repo stringclasses 1
value | github_id int64 3.77B 4.38B | github_node_id stringlengths 18 24 | number int64 75 77.8k | html_url stringlengths 46 49 | api_url stringlengths 56 59 | title stringlengths 1 746 | body stringlengths 1 55.7k ⌀ | state stringclasses 2
values | state_reason stringclasses 4
values | locked bool 2
classes | comments_count int64 0 128 | labels listlengths 0 5 | assignees listlengths 0 2 | created_at stringdate 2026-01-01 16:55:34 2026-05-05 11:59:03 | updated_at stringdate 2026-03-24 00:13:29 2026-05-05 11:59:15 | closed_at stringlengths 20 20 ⌀ | author_association stringclasses 3
values | milestone_title stringclasses 0
values | snapshot_id stringclasses 76
values | extracted_at stringdate 2026-03-24 23:36:49 2026-05-05 12:00:33 | author_login stringlengths 3 38 | author_id int64 346 282M | author_node_id stringlengths 12 20 | author_type stringclasses 2
values | author_site_admin bool 2
classes |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
openclaw/openclaw | 4,221,318,685 | I_kwDOQb6kR877nDYd | 62,778 | https://github.com/openclaw/openclaw/issues/62778 | https://api.github.com/repos/openclaw/openclaw/issues/62778 | Anthropic thinking block signature corruption after cache-TTL context pruning | ## Summary
Sessions with `thinkingLevel` enabled (medium/high) become permanently stuck (400 errors) when cache-TTL context pruning fires between turns. The thinking block's cryptographic signature gets corrupted during the pruning/serialization pipeline, and Anthropic's API rejects all subsequent requests.
## Environ... | closed | completed | true | 2 | [] | [] | 2026-04-08T00:21:57Z | 2026-04-28T05:05:22Z | 2026-04-25T14:44:32Z | NONE | null | 20260428T060020Z | 2026-04-28T06:00:20Z | lexsy-cto | 264,235,225 | U_kgDOD7_o2Q | User | false |
openclaw/openclaw | 4,004,758,076 | I_kwDOQb6kR87us8I8 | 29,809 | https://github.com/openclaw/openclaw/issues/29809 | https://api.github.com/repos/openclaw/openclaw/issues/29809 | [Bug]: origin not allowed (open the Control UI from the gateway host or allow it in gateway.controlUi.allowedOrigins) | ### Summary
### Description
I deployed the Docker version of OpenClaw on a cloud server and opened port `18789` in the firewall. I used FRP for port forwarding to expose the service to the public network.
Although I can successfully open the web page at `http://<my-public-ip>:18789`, I continuously receive the follow... | closed | completed | true | 15 | [
"bug",
"stale"
] | [] | 2026-02-28T14:34:18Z | 2026-04-28T04:30:36Z | 2026-04-25T05:49:24Z | NONE | null | 20260428T060020Z | 2026-04-28T06:00:20Z | Crescentz | 35,165,563 | MDQ6VXNlcjM1MTY1NTYz | User | false |
openclaw/openclaw | 4,185,983,529 | I_kwDOQb6kR875gQop | 58,957 | https://github.com/openclaw/openclaw/issues/58957 | https://api.github.com/repos/openclaw/openclaw/issues/58957 | Model switch can fail silently when carried-over session context is too large | ## Summary
When switching models or continuing in a long-running session, OpenClaw can appear to fail silently when the carried-over context is too large. There is no clear error message indicating that the prompt/context window was exceeded, so users cannot tell whether the problem is caused by context overflow, model... | open | null | false | 6 | [] | [] | 2026-04-01T10:52:53Z | 2026-04-30T23:22:26Z | null | NONE | null | 20260501T113057Z | 2026-05-01T11:30:57Z | LunaLee0130 | 258,261,715 | U_kgDOD2TC0w | User | false |
openclaw/openclaw | 4,202,819,242 | I_kwDOQb6kR876ge6q | 60,612 | https://github.com/openclaw/openclaw/issues/60612 | https://api.github.com/repos/openclaw/openclaw/issues/60612 | Doctor warns about NVM node instead of Homebrew/system node, but cannot be fixed | ## Problem
OpenClaw doctor warns about Gateway using NVM node instead of Homebrew/system node, but the warning cannot be resolved because:
1. OpenClaw regenerates the launchd plist file on each restart
2. The plist always uses NVM path (`~/.nvm/versions/node/v24.14.1/bin/node`)
