Document the markdown→blocks conversion and the fallback-text trap (--help + skill)

[ygpark80]

Problem

chat send/edit --help says only "Message text in markdown format". It doesn't say the text is converted to Slack blocks (markdownToBlocks), nor that the API text field is kept as the raw-markdown fallback (used for notifications only).

That fallback field is a diagnosis trap. Re-fetching a sent message shows:

text:   '**bold** and [link](<https://example.com>)'            ← raw fallback, looks broken
blocks: '*bold* and <https://example.com|link>'                  ← what actually renders, correct

Newlines look collapsed in text too (para one para two) while blocks preserve paragraphs and soft breaks.

Verified fallout (2026-07-16, two independent Claude Code sessions):

• one concluded "holla collapses newlines" and bypassed holla with raw chat.update calls;
• another concluded "markdown conversion doesn't happen, **bold** renders literally" and started hand-writing mrkdwn — which the skill itself warns double-converts.

Both messages were in fact rendering perfectly; both sessions were reading the fallback text field.

Proposal

1. --help for chat send/reply/edit: one NOTE line — "standard markdown, rendered via Slack blocks; the API text field is a raw fallback — verify rendering via blocks, not text".
2. skills/slack/SKILL.md: add a "Verifying what was sent" section stating the same (check blocks, never judge rendering by text).
3. (Optional) chat get / post-send echo: render from blocks instead of message.text so the CLI itself doesn't display the misleading fallback.

🤖 Generated with Claude Code