Ingest

dbt-wiki — Ingest Workflow (v1.0)

User-supplied context capture. Mirrors /repo-wiki:ingest context mode for the dbt-internal scope. Notes attach to specific model / source / macro pages and survive /dbt-wiki:rescan runs (preserved as user-owned ## User Notes body section per SCHEMA.md).

dbt-wiki ingest is context-only in v1.0. It does NOT have:

• git mode (manifest.json snapshot supersedes git for dbt structure;

use /dbt-wiki:rescan after dbt parse instead)

• doc-import mode (v2 backlog — for now, paste the doc summary as

context manually, or write to a docs/ file and reference it)

Pre-condition Check

test -d .dbt-wiki || { echo "Knowledge base not initialized. Run /dbt-wiki:init first."; exit 1; }
test -f .dbt-wiki/index.md || { echo "Missing .dbt-wiki/index.md — re-run /dbt-wiki:init."; exit 1; }

If user invocation has no argument:

> Usage: /dbt-wiki:ingest "" > > Example: /dbt-wiki:ingest "fct_orders sort_key is (order_date, customer_id) because Tableau extract joins on these — see incident #4521" > > The note will be attached to the relevant model / source / macro > page based on names mentioned in your text. If multiple matches > or no match is detected, you'll be asked to clarify.

Exit cleanly if no arg.

Step 1: Identify Target Resource(s)

Parse the user's argument text for resource names. Match against existing pages under:

• .dbt-wiki/_evidence/models/*.md — model names
• .dbt-wiki/_evidence/sources/*.md — source names (in form __)
• .dbt-wiki/_evidence/macros/*.md — macro names
• .dbt-wiki/_evidence/seeds/*.md, _evidence/snapshots/*.md, _evidence/exposures/*.md

Matching strategy:

1. Exact name match: tokens in the argument that exactly match a

resource name (case-sensitive, since dbt is case-sensitive)

1. Tier-prefix match: if user mentions a tier ("marts_finance 整層"),

collect all models under that path prefix

1. No match → ask:

> Couldn't identify a target resource in your note. Available models: > > > Which model / source / macro should this note attach to?

1. Multiple matches (when user mentions 2+ resources, or tier-prefix

matches multiple):

> Your note mentions . Attach to: > [a] all of them (same note copied to each) > [b] just — pick one > [c] cancel

Step 2: Compose Note Entry

For each target resource:

###  

Slug rule: take first 6-8 words of the user's text, kebab-case, lowercase. Skip dbt resource names already in heading. Examples:

| User text | Slug | |---|---| | "fctorders sortkey is (orderdate, customerid) because..." | sort-key-rationale | | "martsfinance permissions need analyticsreadonlygroup..." | permissions-readonly-group | | "stgcustomers has GDPR redaction merge — see incident #4521" | gdpr-redaction-merge |

If slug collides with an existing entry on the same date, append -2, -3, ...

Step 3: Append to Target Page

Read the target page. Locate the ## User Notes section:

• If exists: append the new entry below the last existing entry
• If missing: insert ## User Notes after ## Tests section (per SCHEMA.md

standard ordering); then add the entry

NEVER:

• Modify any other section (Description / SQL Preview / Column Sources /

Tests / Inline Comments / Cross-references / etc.)

• Remove existing User Notes entries
• Re-order existing entries
• Change the section heading text

Step 4: Append Log Entry

## [] ingest | attached note to  page(s)
- Targets: [, ...]
- Slug: 
- Note size:  chars

Step 5: Summary

✓ Note attached to  page(s):
  - .dbt-wiki/_evidence/models/fct_orders.md  (### 2026-05-02 sort-key-rationale)

  To verify: /dbt-wiki:query "fct_orders sort key 為什麼這樣設？"
  To add more: /dbt-wiki:ingest ""

Cross-skill Integration

If .repo-wiki/ exists in the same repo, suggest the user consider whether the note belongs in dbt-wiki (dbt-internal: model behavior, column quirks, materialization) or in repo-wiki (cross-cutting WHY: business decisions, project-level history):

> Note attached to dbt-wiki. If this context is relevant beyond dbt > (e.g., business-level decision touching multiple subsystems), consider > also: /repo-wiki:ingest "" to capture at repo level.

This is informational only — do NOT auto-cross-post.

Rules

NEVER:

• Modify any file outside .dbt-wiki/.md (specifically: never

touch dbt/, target/, or other plugins' directories)

• Rewrite or paraphrase the user's note content (light whitespace

normalization OK; semantic editing NOT OK)

• Auto-attach without confirming target on multi-match or no-match
• Skip the ## User Notes section heading (rescan recognizes it as

user-owned by name)

• Use [[wikilinks]] — only standard markdown links
• Connect to dbt Cloud / warehouse / external API

ALWAYS:

• Verify .dbt-wiki/ exists before any work
• Match resource names case-sensitively (dbt convention)
• Append, never replace, existing User Notes entries
• Date-prefix every new entry (### YYYY-MM-DD )
• Preserve user's original wording and language (zh / ja / en mixed OK)
• Append a log entry per ingest invocation
• Suggest /repo-wiki:ingest cross-link when note has cross-cutting scope

(informational; never auto-post)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: kouko
• Source: kouko/monkey-skills
• License: MIT

Install and usage instructions live in the source repository linked above.