3. Manually changing the plist to Homeb... | open | null | false | 2 | [] | [] | 2026-04-04T00:55:11Z | 2026-05-02T09:00:03Z | null | NONE | null | 20260502T090030Z | 2026-05-02T09:00:30Z | drzeast-png | 264,975,099 | U_kgDOD8sy-w | User | false |
openclaw/openclaw | 4,132,640,547 | I_kwDOQb6kR872Uxcj | 54,253 | https://github.com/openclaw/openclaw/issues/54253 | https://api.github.com/repos/openclaw/openclaw/issues/54253 | [Bug]: OpenClaw returns "run Error : LLM Request Failed" on RISC-V64 System. | ### Bug type
Behavior bug (incorrect output/state without crash)
### Summary
Hello Team,
I'm quite excited to have tried using OpenClaw on my Mac & x86 Systems and have had great experiences. However, I'm working on Installing OpenClaw on my RISC-V64 System. And i'm successfully able to install and setup without a... | open | null | false | 10 | [
"bug",
"bug:behavior"
] | [] | 2026-03-25T05:27:22Z | 2026-05-04T21:19:42Z | null | NONE | null | 20260505T000034Z | 2026-05-05T00:00:34Z | iravikiran | 31,534,896 | MDQ6VXNlcjMxNTM0ODk2 | User | false |
openclaw/openclaw | 4,226,265,800 | I_kwDOQb6kR87757LI | 63,265 | https://github.com/openclaw/openclaw/issues/63265 | https://api.github.com/repos/openclaw/openclaw/issues/63265 | [Bug]: `openclaw doctor --fix` creates minimal config at /root/.openclaw/ that silently overrides real config, breaking elevated permissions | ## Bug type
Behavior bug
## Summary
Running `openclaw doctor --fix` as root (e.g. via SSH on a systemd-managed gateway with `Environment=HOME=/home/node`) creates `/root/.openclaw/openclaw.json` with only `gateway.auth` + `wizard` + `meta`. This minimal config silently takes precedence over the actual config at `/ho... | open | null | false | 3 | [] | [] | 2026-04-08T17:09:00Z | 2026-05-02T23:42:34Z | null | NONE | null | 20260503T000641Z | 2026-05-03T00:06:41Z | thesamgodson | 89,787,877 | MDQ6VXNlcjg5Nzg3ODc3 | User | false |
openclaw/openclaw | 4,201,087,774 | I_kwDOQb6kR876Z4Me | 60,389 | https://github.com/openclaw/openclaw/issues/60389 | https://api.github.com/repos/openclaw/openclaw/issues/60389 | Persistent thread-bound ACP sessions retain delivery context after end, hijacking unrelated agent chats | ## Bug
When a thread-bound ACP session (`mode: 'persistent'`) ends, its `deliveryContext` and `origin` fields are preserved in `sessions.json` indefinitely. On every gateway restart, the gateway regenerates a stub entry for that session key — including the original delivery context. This causes all subsequent messages... | closed | completed | true | 3 | [] | [] | 2026-04-03T15:37:29Z | 2026-04-29T04:39:34Z | 2026-04-26T09:18:29Z | NONE | null | 20260429T060025Z | 2026-04-29T06:00:25Z | CarsonGreenland | 3,248,403 | MDQ6VXNlcjMyNDg0MDM= | User | false |
openclaw/openclaw | 4,069,967,254 | I_kwDOQb6kR87ylsWW | 44,923 | https://github.com/openclaw/openclaw/issues/44923 | https://api.github.com/repos/openclaw/openclaw/issues/44923 | [Bug]: Telegram messages silently deleted after streaming/draft — user loses content permanently (repeated pattern) | ## Environment
- **Version:** OpenClaw 2026.3.11 (29dc654)
- **OS:** Ubuntu 24.04, Telegram forum bot
- **Channel:** Telegram (forum mode with topics)
## Describe the bug
Messages sent to Telegram are **silently deleted** after being posted. This happens repeatedly in different scenarios:
1. **Status messages** with... | closed | completed | true | 2 | [] | [] | 2026-03-13T09:29:13Z | 2026-04-27T04:46:57Z | 2026-04-24T23:54:52Z | NONE | null | 20260427T090020Z | 2026-04-27T09:00:20Z | IIIyban | 137,302,231 | U_kgDOCC8Q1w | User | false |
openclaw/openclaw | 4,042,398,730 | I_kwDOQb6kR87w8hwK | 40,343 | https://github.com/openclaw/openclaw/issues/40343 | https://api.github.com/repos/openclaw/openclaw/issues/40343 | Auto-compaction can delete an already-delivered message on Telegram | ## Description
When auto-compaction triggers while a reply is being streamed/delivered to Telegram, the message that was already visible in the chat gets deleted. The user briefly sees the full reply, then it vanishes and is replaced by the post-compaction verbose notice.
