Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
After upgrading from v2026.6.10 to v2026.6.11, all tool results (exec, read, web_fetch, cron, session_status, canvas) are delivered to the model as image content blocks instead of text when i m viewing in the openclaw native chat ,
making tool use non-functional in Discord channel sessions.
Steps to reproduce
1 - Run OpenClaw v2026.6.11 with a Discord channel session
2- Send any agent message that triggers a tool call (exec, read, web_fetch, cron)
3- Observe that every tool result returns (see attached image) instead of text output
4- Confirm all tool types are equally affected
Expected behavior
In v2026.6.10 and earlier versions , all tool results were delivered as {type:"text"} content blocks and the model could process them programmatically.
Actual behavior
Every tool result is delivered as an image content block. The model receives (see attached image) for every tool call. Confirmed via live tool testing — exec, read, web_fetch, cron, session_status, canvas all affected equally. Even message(action="send") returns an image response.
OpenClaw version
v2026.6.11
Operating system
Ubuntu (Zeabur VPS, Singapore)
Install method
Zeabur managed deployment
Model
deepseek/deepseek-v4-pro
Provider / routing chain
openclaw → deepseek api → deepseek/deepseek-v4-pro
Additional provider/model setup details
Agent: chief-of-staff (CoS) running in Discord channel session (#chief-of-staff). Issue is session-wide, not agent-specific or model-specific. Fresh session sometimes resolves intermittently but is not reliable.
Logs
From transport layer inspection (transport-stream-B5BPwLDf.js line 335): tool results are being converted to image content blocks in the rendering/delivery layer before reaching the model. The exec tool itself only produces {type:"text"} blocks — the conversion happens downstream.
Suspected vector: commit 86ea382 ("preserve progress preview final edits") — the progress preview compositor may be screenshotting/normalizing tool outputs as part of the draft flow. Potentially related PRs in this release: #89911 (Discord reply threading), #94072 (message-tool success reporting), #93002 (Telegram progress-mode).Screenshots, recordings, and evidence
No response
Impact and severity
Affected: Discord channel sessions on v2026.6.11
Severity: High — blocks all tool use; agent cannot execute pipeline work
Frequency: Consistent within affected sessions; intermittently resolves on fresh session start
Consequence: Agent cannot read files, run scripts, check cron jobs, or perform any tool-dependent work
Additional information
Last known good version: v2026.6.10. First known bad version: v2026.6.11. Temporary workaround: start a fresh Discord session — resolves intermittently but not reliably. Issue does not affect non-Discord session types.
Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
After upgrading from v2026.6.10 to v2026.6.11, all tool results (exec, read, web_fetch, cron, session_status, canvas) are delivered to the model as image content blocks instead of text when i m viewing in the openclaw native chat ,
making tool use non-functional in Discord channel sessions.
Steps to reproduce
1 - Run OpenClaw v2026.6.11 with a Discord channel session
2- Send any agent message that triggers a tool call (exec, read, web_fetch, cron)
3- Observe that every tool result returns (see attached image) instead of text output
4- Confirm all tool types are equally affected
Expected behavior
In v2026.6.10 and earlier versions , all tool results were delivered as {type:"text"} content blocks and the model could process them programmatically.
Actual behavior
Every tool result is delivered as an image content block. The model receives (see attached image) for every tool call. Confirmed via live tool testing — exec, read, web_fetch, cron, session_status, canvas all affected equally. Even message(action="send") returns an image response.
OpenClaw version
v2026.6.11
Operating system
Ubuntu (Zeabur VPS, Singapore)
Install method
Zeabur managed deployment
Model
deepseek/deepseek-v4-pro
Provider / routing chain
openclaw → deepseek api → deepseek/deepseek-v4-pro
Additional provider/model setup details
Agent: chief-of-staff (CoS) running in Discord channel session (#chief-of-staff). Issue is session-wide, not agent-specific or model-specific. Fresh session sometimes resolves intermittently but is not reliable.
Logs
Screenshots, recordings, and evidence
No response
Impact and severity
Affected: Discord channel sessions on v2026.6.11
Severity: High — blocks all tool use; agent cannot execute pipeline work
Frequency: Consistent within affected sessions; intermittently resolves on fresh session start
Consequence: Agent cannot read files, run scripts, check cron jobs, or perform any tool-dependent work
Additional information
Last known good version: v2026.6.10. First known bad version: v2026.6.11. Temporary workaround: start a fresh Discord session — resolves intermittently but not reliably. Issue does not affect non-Discord session types.