docs: Adds empty AGENTS.md

[vagmcs]

Description

Please provide a clear and concise description of the changes introduced by this pull request.

• Why are these changes necessary?
• Are there any related issues or pull requests?

Checklist

Before submitting this PR, please ensure you have completed the following:

• I have performed a self-review of my code.
• I have added or updated unit/integration tests to cover my changes.
• I have documented any relevant changes in the documentation.

[claude]

Review Summary

Overall: This PR adds an empty AGENTS.md file with no content.

Issue: The file is completely empty. An AGENTS.md serves as documentation for AI agents (e.g., Codex, Claude Code) about how to work in this repository — conventions, build commands, test commands, environment setup, etc. An empty file provides no value and could mislead agents into thinking documentation exists when it does not.

Suggestion: Either populate the file before merging, or defer the PR until content is ready. At minimum, consider a stub like:

# Agents

<!-- TODO: Add repo conventions, build/test commands, and notes for AI coding agents -->

[github-actions]

Review Summary

Two small issues found; no correctness or security problems.

Critical: None

Notable:

• .github/workflows/claude-review.yml line 25 — continue-on-error: true masks step failures silently. A broken API key or action regression will make the workflow appear green with no review ever posted. Either remove it or add a downstream failure-notification step.
• AGENTS.md line 34 — The **Docstrings:** entry is missing its leading - , so it renders as a standalone paragraph rather than a list item. One-character fix.

The github_token addition and the workflow rename are straightforward and correct.

[github-actions]

continue-on-error: true silently swallows failures of this step — if the API key expires, the action is broken, or the review hits a rate limit, CI will still go green with no indication that reviews stopped running. This makes failures invisible and hard to debug.

Unless there's a specific reason to tolerate failures here, remove this flag so broken review runs are surfaced as a visible CI failure (or at minimum as a workflow annotation).

Suggested change

id: claude-pr-review

[github-actions]

uv run poe bump is listed here as a normal command, but the Boundaries section (line 64) flags it as "Ask first". The placement here implies it's routine, which contradicts the intent. Consider moving it out of the general commands table or adding an inline note that it requires approval.

Suggested change

	uv run pre-commit run --all-files # Run all pre-commit hooks manually
	uv run poe bump # Bump version + generate changelog (ask first — see Boundaries)

[github-actions]

Review Summary

Overall: Small, well-structured PR. Two issues worth addressing.

Critical

None.

Minor

1. continue-on-error: true (workflow) — Silently hides failures of the Claude review step. A broken API key or rate limit won't surface in CI. Remove unless there's a deliberate reason to treat this step as optional.

2. uv run poe bump listed as a standard command (AGENTS.md) — Contradicts the Boundaries section which marks it as 'ask first'. Add an inline note to avoid confusion for agents reading the file top-to-bottom.