Conversation
|
Caution Review failedPull request was closed or merged during review 📝 WalkthroughWalkthroughAdded comprehensive documentation and Claude agent specs for a 10-step dataset onboarding workflow: interactive context gathering, Drive architecture table fetch/create, cleaning to partitioned parquet, BigQuery uploads (dev → prod with approval), DBT model/test generation & runs, metadata registration, and PR creation. All backend calls must use the MCP server. Changes
Sequence Diagram(s)mermaid User->>Agent: start onboarding (slug, path) Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 inconclusive)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 11
🧹 Nitpick comments (15)
.claude/commands/databasis-metadata.md (6)
142-142: Make the required field list explicit.Line 142 states "All other fields must be re-passed as well (the API requires them)" but doesn't specify which fields. An AI agent should explicitly re-pass the fields from Step 1:
- slug, name_pt/en/es, description_pt/en/es - dataset_id, status_id, idAdding this list would eliminate uncertainty about which fields constitute "all other fields."
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-metadata.md at line 142, Update the documentation sentence that reads "All other fields must be re-passed as well (the API requires them)" in the databasis-metadata instructions (the Step 1 / deferred call paragraph) to explicitly list the required fields to re-pass: slug, name_pt, name_en, name_es, description_pt, description_en, description_es, dataset_id, status_id, and id; place this explicit bullet-style list or inline comma-separated list immediately after the sentence so agents know exactly which fields are required when performing the deferred relationship write.
148-148: Add language identifier to output format code block.The fenced code block at Line 148 should specify a language identifier for better rendering and accessibility:
-``` +```text === METADATA REGISTRATION COMPLETE (env=<env>) ===As per static analysis hint MD040 (fenced-code-language).
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-metadata.md at line 148, The fenced code block containing the line "=== METADATA REGISTRATION COMPLETE (env=<env>) ===" needs a language identifier; replace the opening triple backticks for that block with a language-tagged fence (e.g., ```text) so the block becomes ```text ... ``` to satisfy MD040 and improve rendering/accessibility.
135-138: Clarify raw data source selection logic assumptions.The conditional logic assumes:
- Exactly two raw data sources exist (pre-2013 and post-2013)
- All datasets follow this 2013 split pattern
An AI agent might encounter datasets with different source structures:
- Only one raw data source spanning all years
- Multiple sources not delineated by the 2013 boundary
- Sources split by topic rather than time
Consider adding:
- A note about when this logic applies vs. when to ask for user clarification
- Guidance for datasets that don't fit the pre/post-2013 pattern
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-metadata.md around lines 135 - 138, The raw_data_source_ids selection logic (using get_raw_data_sources and start_year) currently assumes a rigid pre-2013/post-2013 split and exactly two sources; update the documentation to make that assumption explicit and add fallback guidance: state that the 2013 rule applies only when get_raw_data_sources returns distinct pre- and post-2013 sources, otherwise (a) if there is a single source spanning all years, select that source; (b) if there are multiple sources not aligned to 2013, prefer the source(s) whose temporal coverage overlaps the table's temporal_coverage and include a note to consult raw source descriptions; and (c) if topic-based splits exist or ambiguity remains, instruct the agent to ask the user for clarification rather than guessing.
158-158: Consider consistent command invocation format.Line 158 shows
/databasis-metadata <slug> --env prodwith a leading slash, suggesting a slash command invocation. However, the document title and earlier references usedatabasis-metadatawithout the slash.If commands are invoked as slash commands in the AI agent interface, consider:
- Using the slash prefix consistently throughout all three command files
- Or removing the slash if commands are invoked without it
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-metadata.md at line 158, The document shows an inconsistent command invocation: a leading slash is used in the example "/databasis-metadata <slug> --env prod" while the title and other references use "databasis-metadata" without the slash; update the examples in all three command files so the invocation format is consistent—either prefix every occurrence of the command with a slash (e.g., "/databasis-metadata <slug> --env prod") if these are true slash commands, or remove the slash everywhere (e.g., "databasis-metadata <slug> --env prod") if not; search for occurrences of databasis-metadata across the three command files and make the same change to each to ensure uniformity.
30-31: Clarify handling of source documentation format.Line 30 recommends "direct copy-paste from raw data source documentation when it captures the content well", but Line 31 specifies "No bullet lists." If source documentation contains bullet lists, an AI agent might be uncertain whether to:
- Copy the bullet content but convert to prose sentences, or
- Avoid copy-pasting entirely in that case
Adding "convert bullet lists to prose" would resolve this.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-metadata.md around lines 30 - 31, The guidance is ambiguous between "direct copy-paste from raw data source documentation" and "No bullet lists"; update the rule text so it explicitly says that when the source contains bullet lists you may copy the content but must convert bullets into 1–3 sentence prose per entity. Locate the phrases "direct copy-paste from raw data source documentation when it captures the content well" and "No bullet lists." and augment them with a short clarifying clause such as "If the source uses bullet lists, transform those bullets into concise prose sentences rather than preserving list formatting."
56-56: Consider moving detailed backend issue notes to a separate troubleshooting document.The known issue note provides valuable context about M2M field behavior, but it includes implementation details (specific Python code, file paths, dates) that may become stale as the backend evolves.
Consider either:
- Moving this to a separate "Known Issues" or "Troubleshooting" document referenced from here
- Keeping a brief note here ("M2M fields require deferred update — see step 8") and moving details elsewhere
This would keep command specs focused on "what to do" rather than "why the backend behaves this way."
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-metadata.md at line 56, The documentation includes a detailed backend troubleshooting note about M2M fields (raw_data_source_ids, published_by_ids, data_cleaned_by_ids) and even references implementation details in perform_mutate (backend/custom/graphql_auto.py) and a date; move the long technical explanation out of .claude/commands/databasis-metadata.md into a separate "Known Issues" or "Troubleshooting" document and replace it with a short actionable note (e.g., "M2M fields require deferred update — see Troubleshooting doc") that links to the new doc so the command spec focuses on what to do not implementation details..claude/commands/databasis-architecture.md (5)
41-41: Define "Brasil-level" terminology.The term "Brasil-level" is used to distinguish federal/national data from state/municipality data, but it's not explicitly defined. An AI agent might misinterpret this.
