Introduce GitLab monorepo agent instructions

Feedback issue: https://gitlab.com/gitlab-org/gitlab/-/work_items/596611+s

What does this MR do and why?

Establishes a minimal standard for AI agent instruction files in the GitLab monorepo. Only markdown instruction files are committed — no skills, hooks, commands, or tool-specific configs.

Commit 1: Instruction files

• Adds top-level AGENTS.md and CLAUDE.md (identical content, kept in sync)
• Creates missing CLAUDE.md parity files for existing subdirectory AGENTS.md files (doc/, workhorse/, ee/frontend_islands/)
• Updates .gitignore with non-rooted patterns for AGENTS.local.md, CLAUDE.local.md, and .ai/*
• Updates .ai/README.md to document conventions

Commit 2: Doctor lint script

• Implements scripts/ai_harness/doctor with --fix support using Railway Oriented Programming (Gitlab::Fp::Result)
• Validates: parity between CLAUDE.md/AGENTS.md pairs, .ai/ reference resolution, .gitignore coverage, and no forbidden tool-specific files committed

References

• Introduce GitLab monorepo agent instructions (#594821) • Chad Woolley
• Supersedes GitLab Monorepo AI Harness Configuration Proof ... (#594027 - closed) • Chad Woolley

Screenshots or screen recordings

Not applicable (CLI tooling, no UI changes).

Exploratory Testing

See AI-executed exploratory testing report here

How to set up and validate locally

Integration spec: spec/tooling/ai_harness/doctor/integration_spec.rb

ENABLE_SPRING=1 bin/rspec spec/tooling/ai_harness/doctor/

Manual validation:

1. Run scripts/ai_harness/doctor from the repo root — all checks should show OK, exit 0
2. Delete CLAUDE.md and run scripts/ai_harness/doctor — parity check shows FAIL, exit 1
3. Run scripts/ai_harness/doctor --fix — CLAUDE.md is recreated, parity check shows FIXED, exit 0
4. Stage a forbidden file: mkdir -p .claude/rules && echo '# rule' > .claude/rules/test.md && git add -f .claude/rules/test.md
5. Run scripts/ai_harness/doctor — forbidden files check shows FAIL with the file path
6. Clean up: git rm -f .claude/rules/test.md && rm -rf .claude/rules
7. Run scripts/ai_harness/doctor --help — prints usage, exit 0

MR acceptance checklist

Evaluate this MR against the MR acceptance checklist. It helps you analyze changes to reduce risks in quality, performance, reliability, security, and maintainability.