## Steps to Reproduce
1. Have a long convers... | closed | completed | true | 3 | [] | [] | 2026-03-08T23:09:52Z | 2026-04-27T04:58:01Z | 2026-04-24T20:46:12Z | CONTRIBUTOR | null | 20260427T090020Z | 2026-04-27T09:00:20Z | jakepresent | 31,285,242 | MDQ6VXNlcjMxMjg1MjQy | User | false |
openclaw/openclaw | 4,166,357,320 | I_kwDOQb6kR874VZFI | 57,386 | https://github.com/openclaw/openclaw/issues/57386 | https://api.github.com/repos/openclaw/openclaw/issues/57386 | [Bug]: Google Chat webhook returns 401 Unauthorized - all versions 2026.3.23+ | ### Bug type
Regression (worked before, now fails)
### Beta release blocker
No
### Summary
Google Chat inbound webhooks consistently return 401 Unauthorized despite correct configuration; Google Cloud Logs confirm webhooks reach the endpoint but OpenClaw rejects all requests before JWT verification across versions... | closed | not_planned | true | 3 | [
"bug",
"regression"
] | [] | 2026-03-30T01:21:27Z | 2026-04-29T04:40:35Z | 2026-04-26T08:48:36Z | NONE | null | 20260429T060025Z | 2026-04-29T06:00:25Z | kristine-jpg | 272,073,704 | U_kgDOEDeD6A | User | false |
openclaw/openclaw | 4,203,179,674 | I_kwDOQb6kR876h26a | 60,656 | https://github.com/openclaw/openclaw/issues/60656 | https://api.github.com/repos/openclaw/openclaw/issues/60656 | MCP stdio servers accumulate across turns and are not cleaned up on config reload (memory leak) | ## Summary
On OpenClaw `2026.4.2`, MCP stdio servers appear to accumulate across turns and are not cleaned up properly.
Observed servers:
- `@modelcontextprotocol/server-sequential-thinking`
- `@upstash/context7-mcp`
- `mcp-deepwiki` / `mcp-instruct`
This caused repeated memory growth on a 15 GiB VPS until restart.
... | closed | completed | true | 3 | [] | [] | 2026-04-04T03:04:06Z | 2026-04-30T04:44:24Z | 2026-04-25T07:41:06Z | NONE | null | 20260430T060018Z | 2026-04-30T06:00:18Z | xieyuanqing | 43,878,017 | MDQ6VXNlcjQzODc4MDE3 | User | false |
openclaw/openclaw | 4,177,018,703 | I_kwDOQb6kR874-D9P | 58,235 | https://github.com/openclaw/openclaw/issues/58235 | https://api.github.com/repos/openclaw/openclaw/issues/58235 | [Bug]: Gemini 3.1 Pro Preview via OpenAI-compatible API missing thought_signature in tool_calls | ### Bug type
Behavior bug (incorrect output/state without crash)
### Beta release blocker
No
### Summary
When using OpenClaw's custom OpenAI protocol service provider to interface with Google Gemini 3.1 Pro Preview, the returned tool_calls are missing the extra_content.google.thought_signature field, causing the G... | closed | completed | true | 2 | [
"bug",
"bug:behavior"
] | [] | 2026-03-31T08:15:23Z | 2026-04-28T05:16:34Z | 2026-04-25T12:33:08Z | NONE | null | 20260428T060020Z | 2026-04-28T06:00:20Z | pijianqiang | 140,389,868 | U_kgDOCF4t7A | User | false |
openclaw/openclaw | 4,026,950,676 | I_kwDOQb6kR87wBmQU | 35,827 | https://github.com/openclaw/openclaw/issues/35827 | https://api.github.com/repos/openclaw/openclaw/issues/35827 | [Bug] Null handling error after LLM timeout - 'Cannot convert undefined or null to object' | ## Description
When an LLM request times out (10 minutes), OpenClaw throws a JavaScript error instead of gracefully handling the null result.
## Error Log
```
[agent/embedded] embedded run timeout: runId=... timeoutMs=600000
[diagnostic] lane task error: "FailoverError: LLM request timed out."