Consider adding a brief parenthetical definition: "Brasil-level (federal/national level, not disaggregated by state or municipality)"
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-architecture.md at line 41, Update the wording where "Brasil-level" appears in the Partition columns line (the line containing "2. **Partition columns:** ... Brasil-level: `ano` only") to include a brief parenthetical definition clarifying it means federal/national data not disaggregated by state or municipality (e.g., add "(Brasil-level = federal/national level, not disaggregated by state or municipality)") so any agent or reader cannot misinterpret the term.
68-70: Consider specifying a structured output format.The output summary is described informally ("listing all tables found/created and the Drive URLs"), which could lead to inconsistent formatting by an AI agent. Since downstream commands need these URLs, a structured format would improve reliability:
=== ARCHITECTURE TABLES === Found: <N> | Created: <N> Tables: ✓ <table_slug> — <drive_url> ✓ <table_slug> — <drive_url>This makes parsing more reliable for subsequent commands.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-architecture.md around lines 68 - 70, Update "Step 6 — Output" to require a deterministic, machine-parsable summary for downstream use (databasis-metadata). Replace the informal sentence with a specified structured format that includes a clear header (e.g., "=== ARCHITECTURE TABLES ==="), counts ("Found: <N> | Created: <N>"), and a line-per-table list of " <table_slug> — <drive_url>" so callers can reliably parse table slugs and Drive URLs; explicitly state that the Drive URLs must be stored for databasis-metadata consumption.
64-64: Clarify file format for architecture tables.The path specifies
.xlsx(Excel format), but Line 66 referencescreate_spreadsheetwhich creates native Google Sheets. An AI agent might be confused about which format to use.Consider either:
- Removing the
.xlsxextension if using native Google Sheets format- Clarifying that the sheet should be created in Google Sheets but exported as
.xlsx- Or specifying just
<table_slug>without extension🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-architecture.md at line 64, The file path example uses an .xlsx extension but the code/workflow uses create_spreadsheet (Google Sheets); update the spec to avoid ambiguity by either removing the .xlsx extension from the path string (`BD/Dados/Conjuntos/<dataset>/architecture/<table_slug>`), stating explicitly that create_spreadsheet should create a native Google Sheet which may be exported as .xlsx if needed, or documenting that `<table_slug>.xlsx` refers to an exported Google Sheets file; modify the documentation where the path is defined and any references around create_spreadsheet to reflect the chosen option so agents know whether to create a native Google Sheet or an exported Excel file.
21-21: Consider documenting edge cases for the directory_column FK format.The format
<dataset>.<table>:<column>is clear for simple cases, but an AI agent might encounter:
- Columns that could reference multiple directory tables
- Composite foreign keys
- Cases where the reference column name differs from the local column name
Adding a sentence about these scenarios would reduce ambiguity.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-architecture.md at line 21, Update the `directory_column` documentation line to explicitly note edge cases for the `<dataset>.<table>:<column>` FK format: mention how to handle columns that may reference multiple directory tables, how composite foreign keys should be represented, and what to do when the referenced column name differs from the local column name (e.g., indicate multiple entries or a clear tuple/notation and mapping convention). Reference the `directory_column` field and the `<dataset>.<table>:<column>` format when adding this single clarifying sentence so readers/agents know how to encode these special cases.
27-27: Clarify interval notation for non-annual data.The temporal coverage notation only shows annual examples with
(1). For an AI agent processing monthly or quarterly data, the interval semantic is ambiguous:
- Does
(12)mean monthly (12 observations per year)?- Does
(3)mean quarterly (every 3 months)?- How to express irregular intervals?
Adding examples like
2020(12)2022for monthly data would eliminate this ambiguity.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-architecture.md at line 27, Clarify the interval semantics in the "Temporal coverage notation: START(INTERVAL)END" line: state that the number inside parentheses is observations per year (e.g., (1)=annual, (12)=monthly, (4)=quarterly) and add explicit examples such as `2020(12)2022` for monthly and `2019(4)2021` for quarterly; also specify how to denote ongoing ranges (keep `2013(1)` meaning from 2013 onward) and provide a convention for irregular data (e.g., use `IRR` or leave interval empty with a note) so agents can unambiguously parse `START(INTERVAL)END`..claude/commands/databasis-clean.md (4)
48-49: Clarify column order vs. column presence for subset validation.Line 48 instructs starting with a small subset, but Line 49 requires "output column order must match architecture exactly." An AI agent might be uncertain whether:
- All architecture columns must be present in the subset (even as null columns), or
- Only columns present in the subset data need to match the architecture order
Adding "Output columns present in the subset must match architecture order" would eliminate ambiguity.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-clean.md around lines 48 - 49, Clarify that the "output column order must match architecture exactly" requirement applies only to the columns present in the chosen subset by updating the second bullet (the line that currently reads "Output column order must match architecture exactly") to say something like "Output columns present in the subset must match architecture order exactly (missing architecture columns may be omitted or null)"; reference the two bullets "Start with a small subset (1 year or the smallest available partition)" and the "Output column order..." sentence so readers know the subset validation rule is about ordering of present columns, not forcing all architecture columns to appear.
