openclaw - 💡(How to fix) Fix Slack/Telegram delivery layer: posts raw errorMessage verbatim and concatenates multiple text content items from assistant messages [1 participants]

Summary

OpenClaw's chat delivery layer (Slack verified; Telegram suspected same code path) posts raw assistant message content without sanitization in two cases, leading to user-visible confusion:

1. Raw API errorMessage surfacing. When an assistant message has stopReason=error and the provider populates errorMessage with raw upstream API error text (e.g., an OpenAI 500 with a request ID and a help.openai.com link), OpenClaw posts that errorMessage verbatim to the configured delivery channel.

2. Multiple text content items concatenated. When an assistant message contains more than one text content item (observed after parallel tool calls with the gpt-5.4 / openai-codex-responses provider — the model emits one text item per parallel tool branch), OpenClaw concatenates all text items into a single chat post instead of selecting one.

Both are in the same code path: the delivery layer doesn't sanitize assistant message content before posting to the user-facing channel.

Environment

• OpenClaw: 2026.4.14
• Provider: openai-codex (using api: openai-codex-responses)
• Model: gpt-5.4
• Delivery channel verified: Slack (same code path likely for Telegram; untested)

Evidence — Bug 1 (errorMessage surfacing)

A scheduled cron fire hit an OpenAI 500. The assistant message written to the session store had:

stopReason: error
content: []
errorMessage: "The server had an error processing your request. Sorry about that!
               You can retry your request, or contact us through our help center at
               help.openai.com if you keep seeing this error. (Please include the
               request ID <uuid> in your email.)"

The delivery layer posted the errorMessage text verbatim to the end user's Slack DM.

Evidence — Bug 2 (multiple text content items)

An interactive session, preceded by an assistant turn with two parallel tool calls, produced one assistant message containing two text content items:

content:
  - type: text
    text: "<single paragraph answer to the user's question, variant 1>"  (~120 chars)
  - type: text
    text: "<same paragraph answer, slightly different wording>"           (~136 chars)
stopReason: stop

Slack delivery concatenated both items into one post, visible to the user as a near-duplicate paragraph with slight wording variation between the two halves. The two content items appear to be emitted one per parallel tool branch — a model behavior we can't rely on not happening.

Expected behavior

For stopReason=error:

• Suppress the message entirely (admin-log server-side only), OR
• Substitute a generic user-facing fallback (e.g., "Temporary issue — will retry shortly."), OR
• Retry with backoff before posting
• Never post the raw errorMessage verbatim to end users

For multi-text-item messages:

• Select one text item to post (most useful: the last non-empty), OR
• Expose a config flag: pick-last | pick-longest | concat (current) | separate-messages
• concat should not be the default

Proposed fix surface

Sanitization step in the delivery path, before posting. Rough sketch:

function buildPostText(msg: AssistantMessage): string | null {
  if (msg.stopReason === 'error') return GENERIC_ERROR_FALLBACK; // or null to suppress
  const texts = (msg.content ?? []).filter(c => c.type === 'text' && c.text?.trim());
  if (texts.length === 0) return null;
  if (texts.length === 1) return texts[0].text;
  return texts[texts.length - 1].text; // pick-last; behind a config flag would be even better
}

Why bundle them

Same code path. Same class of issue. Same sanitization function covers both. Addressing them together gives one review surface and one deploy.

Related / separate

Retry-with-backoff on transient provider errors is a separate/larger feature — not requested here; this issue focuses on the delivery-layer sanitization only.

Offer

Happy to test a PR against our production deployment and provide additional evidence (scrubbed of user data) if useful.