[agent/embedded] embedde... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-05T09:19:01Z | 2026-04-25T04:18:47Z | 2026-04-22T04:31:56Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | vulturesammy | 264,801,135 | U_kgDOD8iLbw | User | false |
openclaw/openclaw | 4,026,870,494 | I_kwDOQb6kR87wBSre | 35,807 | https://github.com/openclaw/openclaw/issues/35807 | https://api.github.com/repos/openclaw/openclaw/issues/35807 | [Bug] Exec tool corrupts PowerShell pipeline variables on Windows - \ stripped, -f operator broken | ## Bug Report
**Supersedes:** #6443 (locked — not resolved. Confirmed by 2 independent users)
### Expected behavior
Exec tool passes PowerShell commands to the shell verbatim. Pipeline variables and operators work as expected.
### Observed behavior
Exec tool corrupts PowerShell commands on Windows when passed inline... | closed | completed | true | 9 | [] | [] | 2026-03-05T09:03:38Z | 2026-04-28T04:43:47Z | 2026-04-26T01:31:07Z | NONE | null | 20260428T060020Z | 2026-04-28T06:00:20Z | henrybottter | 257,275,244 | U_kgDOD1W1bA | User | false |
openclaw/openclaw | 4,026,866,535 | I_kwDOQb6kR87wBRtn | 35,805 | https://github.com/openclaw/openclaw/issues/35805 | https://api.github.com/repos/openclaw/openclaw/issues/35805 | [Bug] No governance file drift detection - stale agent context causes silent coordination failures | ## Bug Report
**Supersedes:** #7050 (locked/closed — not resolved)
### Expected behavior
Platform detects and alerts when agent governance files (SOUL.md, CONSTITUTION.md, AGENTS.md, etc.) have not been updated within a configurable threshold.
### Observed behavior
Governance files drift silently. No detection, no a... | closed | not_planned | true | 7 | [] | [] | 2026-03-05T09:02:50Z | 2026-04-27T05:03:08Z | 2026-04-24T19:23:43Z | NONE | null | 20260427T090020Z | 2026-04-27T09:00:20Z | henrybottter | 257,275,244 | U_kgDOD1W1bA | User | false |
openclaw/openclaw | 4,026,866,302 | I_kwDOQb6kR87wBRp- | 35,804 | https://github.com/openclaw/openclaw/issues/35804 | https://api.github.com/repos/openclaw/openclaw/issues/35804 | [Bug] No reproducible agent onboarding - 10+ manual files, no validation, no automation | ## Bug Report
**Supersedes:** #7049 (locked/closed — not resolved)
### Expected behavior
Onboarding a new agent is automated or at least validated: scaffold files created, required fields checked, boot sequence verified.
### Observed behavior
New agent requires manually creating/populating 10+ files (SOUL.md, CONSTI... | closed | completed | true | 6 | [] | [] | 2026-03-05T09:02:47Z | 2026-04-27T05:03:09Z | 2026-04-24T19:23:38Z | NONE | null | 20260427T090020Z | 2026-04-27T09:00:20Z | henrybottter | 257,275,244 | U_kgDOD1W1bA | User | false |
openclaw/openclaw | 4,026,864,027 | I_kwDOQb6kR87wBRGb | 35,802 | https://github.com/openclaw/openclaw/issues/35802 | https://api.github.com/repos/openclaw/openclaw/issues/35802 | [Bug] No centralized agent registry - orchestrator blind to running agents at scale | ## Bug Report
**Supersedes:** #7048 (locked/closed by stale-bot — not resolved)
### Expected behavior
Orchestrator can query active agent sessions, their status, model, and resource usage via a first-class API or built-in tool.
### Observed behavior
No such mechanism exists. \sessions_list\ returns sessions but not ... | closed | completed | true | 6 | [] | [] | 2026-03-05T09:02:21Z | 2026-04-27T05:03:10Z | 2026-04-24T19:23:34Z | NONE | null | 20260427T090020Z | 2026-04-27T09:00:20Z | henrybottter | 257,275,244 | U_kgDOD1W1bA | User | false |
openclaw/openclaw | 4,026,827,935 | I_kwDOQb6kR87wBISf | 35,796 | https://github.com/openclaw/openclaw/issues/35796 | https://api.github.com/repos/openclaw/openclaw/issues/35796 | [Bug]: Windows node onboarding: no --token flag, config overwritten on restart, confusing devices/nodes split | ### Bug type
Behavior bug (incorrect output/state without crash)
### Summary
Setting up a Windows node and connecting it to a gateway is unnecessarily confusing. The current flow requires undocumented environment variables, has config files silently overwritten on restart, and splits pairing across two different CLI... | closed | not_planned | true | 2 | [
"bug",
"stale",
"bug:behavior"
] | [] | 2026-03-05T08:55:12Z | 2026-04-25T04:18:48Z | 2026-04-22T04:31:50Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | BenediktSchackenberg | 6,381,261 | MDQ6VXNlcjYzODEyNjE= | User | false |
openclaw/openclaw | 4,026,823,792 | I_kwDOQb6kR87wBHRw | 35,795 | https://github.com/openclaw/openclaw/issues/35795 | https://api.github.com/repos/openclaw/openclaw/issues/35795 | [Bug]: Sub-agent completion messages ignore channel messageToolHints | ### Bug type
Behavior bug (incorrect output/state without crash)
### Summary
Sub-agent completion messages are delivered as raw Markdown regardless of the channel's `messageToolHints` formatting requirements.