74-74: Consider adding an example row for the dictionary schema.The dictionary schema uses Portuguese column names (
id_tabela,nome_coluna,chave,valor) while architecture tables use English (name,bigquery_type, etc.). An AI agent might also be uncertain how to populatecobertura_temporalfor dictionary entries.Adding a brief example would clarify:
id_tabela | nome_coluna | chave | cobertura_temporal | valor despesa | estagio | 1 | 2013(1)2022 | Empenhado despesa | estagio | 2 | 2013(1)2022 | Liquidado🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-clean.md at line 74, Add a concrete example row block for the dictionary schema to clarify Portuguese column usage and how to populate cobertura_temporal; update the section that currently shows "Schema: `id_tabela | nome_coluna | chave | cobertura_temporal | valor`" to include sample rows (e.g., for the table/column combination and temporal coverage) so agents know how to fill id_tabela, nome_coluna, chave, cobertura_temporal and valor consistently.
45-59: Provide polars examples or recommend pandas by default.Line 45 instructs the AI agent to "choose whichever fits the data better (polars for large files or complex transformations; pandas otherwise)", but Lines 56-59 provide only pandas-specific type casting patterns.
If an agent selects polars, it has no reference implementation for safe type casting. Consider either:
- Recommending pandas as the default and using polars only for specific large-file scenarios
- Adding equivalent polars type casting examples alongside pandas
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-clean.md around lines 45 - 59, The guidance currently lists pandas-specific safe casting examples (INT64, FLOAT64, STRING, DATE) but also tells agents they can choose polars, leaving polars users without equivalents; update the doc so either pandas is recommended as the default choice (and note polars only for large/complex cases) or add matching polars safe_cast patterns alongside the pandas ones. Specifically, add polars equivalents for the safe_cast logic and the standard column types (INT64, FLOAT64, STRING, DATE) and mention pivot-to-long and pyarrow partitioning applicability in polars, or explicitly state "prefer pandas by default" near the existing examples and mark polars as an advanced option. Ensure the terms safe_cast, INT64, FLOAT64, STRING, DATE, pivot-to-long, and pyarrow are referenced so readers can find and apply the fix.
79-81: Consider specifying output format for validation and handoff.Step 6 instructs to "report final row counts per partition" but doesn't specify the format. For reliable handoff to downstream commands (e.g.,
databasis-upload), a structured summary would help:=== CLEANING COMPLETE === Dataset: <dataset_slug> Tables: ✓ <table_slug> — <total_rows> rows across <N> partitions └─ ano=2020: <rows>, ano=2021: <rows>, ... Output path: <output_path>🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/databasis-clean.md around lines 79 - 81, Update the "Step 6 — Scale to full data" instructions to require emitting a structured, machine- and human-friendly summary (for downstream validation/hand-off to commands like databasis-upload) by adding a concrete output template under the step (include Dataset slug, a Tables section with per-table total rows and per-partition row counts, and the Output path). Reference the existing Step 6 heading ("Step 6 — Scale to full data") and ensure the documentation explicitly shows the exact fields required (Dataset, Tables → <table_slug> total rows and partition breakdown, Output path) and a short marker like "=== CLEANING COMPLETE ===" so downstream tooling can reliably parse it.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.claude/agents/onboarding-pipeline.md:
- Around line 18-20: Add language identifiers to the unlabeled fenced code
blocks so they pass MD040: change the opening backticks for the example,
sequence, and checklist fences to include a language token (e.g., ```text or
```md) instead of plain ```; locate the three fenced blocks named "example",
"sequence" and "checklist" in the onboarding-pipeline content and update their
opening fences to ```text (and do the same for the other similar
example/sequence/checklist fenced blocks in this file).
In @.claude/commands/databasis-context.md:
- Around line 37-47: Update the fenced code block that begins with "=== DATASET
CONTEXT: <slug> ===" so the opening fence includes a language specifier; replace
the existing "```" with "```text" to satisfy the MD040 rule and mark the block
as plain text. Ensure only the opening fence is changed and the closing "```"
remains the same so the block renders correctly.
In @.claude/commands/databasis-dbt-run.md:
- Around line 15-18: Replace the hardcoded pyenv path used to create the
virtualenv with a generic python invocation: change the command using
"~/.pyenv/versions/3.11.6/bin/python -m venv /tmp/dbt_env" to use "python -m
venv /tmp/dbt_env" so the venv honors the repo's Python constraint; similarly
update any subsequent references that assume that specific pyenv path (e.g.,
"/tmp/dbt_env/bin/pip install dbt-bigquery") to remain the same but ensure the
initial venv creation uses the generic python executable.
- Around line 26-27: The documented dbt run/test command uses --profiles-dir
~/.dbt which doesn't match this repo's profiles.yml location; update the command
examples in .claude/commands/databasis-dbt-run.md to point --profiles-dir to the
repository root (or remove the flag so dbt picks up ./profiles.yml) for both
occurrences (the block containing "--profiles-dir ~/.dbt \ --target <target>"
and the similar lines at 37-38), ensuring the --target flag remains unchanged.
In @.claude/commands/databasis-dbt.md:
- Line 2: Replace all occurrences of the alternative schema filename (e.g.,
"schema.yaml") with the canonical "schema.yml" throughout the document and any
template strings so the generator and lookup logic are consistent; specifically
update the description line and the later reference that currently uses
"schema.yaml"/"schema.yml" mismatch and ensure any examples, templates or
variables that emit the filename use "schema.yml" only.
In @.claude/commands/databasis-discover.md:
- Around line 16-18: Label the unlabeled fenced code blocks that contain the
snippet "discover_ids(env=<env>)" (and the other block around lines 41-74) with
an appropriate language tag (e.g., use ```text for plain output or ```bash for
command-like snippets) so the markdown linter MD040 is satisfied; update the two
fenced blocks containing the shown snippet and the longer block later to start
with a language identifier instead of a bare ```.
In @.claude/commands/databasis-pr.md:
- Around line 38-55: The fenced code block that begins with the "## Dataset"
header is missing a language identifier and triggers MD040; update the opening
fence to include a markdown language tag (e.g., use ```md or ```markdown) so the
snippet is properly labeled and linting passes, leaving the rest of the block
content unchanged.
