chore: add AGENTS.md + CLAUDE.md for unified AI agent context

AGENTS.md

@@ -0,0 +1,112 @@
+# BQuest Agent Guide
+
+## Project Overview
+
+**BQuest** (v0.5.8, Beta) is a Python library for validating and testing Google BigQuery queries using pandas DataFrames. It creates temporary test tables in BigQuery from JSON/DataFrame data, executes queries against them, and validates results without touching production tables.
+
+**Status**: Beta — breaking changes expected until 1.0; pin your version.
+
+## Tech Stack
+
+- **Python** 3.10–3.12
+- **Package manager**: `uv` (>=0.6.0)
+- **Build**: `hatchling`
+- **Key deps**: `google-cloud-bigquery[bqstorage,pandas]>=3.8`, `pandas>=2.0`, `pandas-gbq>=0.19`, `sqlvalidator>=0.0.20`
+- **Dev tools**: `ruff>=0.2.2` (line length 120), `mypy>=1.3.0` (strict), `pytest>=7.3.1`, `poethepoet` (task runner), `pip-audit`, `creosote`, `zizmor`
+- **Docs**: MkDocs Material + mkdocstrings, versioned with `mike`
+
+## Repository Structure
+
+```
+bquest/
+├── bquest/
+│   ├── tables.py      # BQTable, BQTableDefinition, BQTableDataframeDefinition, BQTableJsonDefinition, BQTableDefinitionBuilder
+│   ├── runner.py      # BaseRunner, BQConfigRunner, BQConfigFileRunner, BQConfigSubstitutor, SQLRunner, SQLFileRunner
+│   ├── dataframe.py   # standardize_frame_numerics(), assert_frame_equal() (order-independent)
+│   └── util.py        # is_sql() — detect SQL in strings
+├── tests/
+│   ├── unit/          # No external deps (mocked BigQuery)
+│   └── integration/   # Real BigQuery (requires auth + bquest dataset)
+├── docs/              # MkDocs source
+├── pyproject.toml
+└── uv.lock
+```
+
+## Local Setup
+
+```bash
+uv sync
+gcloud auth application-default login   # For integration tests
+```
+
+**BigQuery setup**: Create dataset `bquest` in EU, set table expiration ~1 hour.
+
+## Commands (via `poe`)
+
+```bash
+uv run poe lint              # ruff check
+uv run poe format            # ruff format --fix
+uv run poe type              # mypy strict
+uv run poe test_unit         # Unit tests (fast, no BQ needed)
+uv run poe test_integration  # Integration tests (requires BQ auth)
+uv run poe test              # Both
+uv run poe check_vulnerable_dependencies
+uv run poe check_unused_dependencies
+uv run poe check_githubactions
+```
+
+## Architecture & Key Classes
+
+### Tables (`tables.py`)
+- **`BQTable`** — wraps a BQ table ref; `load_to_bq()`, `to_df()`, `delete()`
+- **`BQTableDataframeDefinition`** — create table from pandas DataFrame (via `pandas_gbq.to_gbq`)
+- **`BQTableJsonDefinition`** — create table from JSON rows
+- **`BQTableDefinitionBuilder`** — factory; use this instead of direct instantiation
+
+### Runners (`runner.py`)
+- **`BQConfigRunner`** — execute BQ config dict
+- **`BQConfigFileRunner`** — load config from `.py` file via `ast.literal_eval`
+- **`BQConfigSubstitutor`** — maps original table IDs → test table IDs
+- **`SQLRunner`** — direct SQL with string replacements and parameter substitution
+- **`SQLFileRunner`** — file-based SQL runner
+
+### DataFrame utils (`dataframe.py`)
+- **`standardize_frame_numerics()`** — rounds floats to 2 decimals, converts Int64→float64
+- **`assert_frame_equal()`** — order-independent comparison (sorts columns + rows); use for test assertions
+
+## Conventions
+
+### Test Table Naming
+`{original_table_id}_{uuid}` — special chars (`.{}$-`) → `_`
+
+### BQConfig Dict Format
+```python
+{
+    "query": "SELECT ...",
+    "source_tables": {"key": "schema.table"},
+    "feature_table_name": "schema.result_table",
+    "start_date": "date_column_name",
+}
+```
+
+### Code Style
+- Line length: 120 (ruff), strict mypy, Ruff rules: S, B, A, C4, ERA, YTT, I, T20, E, W, F, RET
+- `@pytest.mark.unit` — no external deps; `@pytest.mark.integration` — requires BigQuery
+
+## CI/CD
+
+- **`tests.yml`**: Python 3.10–3.12 matrix; integration tests skip on fork PRs
+- **`code-checks.yml`**: ruff, pip-audit, creosote, zizmor
+- **`type.yml`**: mypy strict; **`coverage.yml`**: Codecov 75% target
+- **`release.yml`**: `uv build` → docs versioned with `mike` → `uv publish` (OIDC trusted publishing)
+- Auth: Workload Identity Federation (no static keys)
+
+## Key Gotchas
+
+1. **Float precision**: Always use `standardize_frame_numerics()` + `assert_frame_equal()` — raw floats cause flaky tests
+2. **Curly braces in SQL**: Can't use `{substitutions}` if query contains `{}` (e.g., regex). Use `string_replacements` instead
+3. **Dataset must pre-exist**: `bquest` dataset won't be auto-created
+4. **Location mismatch**: Pass `location` to `BQTableDefinitionBuilder` if not EU
+5. **`pandas-gbq` exempt from creosote**: Dynamically imported; false positive for "unused"
+6. **`BQConfigFileRunner` uses `ast.literal_eval`**: Config files can only contain Python literals
+7. **Partition filter**: `to_df()` temporarily disables `requirePartitionFilter` for test tables

CLAUDE.md

@@ -0,0 +1 @@
+@AGENTS.md