fix(slack): Preserve hyperlink URLs in assistant context#156
fix(slack): Preserve hyperlink URLs in assistant context#156
Conversation
When a user sends a message with Slack hyperlinks (<url|display text>), the URLs were stripped before reaching the model. The adapter's extractPlainText() converts the AST to plain text, discarding link URLs entirely. The assistant only saw the display text with no knowledge of embedded URLs. Use message.formatted (the AST) instead of message.text, and walk the tree with a new extractTextPreservingLinks() helper that renders links as [text](url) while keeping all other content as plain text. Also update stripLeadingBotMention to handle the @USER_ID format produced by the AST (the existing <@USER_ID> regex was dead code since the adapter's toAst already strips angle brackets). Fixes #137 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
visitNode only handled 'text' type nodes for reading node.value. mdast inlineCode and code nodes also store content in value (not in children), so backtick-wrapped content was silently dropped. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add newline separators for block-level AST containers (list, listItem, blockquote, etc.) so bulleted lists and similar structures are not concatenated without whitespace. Remove the dead <@USER_ID> regex in stripLeadingBotMention — the AST never produces angle-bracket mentions, and chaining it with the new @USER_ID regex risked double-stripping. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit b73c604. Configure here.
| "listItem", | ||
| "blockquote", | ||
| "heading", | ||
| ]); |
There was a problem hiding this comment.
Paragraph in BLOCK_TYPES inserts newlines between inline elements
High Severity
paragraph and heading are included in BLOCK_TYPES, which causes their inline children (text, links, emphasis, etc.) to be joined with "\n" instead of "". In a standard markdown AST, paragraphs contain inline children — the paragraph itself is a block element, but its children are not. A message like "check [this PR](url) please" produces "check \n[this PR](url)\n please" with spurious newlines splitting a single sentence. The existing tests don't catch this because they use toContain rather than exact matching. This directly impacts the core goal of the PR — every message with a hyperlink or any inline formatting will have corrupted text sent to the model.
Reviewed by Cursor Bugbot for commit b73c604. Configure here.
1 participant
When a user sends a message with Slack hyperlinks (
<url|display text>), the URLs are stripped before reaching the model. The adapter'sextractPlainText()converts the AST to plain text, discarding link URLs entirely — the assistant only sees the display text with no knowledge of embedded URLs (e.g., PR links, thread links).This switches from
message.text(plain text) tomessage.formatted(the AST) and walks the tree with a newextractTextPreservingLinks()helper that renders links as[text](url)while keeping all other content as unescaped plain text. This preserves URLs for the model without introducing markdown escape artifacts like\_.Three ingress points updated:
slack-runtime.ts— subscribed message handlerreply-executor.ts— direct mention handlerconversation-memory.ts— thread backfillAlso updates
stripLeadingBotMentionto handle the@USER_IDformat produced by the AST — the existing<@USER_ID>regex was dead code since the adapter'stoAst()already strips angle brackets from mention tokens.Fixes #137