- Around line 31-35: Update the gh pr create invocation to open the PR as a
draft, use the required title pattern starting with a bracketed keyword (e.g.,
"[Data] {dataset_slug} {table names}"), and build the PR body from the
repository template: generate /tmp/pr_body.md by loading
.github/pull_request_template.md and inserting the dataset changelog into the
"Detalhes Técnicos" section (keep Descrição do PR, Teste e Validações, Riscos e
Mitigações, Dependências intact), then call gh pr create --title
"<bracketed-keyword> ${dataset_slug} <table names>" --body-file /tmp/pr_body.md
--label "test-dev,table-approve,metadata-test" --draft so the template structure
and draft status are preserved.
In @.claude/commands/databasis-upload.md:
- Around line 15-17: The fenced code block containing the message "Missing
~/.basedosdados/config.toml — run `basedosdados config init` first." is
unlabeled and triggers MD040; update the opening backticks to include a language
such as text or bash (e.g., ```text or ```bash) for the prerequisites snippet so
the fenced block is explicitly labeled.
- Around line 37-41: The monkey-patch references BILLING_PROJECT but it is not
defined, causing a NameError at runtime; define and validate BILLING_PROJECT
before assigning gcs.Client.bucket so _patched_bucket can reference it. Add a
top-level definition for BILLING_PROJECT (e.g., read from configuration or
os.environ and fail fast with a clear error if missing), then apply the
monkey-patch that replaces gcs.Client.bucket with _patched_bucket; ensure
references to _patched_bucket and gcs.Client.bucket occur after BILLING_PROJECT
is set.
In `@AGENTS.md`:
- Around line 242-244: The fenced code block containing "Onboard dataset <slug>.
Raw files at <path>. Drive folder: BD/Dados/Conjuntos/<slug>/." is missing a
language label; update the opening ``` to include a language token (e.g.,
```text or ```md) so the block passes MD040 and is properly labeled in
AGENTS.md.
---
Nitpick comments:
In @.claude/commands/databasis-architecture.md:
- Line 41: Update the wording where "Brasil-level" appears in the Partition
columns line (the line containing "2. **Partition columns:** ... Brasil-level:
`ano` only") to include a brief parenthetical definition clarifying it means
federal/national data not disaggregated by state or municipality (e.g., add
"(Brasil-level = federal/national level, not disaggregated by state or
municipality)") so any agent or reader cannot misinterpret the term.
- Around line 68-70: Update "Step 6 — Output" to require a deterministic,
machine-parsable summary for downstream use (databasis-metadata). Replace the
informal sentence with a specified structured format that includes a clear
header (e.g., "=== ARCHITECTURE TABLES ==="), counts ("Found: <N> | Created:
<N>"), and a line-per-table list of " <table_slug> — <drive_url>" so callers can
reliably parse table slugs and Drive URLs; explicitly state that the Drive URLs
must be stored for databasis-metadata consumption.
- Line 64: The file path example uses an .xlsx extension but the code/workflow
uses create_spreadsheet (Google Sheets); update the spec to avoid ambiguity by
either removing the .xlsx extension from the path string
(`BD/Dados/Conjuntos/<dataset>/architecture/<table_slug>`), stating explicitly
that create_spreadsheet should create a native Google Sheet which may be
exported as .xlsx if needed, or documenting that `<table_slug>.xlsx` refers to
an exported Google Sheets file; modify the documentation where the path is
defined and any references around create_spreadsheet to reflect the chosen
option so agents know whether to create a native Google Sheet or an exported
Excel file.
- Line 21: Update the `directory_column` documentation line to explicitly note
edge cases for the `<dataset>.<table>:<column>` FK format: mention how to handle
columns that may reference multiple directory tables, how composite foreign keys
should be represented, and what to do when the referenced column name differs
from the local column name (e.g., indicate multiple entries or a clear
tuple/notation and mapping convention). Reference the `directory_column` field
and the `<dataset>.<table>:<column>` format when adding this single clarifying
sentence so readers/agents know how to encode these special cases.
- Line 27: Clarify the interval semantics in the "Temporal coverage notation:
START(INTERVAL)END" line: state that the number inside parentheses is
observations per year (e.g., (1)=annual, (12)=monthly, (4)=quarterly) and add
explicit examples such as `2020(12)2022` for monthly and `2019(4)2021` for
quarterly; also specify how to denote ongoing ranges (keep `2013(1)` meaning
from 2013 onward) and provide a convention for irregular data (e.g., use `IRR`
or leave interval empty with a note) so agents can unambiguously parse
`START(INTERVAL)END`.
In @.claude/commands/databasis-clean.md:
- Around line 48-49: Clarify that the "output column order must match
architecture exactly" requirement applies only to the columns present in the
chosen subset by updating the second bullet (the line that currently reads
"Output column order must match architecture exactly") to say something like
"Output columns present in the subset must match architecture order exactly
(missing architecture columns may be omitted or null)"; reference the two
bullets "Start with a small subset (1 year or the smallest available partition)"
and the "Output column order..." sentence so readers know the subset validation
rule is about ordering of present columns, not forcing all architecture columns
to appear.
- Line 74: Add a concrete example row block for the dictionary schema to clarify
Portuguese column usage and how to populate cobertura_temporal; update the
section that currently shows "Schema: `id_tabela | nome_coluna | chave |
cobertura_temporal | valor`" to include sample rows (e.g., for the table/column
combination and temporal coverage) so agents know how to fill id_tabela,
nome_coluna, chave, cobertura_temporal and valor consistently.
- Around line 45-59: The guidance currently lists pandas-specific safe casting
examples (INT64, FLOAT64, STRING, DATE) but also tells agents they can choose
polars, leaving polars users without equivalents; update the doc so either
pandas is recommended as the default choice (and note polars only for
large/complex cases) or add matching polars safe_cast patterns alongside the
pandas ones. Specifically, add polars equivalents for the safe_cast logic and
the standard column types (INT64, FLOAT64, STRING, DATE) and mention
pivot-to-long and pyarrow partitioning applicability in polars, or explicitly
state "prefer pandas by default" near the existing examples and mark polars as
an advanced option. Ensure the terms safe_cast, INT64, FLOAT64, STRING, DATE,
pivot-to-long, and pyarrow are referenced so readers can find and apply the fix.