### Steps to reproduce
1. Configure a channel with `messageToolHints` that requests non-Markdown output (e... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-05T08:54:22Z | 2026-04-25T04:18:48Z | 2026-04-22T04:31:49Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | frankdierolf | 207,307,187 | U_kgDODFtBsw | User | false |
openclaw/openclaw | 4,026,816,808 | I_kwDOQb6kR87wBFko | 35,793 | https://github.com/openclaw/openclaw/issues/35793 | https://api.github.com/repos/openclaw/openclaw/issues/35793 | [Bug]: Read not my files in my workdir. | ### Bug type
Regression (worked before, now fails)
### Summary
Latest version installed. I cannot import files into the workspace. I am told that I should provide the file or change to the relevant directory. I have changed to the work directory with the files to be imported several times and opened OpenClaw. Same m... | closed | not_planned | true | 2 | [
"bug",
"stale",
"regression"
] | [] | 2026-03-05T08:52:59Z | 2026-04-25T04:18:49Z | 2026-04-22T04:31:47Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | TomWest61 | 132,100,682 | U_kgDOB9-ySg | User | false |
openclaw/openclaw | 4,026,749,159 | I_kwDOQb6kR87wA1Dn | 35,773 | https://github.com/openclaw/openclaw/issues/35773 | https://api.github.com/repos/openclaw/openclaw/issues/35773 | Out of memory: Process killed with no error message | ## Description
OpenClaw fails to start due to OOM with no helpful error message.
## System Logs
[Thu Mar 5 08:23:05 2026] Out of memory: Killed process 101493 (openclaw-gatewa)
[Thu Mar 5 08:24:55 2026] Out of memory: Killed process 101553 (openclaw)
[Thu Mar 5 08:26:03 2026] Out of memory: Killed process 101696 (op... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T08:38:59Z | 2026-04-25T04:18:50Z | 2026-04-22T04:31:45Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | liangchaoboy | 17,689,793 | MDQ6VXNlcjE3Njg5Nzkz | User | false |
openclaw/openclaw | 4,026,734,162 | I_kwDOQb6kR87wAxZS | 35,768 | https://github.com/openclaw/openclaw/issues/35768 | https://api.github.com/repos/openclaw/openclaw/issues/35768 | [Bug]: Homebrew version shows "dev" - requires OPENCLAW_VERSION env var | ### Bug type
Regression (worked before, now fails)
### Summary
After upgrading via Homebrew, openclaw --version displays "dev" instead of actual version number unless OPENCLAW_VERSION environment variable is manually set.
### Steps to reproduce
1. Install/upgrade OpenClaw via Homebrew:
brew upgrade openclaw-cli
... | closed | not_planned | true | 2 | [
"bug",
"stale",
"regression"
] | [] | 2026-03-05T08:36:03Z | 2026-04-25T04:18:50Z | 2026-04-22T04:31:43Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | hdot123 | 254,672,097 | U_kgDODy384Q | User | false |
openclaw/openclaw | 4,026,714,629 | I_kwDOQb6kR87wAsoF | 35,764 | https://github.com/openclaw/openclaw/issues/35764 | https://api.github.com/repos/openclaw/openclaw/issues/35764 | Feature: Cross-channel mention forwarding — notify designated channel when agent is @mentioned | ## Feature Request
### Problem
When agent A sends a message in agent B's channel and @mentions them, agent B responds in their own channel. Agent A (and the human operator) has no way to see that response without manually checking every agent's channel.
