Data Platform Scaling Harnesses and Retro

Date: 2026-03-03
Status: Draft / Planning
Related: Delivery & Finance Analytics Plan, Agent-powered data environment (plan folder), PR #236


Context

Two converging streams:

  1. Delivery & Finance Analytics — Rill deployment (DP-195–DP-200), ingest, dashboards. Merged PRs: #247 (foundation), #255 (delivery mart), #258 (Rill dashboard), #250 (dbt init), #249 (Rill DSN).
  2. Agent-Powered Data Environment (PR #236) — Strategy for worker/meta/orchestrator model, dbt PR impact pipeline, self-learning agents. Branch: cursor/agent-powered-data-environment-a357.

Current gaps: No CI for rill/, knowledge/engineering/data-platform/, or dbt. No Cursor rules, commands, or skills for data work. AGENTS.md omits data platform. No playbook prompts for Rill/dbt workflows.


Part 1: Retro — Extract Learnings from Logs and PRs

1.1 Log Sources to Mine

SourceLocationWhat to Extract
Cursor agent transcripts~/.cursor/projects/.../agent-transcripts/*.jsonlFailed flows, repeated corrections, context gaps, tool usage patterns
PR comments and reviewsGitHub API / gh pr viewReviewer feedback, merge-blocking issues, “why did you…” patterns
GTM RUN_LOG patternRUN_LOG.mdAdapt for data runs: Run ID, context, input, output, decisions, outcome
Eden agentic learningslearnings“Ask don’t fill”, scope to input only, cross-contamination pitfalls
Vercel agent learningsVERCEL_LEAD_AGENT_LEARNINGS.mdBuy-in > build, start small, shadow best people

1.2 Retro Process

  1. Gather — Export agent transcripts for data-related sessions (filter by paths: rill/, knowledge/plans/agent-powered-data-environment/, knowledge/engineering/data-platform/, standards/03-knowledge/engineering/setup/). Pull PR comments for #236, #247, #255, #258, #250, #249.
  2. Categorize — Group by: setup/credentials, dbt/Rill workflow, CI/deployment, agent context gaps, reviewer friction.
  3. Synthesize — Produce a retro doc: knowledge/engineering/data-platform/learnings/data-platform-agent-retro-2026-03.md with patterns, pain points, and proposed fixes.
  4. Feed into harnesses — Use retro findings to prioritize docs, rules, CI, and skills.

Part 2: Doc Harnesses

2.1 Playbook Additions

DocPurpose
standards/03-knowledge/engineering/setup/data-platform-quickstart.mdSingle entry point: Snowflake + Rill + dbt + Cursor Cloud Agent. Links to existing setup docs.
standards/04-prompts/tickets/data-platform-ticket-agent.mdPrompt for creating Linear tickets for data work (DE, dbt, Rill, ingest).
standards/04-prompts/data/dbt-pr-impact-prompt.mdPrompt for “run dbt on changed models and summarize impact” (aligns with PR #236 dbt PR impact worker).
standards/04-prompts/data/rill-deployment-prompt.mdPrompt for “deploy Rill project” or “validate Rill sources” — checklist from rill-setup + governance.

2.2 Vault Learnings

  • knowledge/engineering/data-platform/learnings/ — New folder for data-specific learnings.
  • data-platform-agent-retro-2026-03.md — Output of Part 1 retro.
  • RUN_LOG — Extend GTM pattern: add knowledge/engineering/data-platform/RUN_LOG.md for data agent runs (dbt, Rill, ETL).

Part 3: Cursor Agent Harnesses

3.1 Rules

RulePurpose
.cursor/rules/data-platform-setup.mdcWhen working in rill/, knowledge/engineering/data-platform/, or dbt: check playbook setup first; use 1Password for Snowflake; never commit secrets.
.cursor/rules/data-platform-linear.mdcWhen creating data tickets: use Data Platform team; assign to “Delivery & Finance Analytics — Phase 1 Foundation” project.

3.2 Commands

CommandPurpose
.cursor/commands/data-platform-dbt-run.mdRun dbt run and dbt test on changed models (or full project); output summary.
.cursor/commands/data-platform-rill-validate.mdValidate Rill sources (syntax, connector config); optionally run rill start and check health.
.cursor/commands/data-platform-retro-log.mdAdd a row to data platform RUN_LOG from current session context.