- Around line 79-81: Update the "Step 6 — Scale to full data" instructions to
require emitting a structured, machine- and human-friendly summary (for
downstream validation/hand-off to commands like databasis-upload) by adding a
concrete output template under the step (include Dataset slug, a Tables section
with per-table total rows and per-partition row counts, and the Output path).
Reference the existing Step 6 heading ("Step 6 — Scale to full data") and ensure
the documentation explicitly shows the exact fields required (Dataset, Tables →
<table_slug> total rows and partition breakdown, Output path) and a short marker
like "=== CLEANING COMPLETE ===" so downstream tooling can reliably parse it.
In @.claude/commands/databasis-metadata.md:
- Line 142: Update the documentation sentence that reads "All other fields must
be re-passed as well (the API requires them)" in the databasis-metadata
instructions (the Step 1 / deferred call paragraph) to explicitly list the
required fields to re-pass: slug, name_pt, name_en, name_es, description_pt,
description_en, description_es, dataset_id, status_id, and id; place this
explicit bullet-style list or inline comma-separated list immediately after the
sentence so agents know exactly which fields are required when performing the
deferred relationship write.
- Line 148: The fenced code block containing the line "=== METADATA REGISTRATION
COMPLETE (env=<env>) ===" needs a language identifier; replace the opening
triple backticks for that block with a language-tagged fence (e.g., ```text) so
the block becomes ```text ... ``` to satisfy MD040 and improve
rendering/accessibility.
- Around line 135-138: The raw_data_source_ids selection logic (using
get_raw_data_sources and start_year) currently assumes a rigid
pre-2013/post-2013 split and exactly two sources; update the documentation to
make that assumption explicit and add fallback guidance: state that the 2013
rule applies only when get_raw_data_sources returns distinct pre- and post-2013
sources, otherwise (a) if there is a single source spanning all years, select
that source; (b) if there are multiple sources not aligned to 2013, prefer the
source(s) whose temporal coverage overlaps the table's temporal_coverage and
include a note to consult raw source descriptions; and (c) if topic-based splits
exist or ambiguity remains, instruct the agent to ask the user for clarification
rather than guessing.
- Line 158: The document shows an inconsistent command invocation: a leading
slash is used in the example "/databasis-metadata <slug> --env prod" while the
title and other references use "databasis-metadata" without the slash; update
the examples in all three command files so the invocation format is
consistent—either prefix every occurrence of the command with a slash (e.g.,
"/databasis-metadata <slug> --env prod") if these are true slash commands, or
remove the slash everywhere (e.g., "databasis-metadata <slug> --env prod") if
not; search for occurrences of databasis-metadata across the three command files
and make the same change to each to ensure uniformity.
- Around line 30-31: The guidance is ambiguous between "direct copy-paste from
raw data source documentation" and "No bullet lists"; update the rule text so it
explicitly says that when the source contains bullet lists you may copy the
content but must convert bullets into 1–3 sentence prose per entity. Locate the
phrases "direct copy-paste from raw data source documentation when it captures
the content well" and "No bullet lists." and augment them with a short
clarifying clause such as "If the source uses bullet lists, transform those
bullets into concise prose sentences rather than preserving list formatting."
- Line 56: The documentation includes a detailed backend troubleshooting note
about M2M fields (raw_data_source_ids, published_by_ids, data_cleaned_by_ids)
and even references implementation details in perform_mutate
(backend/custom/graphql_auto.py) and a date; move the long technical explanation
out of .claude/commands/databasis-metadata.md into a separate "Known Issues" or
"Troubleshooting" document and replace it with a short actionable note (e.g.,
"M2M fields require deferred update — see Troubleshooting doc") that links to
the new doc so the command spec focuses on what to do not implementation
details.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 1f8dc0cf-906d-47ab-b612-4c07517ebfb3
📒 Files selected for processing (11)
.claude/agents/onboarding-pipeline.md.claude/commands/databasis-architecture.md.claude/commands/databasis-clean.md.claude/commands/databasis-context.md.claude/commands/databasis-dbt-run.md.claude/commands/databasis-dbt.md.claude/commands/databasis-discover.md.claude/commands/databasis-metadata.md.claude/commands/databasis-pr.md.claude/commands/databasis-upload.mdAGENTS.md
Comment on lines
+26
to
+27
| --profiles-dir ~/.dbt \ | ||
| --target <target> |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "Repo-level profiles file:"
fd -a '^profiles\.yml$' .
echo
echo "Command currently expects ~/.dbt/profiles.yml:"
if [ -f "$HOME/.dbt/profiles.yml" ]; then
echo "Found ~/.dbt/profiles.yml"
else
echo "Missing ~/.dbt/profiles.yml"
fiRepository: basedosdados/pipelines
Length of output: 199
🏁 Script executed:
# Check the exact location of profiles.yml
ls -la ./profiles.yml
echo "---"
# Read the command file at the specified lines
echo "Lines 20-35 of the command file:"
sed -n '20,35p' ./.claude/commands/databasis-dbt-run.md
echo "---"
echo "Lines 35-45 of the command file:"
sed -n '35,45p' ./.claude/commands/databasis-dbt-run.mdRepository: basedosdados/pipelines
Length of output: 787
--profiles-dir ~/.dbt is misaligned with this repo's profiles.yml location.
The repo provides profiles.yml at the project root, but the command points to ~/.dbt. This will cause runs and tests to fail unless ~/.dbt/profiles.yml is pre-seeded externally.