Currently the only workaround is prompt-based routing rules (ag... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T08:32:22Z | 2026-04-25T04:18:51Z | 2026-04-22T04:31:41Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | ToneLoke | 10,687,151 | MDQ6VXNlcjEwNjg3MTUx | User | false |
openclaw/openclaw | 4,026,710,427 | I_kwDOQb6kR87wArmb | 35,763 | https://github.com/openclaw/openclaw/issues/35763 | https://api.github.com/repos/openclaw/openclaw/issues/35763 | [Bug]: Device pairing required on loopback despite docs saying auto-approved | ### Bug type
Regression (worked before, now fails)
### Summary
Device pairing required on loopback despite docs saying auto-approved
### Steps to reproduce
1. Run OpenClaw in Docker with gateway.bind = "loopback"
2. Configure gateway.auth.mode = "token"
3. Send an inbound WhatsApp message to trigger an agent tool ... | closed | not_planned | true | 2 | [
"bug",
"stale",
"regression"
] | [] | 2026-03-05T08:31:24Z | 2026-04-25T04:18:52Z | 2026-04-22T04:31:40Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | Oliversums | 181,245,463 | U_kgDOCs2WFw | User | false |
openclaw/openclaw | 4,026,501,695 | I_kwDOQb6kR87v_4o_ | 35,717 | https://github.com/openclaw/openclaw/issues/35717 | https://api.github.com/repos/openclaw/openclaw/issues/35717 | Improve sessions_send cross-agent visibility error guidance | ## Summary
When `sessions_send` targets another agent session and cross-agent visibility is still restricted, the error only points to `tools.sessions.visibility=all`.
In practice, cross-agent sends also require `tools.agentToAgent.enabled=true`, so users often fix one setting, retry, and hit a second error.
## Impro... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T07:50:02Z | 2026-04-25T04:18:53Z | 2026-04-22T04:31:34Z | CONTRIBUTOR | null | 20260425T060021Z | 2026-04-25T06:00:21Z | shuofengzhang | 24,763,026 | MDQ6VXNlcjI0NzYzMDI2 | User | false |
openclaw/openclaw | 4,026,471,919 | I_kwDOQb6kR87v_xXv | 35,707 | https://github.com/openclaw/openclaw/issues/35707 | https://api.github.com/repos/openclaw/openclaw/issues/35707 | Feature Request: read tool should return image content (base64 or binary) | ## Current Behavior
When using the `read` tool to read image files (jpg, png, gif, webp), it only returns the file type/mime type without the actual image content.
## Expected Behavior
The `read` tool should return the actual image content, either as:
- Base64-encoded string
- Binary data
- Or add a new parameter like... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-05T07:43:20Z | 2026-04-25T04:18:53Z | 2026-04-22T04:31:32Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | eternaling | 73,522,556 | MDQ6VXNlcjczNTIyNTU2 | User | false |
openclaw/openclaw | 4,026,347,210 | I_kwDOQb6kR87v_S7K | 35,656 | https://github.com/openclaw/openclaw/issues/35656 | https://api.github.com/repos/openclaw/openclaw/issues/35656 | [Bug]: Feishu logs refresh frequently | ### Bug type
Regression (worked before, now fails)
### Summary
Config warnings:\n- plugins.entries.feishu: plugin feishu: duplicate plugin id detected; later plugin may be overridden (/root/.openclaw/extensions/feishu/index.ts)
07:10:53 warn Config warnings:\n- plugins.entries.feishu: plugin feishu: duplicate plugin... | closed | not_planned | true | 2 | [
"bug",
"stale",
"regression"
] | [] | 2026-03-05T07:15:27Z | 2026-04-25T04:18:56Z | 2026-04-22T04:31:30Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | 495188853 | 48,780,120 | MDQ6VXNlcjQ4NzgwMTIw | User | false |
openclaw/openclaw | 4,026,345,576 | I_kwDOQb6kR87v_Sho | 35,655 | https://github.com/openclaw/openclaw/issues/35655 | https://api.github.com/repos/openclaw/openclaw/issues/35655 | Failed delivery notifications leak into group chats (Telegram) | ## Bug Description
When a message send fails (e.g. message too long, format issue, or deleted message reference), OpenClaw displays a notification like:
```
⚠️ ✉️ Message: 2201 failed
```
**Problem:** This notification appears in the group chat itself (visible to all members), not just in the admin/owner DM session.... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-05T07:15:05Z | 2026-04-25T04:18:57Z | 2026-04-22T04:31:28Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | balthazar-bot | 258,649,781 | U_kgDOD2qutQ | User | false |
openclaw/openclaw | 4,026,315,046 | I_kwDOQb6kR87v_LEm | 35,647 | https://github.com/openclaw/openclaw/issues/35647 | https://api.github.com/repos/openclaw/openclaw/issues/35647 | [Feature]: web_search provider fallback chain (auto-switch on quota/rate-limit errors) | ### Summary
Add a provider fallback chain to `web_search` so that when the primary provider's free quota is exhausted (HTTP 429/402), the tool automatically retries with the next available provider instead of failing entirely.