3.3 AGENTS.md Update

Add a Data Platform section to AGENTS.md:

  • Working directory for Rill: rill/; for dbt: knowledge/engineering/data-platform/dbt/
  • Env: SNOWFLAKE_*, SNOWFLAKE_CLOUD_AGENT_* (see playbook)
  • Pre-push: platform build only when apps/platform changes; data changes skip platform build
  • Reference: playbook data setup, data-platform rules

Part 4: CI Harnesses

4.1 New Workflow: Data Platform CI

File: .github/workflows/data-platform-ci.yml

  • Triggers: PRs touching rill/**, knowledge/engineering/data-platform/**, standards/03-knowledge/engineering/setup/**
  • Jobs: rill-validate (rill compile), dbt-compile (dbt compile)
  • Scope: Start with compile-only (no Snowflake credentials in CI initially). Add dbt run/dbt test when a read-only Snowflake CI role exists.

4.2 Path Triggers

  • rill/ — Rill validation
  • knowledge/engineering/data-platform/dbt/ — dbt compile (and later run/test)
  • standards/ — Optional: link checker or lint; low priority

Part 5: Skills

5.1 Data Platform Skill

File: .cursor/skills/data-platform/SKILL.md

  • When to use: Rill, dbt, Snowflake, data platform setup, delivery/finance analytics
  • Key paths: rill/, knowledge/engineering/data-platform/, playbook setup docs
  • Env: 1Password for Snowflake; SNOWFLAKE_CLOUD_AGENT_* for Cloud Agents
  • Workflow: setup → compile/validate → run → deploy (Rill Cloud)
  • References: rill-setup, snowflake/cursor-cloud-agent-setup, dbt brainforge-internal-data-platform-setup

5.2 Optional: Linear + Data Skill

Extend meeting-prep or create a thin wrapper: when user asks “create data ticket” or “log data work”, route to Data Platform team and project.


Part 6: Platform Harnesses (Light Touch)

Platform (apps/platform) is Supabase-centric. For scaling data work, minimal additions:

ChangePurpose
Link to Rill CloudAdd a nav item or dashboard widget linking to brainforge Rill org (delivery-analytics, finance-analytics) when deployed.
Deck storagePR #236 mentions analyst deck flow needs persistent storage. Defer to Phase 2 of agent-powered data; document as open item.

No heavy platform integration in this phase.


Part 7: Integration with PR #236

Merge or align PR #236 with this plan:

  • Canonical docs — Strategy and research for this initiative live under knowledge/plans/agent-powered-data-environment/ (same content scope as PR #236’s vault additions; prefer this path for links going forward).
  • Workers — dbt PR impact worker, Rill deployment worker: implement as Cursor commands + prompts first; automate later.
  • Feedback loop — Use RUN_LOG.md for data runs; periodic pattern review per FEEDBACK_LOOP_PROCESS.md.
  • CI — PR #236 calls for “dbt PR impact pipeline”; Part 4 CI is the first step.

Execution Order

  1. Retro first — Informs prioritization of docs, rules, CI.
  2. Docs — Quickstart, prompts. Enables agents to self-serve.
  3. Cursor — Rules and commands. Highest leverage for daily use.
  4. CI — Compile/validate gates. Catches errors before merge.
  5. Skills — Bundles context for “data platform” asks.
  6. Platform — Link to Rill; defer deck storage.

Deliverables Summary

DeliverableLocation
Strategy bundle (PR #236)knowledge/plans/agent-powered-data-environment/
Retro docknowledge/engineering/data-platform/learnings/data-platform-agent-retro-2026-03.md
Data RUN_LOGknowledge/engineering/data-platform/RUN_LOG.md
Quickstartstandards/03-knowledge/engineering/setup/data-platform-quickstart.md
Promptsstandards/04-prompts/data/ (dbt, Rill, ticket)
Rules.cursor/rules/data-platform-*.mdc
Commands.cursor/commands/data-platform-*.md
CI.github/workflows/data-platform-ci.yml
Skill.cursor/skills/data-platform/SKILL.md
AGENTS.mdData Platform section