Suggested update
/tmp/dbt_env/bin/dbt run \
--select <dataset_slug> \
- --profiles-dir ~/.dbt \
+ --profiles-dir <pipelines_root> \
--target <target> /tmp/dbt_env/bin/dbt test \
--select <dataset_slug> \
- --profiles-dir ~/.dbt \
+ --profiles-dir <pipelines_root> \
--target <target>Also applies to: lines 37-38
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.claude/commands/databasis-dbt-run.md around lines 26 - 27, The documented
dbt run/test command uses --profiles-dir ~/.dbt which doesn't match this repo's
profiles.yml location; update the command examples in
.claude/commands/databasis-dbt-run.md to point --profiles-dir to the repository
root (or remove the flag so dbt picks up ./profiles.yml) for both occurrences
(the block containing "--profiles-dir ~/.dbt \ --target <target>" and the
similar lines at 37-38), ensuring the --target flag remains unchanged.
| @@ -0,0 +1,99 @@ | |||
| --- | |||
| description: Write DBT .sql and schema.yaml files for a Data Basis dataset | |||
There was a problem hiding this comment.
Use one schema filename convention (schema.yml) consistently.
Line 2 says schema.yaml, while Line 47 uses schema.yml. Keep one canonical extension to avoid generator drift and missing file lookups. Based on learnings: Every dbt model must have a schema.yml entry inside its dataset directory declaring model metadata and all data quality tests.
Also applies to: 47-48
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.claude/commands/databasis-dbt.md at line 2, Replace all occurrences of the
alternative schema filename (e.g., "schema.yaml") with the canonical
"schema.yml" throughout the document and any template strings so the generator
and lookup logic are consistent; specifically update the description line and
the later reference that currently uses "schema.yaml"/"schema.yml" mismatch and
ensure any examples, templates or variables that emit the filename use
"schema.yml" only.
Comment on lines
+31
to
+35
| gh pr create \ | ||
| --title "[$dataset_slug] <table names>" \ | ||
| --body "<changelog>" \ | ||
| --label "test-dev,table-approve,metadata-test" | ||
| ``` |
There was a problem hiding this comment.
🧩 Analysis chain
🌐 Web query:
Does gh pr create --body "" replace the repository pull request template content instead of auto-populating it?
💡 Result:
Yes, gh pr create --body "" replaces the repository pull request template content instead of auto-populating it. The --body flag supplies the literal body text for the pull request, skipping any interactive prompts or automatic template population. Repository templates are only used as a starting point in interactive mode (when no --body or --title is provided) or explicitly via --template, where they are loaded into the editor. Official documentation states that --body takes precedence over other autofill mechanisms like --fill, overwriting them, and issues confirm --body or --body-file uses the provided content directly without incorporating templates unless manually combined.
Citations:
- 1: https://cli.github.com/manual/gh_pr_create
- 2: https://man.archlinux.org/man/gh-pr-create.1.en
- 3: https://www.gsp.com/cgi-bin/man.cgi?section=1&topic=GH-PR-CREATE
- 4: https://www.mankier.com/1/gh-pr-create
- 5: Allow "gh pr create" to specify a pull request template to use cli/cli#2567
- 6:
gh pr createwith--body-fileshould allow--editor--body-templatecli/cli#6408
🏁 Script executed:
# Check if the PR template exists
find . -name "pull_request_template.md" -o -name "*pull_request*" | head -20Repository: basedosdados/pipelines
Length of output: 100
🏁 Script executed:
# Read the databasis-pr.md file to see the actual content and context
if [ -f ".claude/commands/databasis-pr.md" ]; then
echo "=== File exists, reading content ==="
wc -l ".claude/commands/databasis-pr.md"
echo ""
echo "=== Lines 25-60 ==="
sed -n '25,60p' ".claude/commands/databasis-pr.md"
fiRepository: basedosdados/pipelines
Length of output: 918
🏁 Script executed:
# Check directory structure for .github and related files
ls -la .github/ 2>/dev/null | head -20Repository: basedosdados/pipelines
Length of output: 378
🏁 Script executed:
# Read the PR template to see what is actually required
cat ".github/pull_request_template.md"Repository: basedosdados/pipelines
Length of output: 2945
Add --draft flag and build PR body from template structure; fix title pattern to match repository requirements.
The --body "<changelog>" parameter bypasses the repository's pull request template entirely instead of populating it. Additionally, the title pattern [$dataset_slug] <table names> does not match the required format and the --draft flag is missing.
Per .github/pull_request_template.md, PR titles must start with a keyword in brackets ([Feature], [Data], [Bugfix], etc.), PRs must open as draft, and body must include specific sections: Descrição do PR, Detalhes Técnicos, Teste e Validações, Riscos e Mitigações, and Dependências.
Suggested update
gh pr create \
+ --draft \
- --title "[$dataset_slug] <table names>" \
+ --title "[Data] $dataset_slug" \
- --body "<changelog>" \
+ --body-file /tmp/pr_body.md \
--label "test-dev,table-approve,metadata-test"Build /tmp/pr_body.md to include all sections from .github/pull_request_template.md, embedding the dataset changelog in the "Detalhes Técnicos" section instead of replacing the template structure.
Applies to lines 31–35 and the suggested PR body format in lines 37–55.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.claude/commands/databasis-pr.md around lines 31 - 35, Update the gh pr
create invocation to open the PR as a draft, use the required title pattern
starting with a bracketed keyword (e.g., "[Data] {dataset_slug} {table names}"),
and build the PR body from the repository template: generate /tmp/pr_body.md by
loading .github/pull_request_template.md and inserting the dataset changelog
into the "Detalhes Técnicos" section (keep Descrição do PR, Teste e Validações,
Riscos e Mitigações, Dependências intact), then call gh pr create --title
"<bracketed-keyword> ${dataset_slug} <table names>" --body-file /tmp/pr_body.md
--label "test-dev,table-approve,metadata-test" --draft so the template structure
and draft status are preserved.
Comment on lines
+15
to
+17
| ``` | ||
| Missing ~/.basedosdados/config.toml — run `basedosdados config init` first. | ||
| ``` |
There was a problem hiding this comment.