### Problem to solve
Currently `web_search` binds to a single provider at startup (`resolv... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-05T07:08:20Z | 2026-04-25T04:18:57Z | 2026-04-22T04:31:27Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | cathyjay | 114,842,176 | U_kgDOBthaQA | User | false |
openclaw/openclaw | 4,026,314,404 | I_kwDOQb6kR87v_K6k | 35,646 | https://github.com/openclaw/openclaw/issues/35646 | https://api.github.com/repos/openclaw/openclaw/issues/35646 | [Bug] Session Management: sessions_spawn run-mode cleanup and sessions_send visibility restrictions | # Bug Report: Session Management Issues - sessions_spawn cleanup and sessions_send visibility
## Describe the bug
Two related session management issues discovered during multi-agent coordination:
### Issue 1: sessions_spawn run mode sessions not cleaned from sessions_list
**Expected behavior:**
- When using `sessio... | closed | not_planned | true | 6 | [
"stale"
] | [] | 2026-03-05T07:08:12Z | 2026-04-25T04:18:58Z | 2026-04-22T04:31:25Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | OliverVin | 20,222,136 | MDQ6VXNlcjIwMjIyMTM2 | User | false |
openclaw/openclaw | 4,026,269,570 | I_kwDOQb6kR87v-_-C | 35,632 | https://github.com/openclaw/openclaw/issues/35632 | https://api.github.com/repos/openclaw/openclaw/issues/35632 | [Bug]: [[reply_to_current]] reply tag shows "Deleted message" in Telegram | ## Description
When the agent uses `[[reply_to_current]]` reply tag in Telegram:
1. The message is sent with Quote and Reply correctly
2. **But the Quoted Block shows "Deleted message"** instead of the original sender's message
3. After a short time, the Quoted Block disappears entirely
## Root Cause
The issue is... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-05T06:57:45Z | 2026-04-26T04:31:26Z | 2026-04-23T04:23:03Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | JDDevHub | 130,726,543 | U_kgDOB8q6jw | User | false |
openclaw/openclaw | 4,026,251,145 | I_kwDOQb6kR87v-7eJ | 35,624 | https://github.com/openclaw/openclaw/issues/35624 | https://api.github.com/repos/openclaw/openclaw/issues/35624 | BlueBubbles: chat_guid normalization loses iMessage service info, causing SMS fallback | # BlueBubbles: chat_guid normalization loses iMessage service info, causing SMS fallback
## Environment
- **OpenClaw version**: 2026.3.2
- **BlueBubbles server**: macOS 15.7.4, Private API enabled
- **Channel**: BlueBubbles (default account)
## Problem
When sending proactive messages to a contact that has **both** i... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-05T06:53:28Z | 2026-04-22T04:30:39Z | 2026-04-19T16:16:43Z | NONE | null | 20260422T060039Z | 2026-04-22T06:00:39Z | rainlee | 2,148,148 | MDQ6VXNlcjIxNDgxNDg= | User | false |
openclaw/openclaw | 4,026,250,350 | I_kwDOQb6kR87v-7Ru | 35,623 | https://github.com/openclaw/openclaw/issues/35623 | https://api.github.com/repos/openclaw/openclaw/issues/35623 | BlueBubbles: chat_guid normalization loses iMessage service info, causing SMS fallback | # BlueBubbles: chat_guid normalization loses iMessage service info, causing SMS fallback
## Environment
- **OpenClaw version**: 2026.3.2
- **BlueBubbles server**: macOS 15.7.4, Private API enabled
- **Channel**: BlueBubbles (default account)
## Problem
When sending proactive messages to a contact that has **both** i... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-05T06:53:18Z | 2026-04-22T04:30:40Z | 2026-04-19T16:16:41Z | NONE | null | 20260422T060039Z | 2026-04-22T06:00:39Z | rainlee | 2,148,148 | MDQ6VXNlcjIxNDgxNDg= | User | false |
openclaw/openclaw | 4,026,239,366 | I_kwDOQb6kR87v-4mG | 35,617 | https://github.com/openclaw/openclaw/issues/35617 | https://api.github.com/repos/openclaw/openclaw/issues/35617 | Bug: MiniMax Integration Fails Due to Hardcoded -chat Suffix in openai-completions Adapter | ### Bug Report: MiniMax Integration Fails Due to Hardcoded `-chat` Suffix in `openai-completions` Adapter
**System Information:**
- **OpenClaw Version:** (Please insert version if known)
- **Host OS:** macOS 25.3.0 (arm64)
- **Provider:** MiniMax (`https://api.minimax.io/v1`)
**Describe the Bug:**
When configuring Mi... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-05T06:50:48Z | 2026-04-26T04:31:26Z | 2026-04-23T04:23:02Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | topsecretM | 265,588,714 | U_kgDOD9SP6g | User | false |
openclaw/openclaw | 4,026,223,299 | I_kwDOQb6kR87v-0rD | 35,609 | https://github.com/openclaw/openclaw/issues/35609 | https://api.github.com/repos/openclaw/openclaw/issues/35609 | 'openclaw status' command hangs after outputting content (v2026.3.2) | ## Bug Description
The `openclaw status` command outputs all content successfully but **never exits**, hanging indefinitely until manually terminated (Ctrl+C or kill).
