Skill trigger precedence should override domain-looking text

Bug Description

When a user message starts with an English Coach trigger such as polish:, Hermes can incorrectly route the request to a domain-specific skill if the text after the trigger looks like a config/code/GitHub task.

Example:

polish: what's the current config's max_turns & approvals.mode

Expected behavior: English Coach handles this as a polish request and rewrites the sentence.

Actual behavior: Hermes treated it as a Hermes config question, loaded the Hermes config skill, read ~/.hermes/config.yaml, and answered with the config values.

Why this matters

Explicit prefix triggers like polish:, fix:, check:, word:, say:, en:, and zh: should have higher priority than semantic/domain routing. Otherwise learner-facing commands become unreliable whenever the text being corrected mentions technical topics.

Proposed behavior

If a user message begins with an English Coach trigger followed by :, treat everything after the colon as the text payload for that mode, even if the payload mentions configs, code, commands, GitHub, or other domain-specific terms.

Suggested rule:

When the user message begins with an English Coach trigger like `polish:`, `fix:`, `check:`, `correct:`, `proofread:`, `say:`, `word:`, `words:`, `en:`, or `zh:`, treat everything after the colon as text to process. Do not answer the domain question unless the user asks without an English Coach trigger.

Local workaround already applied

The local english-coach skill was patched with this precedence rule.

Environment

• Interface: Telegram gateway
• Affected area: skill routing / trigger precedence