Add fence language in prerequisites snippet.
The fenced block at Line 15 is unlabeled (MD040). Use text or bash.
🧰 Tools
🪛 markdownlint-cli2 (0.22.0)
[warning] 15-15: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.claude/commands/databasis-upload.md around lines 15 - 17, The fenced code
block containing the message "Missing ~/.basedosdados/config.toml — run
`basedosdados config init` first." is unlabeled and triggers MD040; update the
opening backticks to include a language such as text or bash (e.g., ```text or
```bash) for the prerequisites snippet so the fenced block is explicitly
labeled.
Comment on lines
+37
to
+41
| # Monkey-patch for requester-pays bucket | ||
| _orig_bucket = gcs.Client.bucket | ||
| def _patched_bucket(self, bucket_name, user_project=None): | ||
| return _orig_bucket(self, bucket_name, user_project=BILLING_PROJECT) | ||
| gcs.Client.bucket = _patched_bucket |
There was a problem hiding this comment.
Define BILLING_PROJECT before monkey-patching GCS client.
Line 40 references BILLING_PROJECT, but the snippet never defines it. This will fail at runtime with NameError and abort upload flow.
Suggested fix
import basedosdados as bd
import google.cloud.storage as gcs
from pathlib import Path
import argparse
+import os
+# Resolve billing project before patch
+BILLING_PROJECT = os.getenv("BILLING_PROJECT")
+if not BILLING_PROJECT:
+ raise ValueError("BILLING_PROJECT must be set")
+
# Monkey-patch for requester-pays bucket
_orig_bucket = gcs.Client.bucket
def _patched_bucket(self, bucket_name, user_project=None):
return _orig_bucket(self, bucket_name, user_project=BILLING_PROJECT)
gcs.Client.bucket = _patched_bucket🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.claude/commands/databasis-upload.md around lines 37 - 41, The monkey-patch
references BILLING_PROJECT but it is not defined, causing a NameError at
runtime; define and validate BILLING_PROJECT before assigning gcs.Client.bucket
so _patched_bucket can reference it. Add a top-level definition for
BILLING_PROJECT (e.g., read from configuration or os.environ and fail fast with
a clear error if missing), then apply the monkey-patch that replaces
gcs.Client.bucket with _patched_bucket; ensure references to _patched_bucket and
gcs.Client.bucket occur after BILLING_PROJECT is set.
There was a problem hiding this comment.
Actionable comments posted: 2
♻️ Duplicate comments (7)
.claude/commands/onboarding-pr.md (2)
38-55:⚠️ Potential issue | 🟡 MinorAdd language identifier to fenced code block.
The PR body template fenced block is missing a language specifier, triggering MD040. Use
mdortext.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/onboarding-pr.md around lines 38 - 55, The fenced code block in .claude/commands/onboarding-pr.md is missing a language specifier and triggers MD040; update the triple-backtick opening fence to include a language identifier (for example change ``` to ```md or ```text) in the PR body template so the block is marked as markdown/plain text and the linter warning is resolved.
31-35:⚠️ Potential issue | 🟠 MajorAdd
--draftflag and build PR body from template structure.The
--body "<changelog>"parameter bypasses the repository's pull request template. Additionally, the--draftflag is missing, and the title pattern should start with a bracketed keyword (e.g.,[Data]).🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/onboarding-pr.md around lines 31 - 35, The gh pr create command is bypassing the repo PR template and missing a draft flag; replace the literal --body "<changelog>" with either --fill (to use the repository PR template) or --body-file <generated_body_path> (if you build the body from your template structure), add the --draft flag, and change the title pattern "[$dataset_slug] <table names>" to start with a bracketed keyword (e.g., "[Data] ${dataset_slug} <table names>"). Ensure you update the invocation of gh pr create to include --draft and use --fill or --body-file instead of --body "<changelog>" so the template is respected.AGENTS.md (1)
242-244:⚠️ Potential issue | 🟡 MinorAdd language identifier to fenced code block.
The command example fenced block is missing a language specifier, triggering MD040. Use
textorbash.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@AGENTS.md` around lines 242 - 244, The fenced code block that contains the example command "Onboard dataset <slug>. Raw files at <path>. Drive folder: BD/Dados/Conjuntos/<slug>/." is missing a language specifier and should be updated to include one (e.g., add ```text or ```bash before the block) so the markdown linter (MD040) stops flagging it; locate the block in AGENTS.md and prepend the appropriate language identifier..claude/commands/onboarding-context.md (1)
37-47:⚠️ Potential issue | 🟡 MinorAdd language identifier to fenced code block.
The fenced block is missing a language specifier, triggering MD040. Use
textfor this structured output example.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/onboarding-context.md around lines 37 - 47, The fenced code block starting with "=== DATASET CONTEXT: <slug> ===" is missing a language specifier which triggers MD040; update the opening fence from ``` to ```text so the block is marked as plain text (i.e., change ``` to ```text immediately before the DATASET CONTEXT block) to satisfy the linter..claude/commands/onboarding-upload.md (2)
15-17:⚠️ Potential issue | 🟡 MinorAdd language identifier to fenced code block.
The error message fenced block is missing a language specifier, triggering MD040. Use
text.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/onboarding-upload.md around lines 15 - 17, Update the fenced code block that currently contains "Missing ~/.basedosdados/config.toml — run `basedosdados config init` first." so the opening fence includes the language identifier text (i.e., change ``` to ```text) to satisfy the MD040 lint rule; no other changes to the block content are needed.
