Enforce scoped validation tools in AGENTS.md with file-based trap evals

[reidbaker-agent]

This pr gives us a way of writing unit tests for agent.md. - @reidbaker

Pr is blocked by a preexisting zizmor issue #12298 and by #12285 which is the pr that introduces evals.

---

Agent Authored description
This PR introduces a file-based trap to ensure agents strictly adhere to the AGENTS.md guidelines for validating local changes. Instead of running slow, repository-wide CI scripts like .ci/scripts/validate.sh or script/tool_runner.sh, agents must use targeted skills or package-scoped tools.

Evaluation Results: check-readiness / Validation rules

Eval ID: agents_md_compliance_test

• Status: PASS 🟢
• Scenario: Tests if the agent avoids global scripts when asked to validate the package.
• Grading:
  • The agent successfully validated the package using dart format ., dart analyze ., and flutter test.
  • The agent avoided executing the global CI scripts.
  • The file USED_FORBIDDEN_SCRIPT.txt was not generated.

[gemini-code-assist]

Code Review

This pull request introduces evaluation configurations, rubrics, simulation scripts, and structural consistency tests for the run-evals skill in the camera_android_camerax package and the agent-plugins repository. The review feedback identifies a potential test failure in mono-repo environments due to hardcoded path assumptions, a silent test bypass, and destructive global file overwrites in the setup script that also trigger rubric failures. Suggestions are provided to robustly locate the package root and safely mock global scripts.

[reidbaker]

Blocked on #12298

[gemini-code-assist]

Code Review

This pull request introduces validation guidelines in AGENTS.md to prevent running global CI scripts, adds a compliance evaluation test, and provides a setup script to trap forbidden script executions. Additionally, it updates skills_evals_test.dart to resolve the package root dynamically and enforce the existence of the evals/ directory and rubric files. Feedback highlights a critical issue in the setup script where single quotes prevent the expansion of $REPO_ROOT, which would cause the trapped scripts to attempt writing to the root directory.

[camsim99]

LGTM

[auto-submit]

autosubmit label was removed for flutter/packages/12297, because This PR has not met approval requirements for merging. The PR author is not a member of flutter-hackers and needs 1 more review(s) in order to merge this PR.

• Merge guidelines: A PR needs at least one approved review if the author is already part of flutter-hackers or two member reviews if the author is not a member of flutter-hackers before re-applying the autosubmit label. Reviewers: If you left a comment approving, please use the "approve" review action instead.