docs: add cli-help-text skill (#1051)

[sdairs]

Closes #1051. Stacked on #1048.

Moves the help-text standard out of the root AGENTS.md into a repo skill that any coding agent can load, in the same layout as #1048.

Changes

• New skill: .agents/skills/cli-help-text/SKILL.md. It keeps every rule from the old "Writing help text" section: screen shape, about and flag rules, option order and help_order ranks, the CONTEXT FOR AGENTS: limits, the permission lines from permissions.rs, and the test policy. New in the skill:
  • A before/after example of an agent block.
  • A short checklist: render each changed screen with cargo run -q -p clickhousectl -- <path> --help, check each line, run the tests.
  • Two rules taken from current vendor guidance: write agent context plainly, with no capitals or "MUST" for emphasis (Anthropic's guidance for Claude 4.5 and later, OpenAI's for GPT-5.5); and use value_enum instead of listing valid values in prose.
  • Runtime gotchas worth stating now include retry safety and long waits, matching what OpenAI says a tool description should cover.
• Claude Code: .claude/skills/cli-help-text is a symlink to the skill.
• .gitignore: un-ignores the new skill and its symlink.
• AGENTS.md: the section becomes a two-line pointer (17.4 KB → 13.9 KB). The Tests section now points at the skill.
• openapi-drift-remediation skill: its CLI PR step now points at the new skill.

The skill description is in the third person and leads with its triggers. Codex shortens skill descriptions when many skills are installed, so the triggers need to come first.

Gates

• Python tests pass; cargo fmt --all --check is clean.
• The Cloud classifier selects no suites for this diff.
• No Rust code changed.

🤖 Generated with Claude Code