Routes dbt + warehouse tasks from Codex to altimate-code, a specialized open-source CLI agent with 100+ deterministic data-engineering tools: SQL anti-pattern detection, column-level lineage, dbt build/test/run against real warehouses, FinOps and cost analysis, cross-dialect translation and data parity, and PII detection.

Why delegate? General coding agents can edit SQL files; they can't inspect your live warehouse state. altimate-code gives Codex a deterministic data-engineering layer — indexed schemas, real query history, real cost data — instead of guesses.

# 1. Add the marketplace to your Codex
codex plugin marketplace add AltimateAI/altimate-codex-plugin

# 2. Open Codex, type /plugins, select "Altimate Code", install
codex

Adding the marketplace makes the plugin discoverable in the /plugins browser but does not install it — you install it explicitly from the browser, and Codex will prompt you to review and trust its SessionStart hook before it runs.

The altimate-code CLI must be installed and configured on your PATH:

npm install -g altimate-code     # requires Node 20+
altimate-code                    # launch the TUI, then run /connect to configure your LLM provider

Or configure a provider via environment variable (ANTHROPIC_API_KEY, OPENAI_API_KEY, and 20+ others — see the altimate-code docs).

If altimate-code is not on PATH or not configured, the plugin surfaces the failure with fix instructions rather than silently falling back to Codex's native tools.

1. A SessionStart hook adds a static description of the plugin's two skills to the session, so Codex knows when a task fits. Codex prompts you to review and trust the hook the first time you start a session with the plugin installed; approve it once and it fires in every subsequent session.
2. Delegation — when a data-engineering task fires the altimate-code skill (or you invoke /altimate-code:altimate <task>), Codex runs altimate-code run "<your task>" locally, selecting the cheapest suitable agent persona (fast-edit for dbt/SQL edits, analyst for aggregation-heavy correctness work, builder for warehouse-state investigation).
3. Review — when the user asks to review dbt changes / PR / commit, the review skill (or /altimate-code:review) runs altimate-code review on the current dbt project and surfaces the signed verdict envelope verbatim (APPROVE / COMMENT / REQUEST_CHANGES).
4. altimate-code plans and executes each task using your configured LLM provider. Warehouse queries, dbt commands, and file edits run locally under your credentials.
5. The result is written to a private local temp file (delegation) or streamed to stdout (review) and surfaced back in your Codex conversation.

Once installed, try these in Codex — the skills route them to altimate-code automatically. You can also invoke a skill explicitly with /altimate-code:altimate <task> or /altimate-code:review [scope].

1. Column-level lineage — "Trace the revenue_usd column in models/marts/fct_revenue.sql back to its sources, and list every downstream model that would be affected if I change its rounding."
2. Warehouse cost analysis — "Give me a cost report for our Snowflake account: the ten most expensive queries this month and any warehouses that look over-provisioned."
3. Cross-dialect validation — "Compare prod.orders in Snowflake with analytics.orders in BigQuery row-by-row using id as the key, and summarize any mismatches."
4. dbt test generation — "Generate dbt tests for models/staging/stg_orders.sql, including unit tests for the CASE/WHEN logic."
5. dbt PR review — "Review my dbt changes." or "Review PR #1158." → the review skill produces a signed verdict envelope.

Delegating a task passes your prompt text to the altimate-code CLI, which sends it to your configured LLM provider — the Altimate LLM Gateway by default, or your own provider (BYOK), in which case Altimate never sees it. Warehouse credentials and query results stay on your machine.

Full details: PRIVACY.md · Security policy and disclosure: SECURITY.md · Terms: TERMS.md

Still stuck? See Support.

• Skill — skills/altimate-code/SKILL.md — delegation logic and agent-persona selection (byte-identical routing behavior with the Claude Code and opencode variants of this plugin, per the agentskills.io open standard)
• Skill — skills/review/SKILL.md — dbt/SQL PR reviewer that runs altimate-code review and surfaces signed verdicts; declines non-dbt code review
• Command — commands/altimate.md — /altimate-code:altimate <task> explicit-delegation slash command with agent-persona selection
• Command — commands/review.md — /altimate-code:review [scope] explicit-review slash command; scope can be empty (working tree), a commit SHA, a branch name, or a GitHub PR URL / number
• Hook — hooks/hooks.json + hooks/altimate-code-session-start.sh — SessionStart capability description (static text only; no side effects)

• Docs: docs.altimate.sh
• Issues: GitHub Issues
• Community: Slack
• Email: support@altimate.ai · security reports: security@altimate.ai

MIT — see LICENSE.