GeezerVison

Summary

Two Pydantic field validators were copy-pasted across request/model modules. This folds them into core/:

• Non-negative limit — the same validator appeared 6 times (services/search, services/index, services/runs, and three in services/viewer). Consolidated into core.models:

  def _require_non_negative_limit(value): ...  # "limit must be non-negative"
  NonNegativeLimit         = Annotated[int | None, AfterValidator(...)]  # optional, default None
  RequiredNonNegativeLimit = Annotated[int,        AfterValidator(...)]  # required, e.g. default 50

  Fields become limit: NonNegativeLimit = None / limit: RequiredNonNegativeLimit = 50, dropping the per-model @field_validator/@classmethod boilerplate.

• Optional-text — the if value is None: return None; return required_text(...) pattern appeared 5 times (services/runs/models, workflows/build ×2, workflows/garden, workflows/ingest). Consolidated into core.text.optional_text(value, field_name); each validator now delegates in one line, preserving its per-model error label.

Field-specific validators with distinct messages (page_limit, transcript/filesystem size, harness count/duration) were intentionally left alone — merging them would change error strings and over-abstract.

Why

Reduce copy-paste and give these validators one home so behavior/wording stays consistent as new request types are added. Net −34 lines across 10 files.

Note: the per-source errors.py duplication (unavailable_runtime/first_error_line) was deliberately not touched — tests/test_architecture.py enforces that split-by-responsibility layout.

Verification

uv run pytest        # 564 passed
uv run ruff check .  # All checks passed!

Also confirmed error messages/accept-reject behavior are byte-for-byte identical (negative limit → "limit must be non-negative"; required-int limit=None still rejected; empty title → "build request text must not be empty"; None accepted).

Docs and wiki

• Not applicable.

Notes for reviewers

NonNegativeLimit keeps the validator inside Annotated[int | None, ...] (rather than Annotated[int, ...] | None) so a single clean ValidationError is produced — matching the previous @field_validator output that surfaces via cli/execution.py.

Link to Devin session: https://app.devin.ai/sessions/2ec4dbc36bd94d9da6e2b3e9ad796454
Requested by: @rohans0509