[leemour]

Task

Two cli-core gaps whatsapp-cli (wa) hit: three copies of one Markdown parser, and messages sync advice that store-history messengers cannot follow (UX-59). (The MCP delete-for-everyone item was dropped: #1439 settled it, since agents never delete for everyone.)

Why

• Markdown. tg-cli, max-cli and wa each carry a copy of the same parser (wa's is tg's plus max's # headings).
• UX-59. fix: gaps whatsapp-cli hit — providerRef leak, store-history sync, MCP delete for everyone, stored lookup in connect #1423 removed --sync-first and the search advice for history: "store". Five more answers still named messages sync, which such a messenger refuses.

What

• parseMarkdownSpans(text): FormattedText, exported from .. It reads tg's dialect plus # headings and a one-line ``` block, sorts the spans, and validates them. Telegram's nesting refusal, tg's trailing-space trim and max's nested-link refusal stay with each messenger's renderer. I checked tg's and wa's own test suites with this parser plus their post-processing: all 20 cases pass. max's dialect is different (* italic, __ bold, ++, ^^), so max does not switch yet: that is the owner's call (NEED-1414).
• UX-59: the new syncsOneChat(messenger) is true when history !== "store" || historyOnDemand. <chat> advice needs it; --all advice needs history !== "store". The services already know the messenger, so the members audit and chat stats set fetch only when a sync can run. Their answer builders keep the "store does not hold the chat whole" caveat (missing with no step) and drop next. I did this instead of passing the messenger into those builders. search conversations / conversations related (CLI and MCP) do take the messenger now. MCP chats_documents_list, store check (with historyOnDemand it names the first chat that is behind) and the message-search coverage line are gated the same way.

Testing

• pnpm typecheck, biome on the changed files.
• New: markdown-spans.test.ts (tests ported from tg, plus headings, the one-line block and nesting left to the messenger), UX-59 tests in chat-stats.test.ts, store-mode.test.ts (store check), conversation-answers.test.ts and chat-answers.test.ts.
• Full vitest run locally before the rebase (4939 passed); after rebasing on fix(mcp): one vocabulary for agent commands; agents never delete for everyone; --permission only tightens #1439, typecheck plus the touched suites (580 passed).

Consumer impact

• tg-cli: can swap the parser in src/telegram/format-markdown.ts for parseMarkdownSpans from @wirecat/cli-core, keeping its nesting check and trailing-space trim. One change: # x becomes a heading span, and a one-line ```x``` becomes pre (that used to be an error). Its entity mapping has to handle heading.
• wa: can swap src/adapter/format-markdown.ts's parser for parseMarkdownSpans and keep toWhatsApp. Same output.
• max-cli: no change until the owner decides on its dialect (NEED-1414).
• cli-memo: none.
• UX-59 changes only what answers print. No API changed for consumers.

🤖 Generated with Claude Code