TDD: tool_max_rounds config silently ignored when falsy, and undocumented in the LLM agent specification #145

New issue

Open

opened 2026-08-21 22:36:06 +00:00 by CoreRasurae · 0 comments

CoreRasurae commented

2026-08-21 22:36:06 +00:00

Member

Metadata

Commit Message: test(agents): capture tool_max_rounds zero-value fallback (#144)

Branch: tdd/m1-tool-max-rounds-precedence

Background and context

Companion TDD issue for #144 per the project's mandatory Bug Fix TDD workflow. This issue captures a failing Behave scenario proving that tool_max_rounds: 0 in an LLM agent config is silently overridden by TOOL_MAX_ROUNDS/the default, instead of being clamped to 1 as documented.

Current behavior

See #144. LLMAgent._execute_tool_loop() uses self.config.get("tool_max_rounds") or os.environ.get("TOOL_MAX_ROUNDS", "20"), so an explicitly-configured 0 is discarded by Python's or truthiness check and the loop runs with the env var/default limit instead of being clamped to 1.

Expected behavior

A Behave scenario in features/llm_agent_tool_loop.feature, tagged @tdd_issue @tdd_issue_144 @tdd_expected_fail, asserts that with tool_max_rounds: 0 configured, _execute_tool_loop() permits exactly 1 tool round. The assertion (using assert/AssertionError only) fails while the bug is present, and the CI TDD-inversion hook makes the scenario pass as a result.

Acceptance criteria

Scenario tagged with @tdd_issue, @tdd_issue_144, and @tdd_expected_fail
The failing step uses assert / raises AssertionError only (never ValueError, RuntimeError, etc.)
Scenario genuinely fails (reproduces the bug) when run without @tdd_expected_fail
nox -s unit_tests passes (CI passes via tag inversion while the bug is unfixed)
Committed to the tdd/m1-tool-max-rounds-precedence branch

Supporting information

Bug issue: #144 (Forgejo dependency: #144 depends on this TDD issue).

Subtasks

Add a failing Behave scenario in features/llm_agent_tool_loop.feature for tool_max_rounds: 0
Apply all three required tags (@tdd_issue, @tdd_issue_144, @tdd_expected_fail)
Confirm the scenario fails via AssertionError without the bug fix
Run nox -s unit_tests, confirm CI passes via tag inversion

Definition of Done

This issue is complete when:

The failing test scenario is committed with all three required tags.
The commit first line matches the Commit Message in Metadata exactly.
The commit is pushed to the branch matching the Branch in Metadata exactly.
The commit is submitted as a PR to master, reviewed, and merged.
This TDD issue is closed.

PR: #146 (awaiting review — not self-approved).

## Metadata Commit Message: `test(agents): capture tool_max_rounds zero-value fallback (#144)` Branch: `tdd/m1-tool-max-rounds-precedence` ## Background and context Companion TDD issue for #144 per the project's mandatory Bug Fix TDD workflow. This issue captures a failing Behave scenario proving that `tool_max_rounds: 0` in an LLM agent config is silently overridden by `TOOL_MAX_ROUNDS`/the default, instead of being clamped to 1 as documented. ## Current behavior See #144. `LLMAgent._execute_tool_loop()` uses `self.config.get("tool_max_rounds") or os.environ.get("TOOL_MAX_ROUNDS", "20")`, so an explicitly-configured `0` is discarded by Python's `or` truthiness check and the loop runs with the env var/default limit instead of being clamped to 1. ## Expected behavior A Behave scenario in `features/llm_agent_tool_loop.feature`, tagged `@tdd_issue @tdd_issue_144 @tdd_expected_fail`, asserts that with `tool_max_rounds: 0` configured, `_execute_tool_loop()` permits exactly 1 tool round. The assertion (using `assert`/`AssertionError` only) fails while the bug is present, and the CI TDD-inversion hook makes the scenario pass as a result. ## Acceptance criteria - [x] Scenario tagged with `@tdd_issue`, `@tdd_issue_144`, and `@tdd_expected_fail` - [x] The failing step uses `assert` / raises `AssertionError` only (never `ValueError`, `RuntimeError`, etc.) - [x] Scenario genuinely fails (reproduces the bug) when run without `@tdd_expected_fail` - [x] `nox -s unit_tests` passes (CI passes via tag inversion while the bug is unfixed) - [x] Committed to the `tdd/m1-tool-max-rounds-precedence` branch ## Supporting information Bug issue: #144 (Forgejo dependency: #144 depends on this TDD issue). ## Subtasks - [x] Add a failing Behave scenario in `features/llm_agent_tool_loop.feature` for `tool_max_rounds: 0` - [x] Apply all three required tags (`@tdd_issue`, `@tdd_issue_144`, `@tdd_expected_fail`) - [x] Confirm the scenario fails via `AssertionError` without the bug fix - [x] Run `nox -s unit_tests`, confirm CI passes via tag inversion ## Definition of Done This issue is complete when: - The failing test scenario is committed with all three required tags. - The commit first line matches the Commit Message in Metadata exactly. - The commit is pushed to the branch matching the Branch in Metadata exactly. - The commit is submitted as a PR to master, reviewed, and merged. - This TDD issue is closed. --- PR: #146 (awaiting review — not self-approved).

Rows
Columns