data being turned into a ranking of engineers.
 
 Follow [workflow.md](workflow.md).
 
+**No credentials yet?** [sample-data/](sample-data/README.md) runs the template
+on fake usage in DuckDB. Use it to show the user what they will get; never
+present its numbers as theirs.
+
 > **Important: do not ask a fixed list of questions.** Read the template as it
 > is today and work out which of its choices this user needs to confirm.
 > Templates change, so the questions change with them.
@@ -33,6 +37,7 @@ theirs to build.
 - Coverage is Claude Code on the first-party Anthropic API, and Cursor; Bedrock, Vertex AI, Foundry, Claude Platform on AWS, Claude Enterprise claude.ai users and every other tool are silently missing.
 - Cost columns are estimates, not the bill, built differently per vendor, with no seat fees; no seat list is ingested.
 - History lives only in the marts; `--full-refresh` drops it.
+- A first run fails on the per-day marts until they exist; Bruin's materialization docs say to create them with `--full-refresh`, the template README does not. Ask first, and only while there is no history.
 
 **Usually needs the user's input:**