37-41:⚠️ Potential issue | 🔴 CriticalDefine
BILLING_PROJECTbefore using it in the monkey-patch.Line 40 references
BILLING_PROJECT, but it's never defined in the snippet. This will cause aNameErrorat runtime.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/onboarding-upload.md around lines 37 - 41, The monkey-patch uses BILLING_PROJECT but never defines it; declare and initialize BILLING_PROJECT before the patch (e.g., read from configuration or environment) and ensure it's available to the closure used by _patched_bucket; update references around _orig_bucket, _patched_bucket and gcs.Client.bucket to use that defined value (or validate and raise a clear error if missing) so the patched function can call _orig_bucket(self, bucket_name, user_project=BILLING_PROJECT) without NameError..claude/agents/onboarding.md (1)
18-20:⚠️ Potential issue | 🟡 MinorAdd language identifiers to fenced code blocks.
Multiple fenced blocks throughout the file are missing language specifiers, triggering MD040. Use
textfor the example invocation (Line 18), step sequence (Line 28), and verification checklist (Line 46).Also applies to lines 28-40 and 46-55.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/agents/onboarding.md around lines 18 - 20, Add language identifiers "text" to the three fenced code blocks that currently lack them: the example invocation block containing "Onboard dataset <slug>. Raw files at <path>. Drive folder: BD/Dados/Conjuntos/<slug>/.", the step sequence block that lists the onboarding steps, and the verification checklist block; update the opening backticks for each of these fenced blocks to "```text" so the markdown linter (MD040) no longer flags them.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.claude/commands/onboarding-clean.md:
- Around line 14-24: The fenced code block showing the dataset folder tree in
.claude/commands/onboarding-clean.md lacks a language identifier (causing
MD040); edit that block's opening backticks to include the language "text"
(i.e., change ``` to ```text) so the diagram is properly labeled while leaving
the internal directory content unchanged.
In @.claude/commands/onboarding-metadata.md:
- Around line 148-159: Update the fenced code block that begins with "===
METADATA REGISTRATION COMPLETE (env=<env>) ===" to include a language identifier
to satisfy MD040; add "text" immediately after the opening triple backticks so
the block becomes a labeled text code block; ensure the rest of the block
content (Dataset/Files/Tables/Next step lines) remains unchanged.
---
Duplicate comments:
In @.claude/agents/onboarding.md:
- Around line 18-20: Add language identifiers "text" to the three fenced code
blocks that currently lack them: the example invocation block containing
"Onboard dataset <slug>. Raw files at <path>. Drive folder:
BD/Dados/Conjuntos/<slug>/.", the step sequence block that lists the onboarding
steps, and the verification checklist block; update the opening backticks for
each of these fenced blocks to "```text" so the markdown linter (MD040) no
longer flags them.
In @.claude/commands/onboarding-context.md:
- Around line 37-47: The fenced code block starting with "=== DATASET CONTEXT:
<slug> ===" is missing a language specifier which triggers MD040; update the
opening fence from ``` to ```text so the block is marked as plain text (i.e.,
change ``` to ```text immediately before the DATASET CONTEXT block) to satisfy
the linter.
In @.claude/commands/onboarding-pr.md:
- Around line 38-55: The fenced code block in .claude/commands/onboarding-pr.md
is missing a language specifier and triggers MD040; update the triple-backtick
opening fence to include a language identifier (for example change ``` to ```md
or ```text) in the PR body template so the block is marked as markdown/plain
text and the linter warning is resolved.
- Around line 31-35: The gh pr create command is bypassing the repo PR template
and missing a draft flag; replace the literal --body "<changelog>" with either
--fill (to use the repository PR template) or --body-file <generated_body_path>
(if you build the body from your template structure), add the --draft flag, and
change the title pattern "[$dataset_slug] <table names>" to start with a
bracketed keyword (e.g., "[Data] ${dataset_slug} <table names>"). Ensure you
update the invocation of gh pr create to include --draft and use --fill or
--body-file instead of --body "<changelog>" so the template is respected.
In @.claude/commands/onboarding-upload.md:
- Around line 15-17: Update the fenced code block that currently contains
"Missing ~/.basedosdados/config.toml — run `basedosdados config init` first." so
the opening fence includes the language identifier text (i.e., change ``` to
```text) to satisfy the MD040 lint rule; no other changes to the block content
are needed.
- Around line 37-41: The monkey-patch uses BILLING_PROJECT but never defines it;
declare and initialize BILLING_PROJECT before the patch (e.g., read from
configuration or environment) and ensure it's available to the closure used by
_patched_bucket; update references around _orig_bucket, _patched_bucket and
gcs.Client.bucket to use that defined value (or validate and raise a clear error
if missing) so the patched function can call _orig_bucket(self, bucket_name,
user_project=BILLING_PROJECT) without NameError.
In `@AGENTS.md`:
- Around line 242-244: The fenced code block that contains the example command
"Onboard dataset <slug>. Raw files at <path>. Drive folder:
BD/Dados/Conjuntos/<slug>/." is missing a language specifier and should be
updated to include one (e.g., add ```text or ```bash before the block) so the
markdown linter (MD040) stops flagging it; locate the block in AGENTS.md and
prepend the appropriate language identifier.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 10ed3e49-8100-4690-8587-e8136b06e249
📒 Files selected for processing (11)
.claude/agents/onboarding.md.claude/commands/onboarding-architecture.md.claude/commands/onboarding-clean.md.claude/commands/onboarding-context.md.claude/commands/onboarding-dbt-run.md.claude/commands/onboarding-dbt.md.claude/commands/onboarding-discover.md.claude/commands/onboarding-metadata.md.claude/commands/onboarding-pr.md.claude/commands/onboarding-upload.mdAGENTS.md
✅ Files skipped from review due to trivial changes (4)
- .claude/commands/onboarding-dbt-run.md
- .claude/commands/onboarding-dbt.md
- .claude/commands/onboarding-architecture.md
- .claude/commands/onboarding-discover.md
This was referenced Mar 31, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Descrição do PR:
Esse PR cria o fluxo e comandos que agentes de IA podem usar para subir dados na BD a partir de um ambiente local configurado.
Relacionado com:
Summary by CodeRabbit
New Features
Documentation