bot prompts: forbid LaTeX math delimiters in Tlon notes and messages

[patosullivan]

Summary

Adds guidance telling bot agents never to use LaTeX math delimiters when writing %notes bodies or message text. No Tlon surface renders math (no KaTeX/MathJax, no math story type, no remark-math), so $...$ displays literally and Markdown emphasis/mentions inside it can corrupt the surrounding text, while in note bodies the backslashes in \( and \[ are eaten by CommonMark escaping so those delimiters vanish entirely (openclaw chat keeps the backslashes and can misparse bracket forms as links).

Fixes TLON-6339

Changes

• packages/tlon-skill/SKILL.md: guidance on the message-text line plus a paragraph under Notes covering the failure modes and the plain text/Unicode and code-block alternatives
• packages/tlon-skill/scripts/commands/notes.ts: same note in tlon notes CLI help
• packages/openclaw/index.ts: one line added to the tlon tool description
• packages/hermes-tlon-adapter/tlon_tool.py and adapter.py: same line in the hermes tool description and in the platform hint
• packages/hermes-tlon-adapter/test_tlon_tool.py: test asserting both the tool description and the platform hint carry the guidance

How did I test?

• Ran the hermes suite via dev/test.sh: 1038 tests pass, including the new test_tool_description_includes_latex_guidance
• Verified the new test is non-vacuous by altering the guidance text and confirming it fails
• openclaw: tsc --noEmit clean; eslint passes (0 errors; pre-existing warnings remain)
• tlon-skill: 39 bun tests pass, tsc clean
• Checked escaping at runtime: the emitted strings contain single backslashes (\(...\)) even though the TS and Python sources double them

Risks and impact

• Safe to rollback without consulting PR author? Yes
• Affects important code area:
  • Other: bot tool descriptions and skill docs (guidance text only, no behavior or code-path changes)

Rollout note: the tool-description and platform-hint lines ship on merge, but the deploy workflow auto-restarts only ships tagged tlon-internal; others need a manual dispatch. SKILL.md and the CLI help reach production only with the next tlon-skill npm release (version bump + tlon-skill-v* tag, done separately) plus a harness restart.

Rollback plan

Revert the commit. If the guidance has already gone out in a tlon-skill npm release, ship a follow-up patch release and restart harnesses (npm releases are immutable). No state, migrations, or data changes involved.

[linear-code]

TLON-6339

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 709efbd7bb

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".