chore: add some skills and update agents doc

[rickeylev]

Some misc instructions for agents to follow.

• Tell them about the multiple modules. I've found they can get
  confused trying to run code in different sub-modules.
• Tell them to never run expunge. This hard-wipes caches and trashes
  build performance, not to mention takes many minutes itself
• Tell them to use the fast-tests config, which captures more than
  just integration tests now.

Add skills for interacting with build kite, which makes automating
PRs a bit easier.

[gemini-code-assist]

Code Review

This pull request introduces new agent skills for retrieving Buildkite results and retrying failed jobs, along with updates to the AGENTS.md documentation regarding Bazel configurations and repository structure. The review feedback suggests several improvements to the Python scripts, such as adding timeouts to network requests for robustness, handling URL fragments more safely, and ensuring that job filtering logic correctly implements the documented regex support.

[gemini-code-assist]

The URL construction for the JSON endpoint should handle potential trailing slashes in the build URL to ensure a valid path.

    json_url = build_url if build_url.endswith(".json") else build_url.rstrip("/") + ".json"

[gemini-code-assist]

Network requests should include a timeout to prevent the script from hanging indefinitely if the server is unresponsive.

Suggested change

	with urllib.request.urlopen(json_url) as response:
	with urllib.request.urlopen(json_url, timeout=10) as response:

[gemini-code-assist]

Using split("#") can cause a ValueError if the URL contains multiple hash characters. rsplit("#", 1) is safer for extracting the fragment identifier.

Suggested change

	base, job_id = job_url.split("#")
	base, job_id = job_url.rsplit("#", 1)

[gemini-code-assist]

Network requests should include a timeout to prevent the script from hanging indefinitely.

Suggested change

	with urllib.request.urlopen(raw_url) as response:
	with urllib.request.urlopen(raw_url, timeout=10) as response:

[gemini-code-assist]

The re module is required to support regex matching for job names as described in the help text.

Suggested change

	import os
	import os
	import re

[gemini-code-assist]

Network requests should include a timeout to prevent the script from hanging indefinitely.

Suggested change

	with urllib.request.urlopen(req) as response:
	with urllib.request.urlopen(req, timeout=10) as response:

[gemini-code-assist]

The implementation should use regex matching to align with the help text for the --job-name argument, which claims to support regex.

        if args.job_name and not re.search(args.job_name, job_name, re.IGNORECASE) and \
           not re.search(args.job_name, job.get("step_key", ""), re.IGNORECASE):
            continue