## Environment
- **OpenClaw Version:** 2026.3.2 (latest)
- **OS:** macOS 26.3 (arm64)
- **Node.js:** 22.22.0
- **Install Method:** npm global (via ho... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-05T06:47:39Z | 2026-04-26T04:31:27Z | 2026-04-23T04:23:00Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | TubeLiu | 14,797,490 | MDQ6VXNlcjE0Nzk3NDkw | User | false |
openclaw/openclaw | 4,026,199,422 | I_kwDOQb6kR87v-u1- | 35,598 | https://github.com/openclaw/openclaw/issues/35598 | https://api.github.com/repos/openclaw/openclaw/issues/35598 | Feishu outbound adapter ignores threadId — ACP replies land in group root instead of topic | ## Bug Description
When an ACP agent (e.g., Codex) completes a task dispatched from a **Feishu group topic session**, the reply is sent to the **group chat root** instead of the originating **topic/thread**. The message appears in the Lark app's group view but not in the topic where the conversation originated.
## Ro... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-05T06:42:09Z | 2026-04-26T04:31:28Z | 2026-04-23T04:22:58Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | t2wei | 65,210,143 | MDQ6VXNlcjY1MjEwMTQz | User | false |
openclaw/openclaw | 4,026,189,345 | I_kwDOQb6kR87v-sYh | 35,595 | https://github.com/openclaw/openclaw/issues/35595 | https://api.github.com/repos/openclaw/openclaw/issues/35595 | Feature Request: Pre-flight data gate — block agent session before start if required files are missing | ## Problem
Currently, Zero-Trust architectures built on OpenClaw rely on LLM agents to self-enforce data preconditions (e.g., checking if a pre-fetched data file exists before running analysis). This is still **AI self-discipline**, not **system enforcement**.
Example pattern we use today (still prompt-based):
```bas... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T06:39:50Z | 2026-04-26T04:31:28Z | 2026-04-23T04:22:56Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | Sven1679 | 246,955,451 | U_kgDODrg9uw | User | false |
openclaw/openclaw | 4,026,172,898 | I_kwDOQb6kR87v-oXi | 35,589 | https://github.com/openclaw/openclaw/issues/35589 | https://api.github.com/repos/openclaw/openclaw/issues/35589 | [Bug]: TUI responses delivered to iMessage when deliveryContext persists and heartbeat target is "last" | ### Bug type
Behavior bug (incorrect output/state without crash)
### Summary
When using `openclaw tui` without `--deliver`, agent responses are also sent to the iMessage channel because `deliveryContext` is persisted in the session from a prior `--deliver` session and the heartbeat with `target: "last"` continuously... | closed | not_planned | true | 2 | [
"bug",
"stale",
"bug:behavior"
] | [] | 2026-03-05T06:35:55Z | 2026-04-26T04:31:29Z | 2026-04-23T04:22:54Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | ewood70 | 26,074,458 | MDQ6VXNlcjI2MDc0NDU4 | User | false |
openclaw/openclaw | 4,026,167,952 | I_kwDOQb6kR87v-nKQ | 35,585 | https://github.com/openclaw/openclaw/issues/35585 | https://api.github.com/repos/openclaw/openclaw/issues/35585 | [Bug]: Double-click-drag to extend word selection not working in webchat composer (macOS) | ## Description
In the webchat message composer, double-clicking selects a word correctly, but **holding the second click and dragging does not extend the selection word-by-word**. This is standard macOS text editing behavior supported natively by `NSTextView`.
## Steps to Reproduce
1. Open OpenClaw desktop app (webc... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T06:34:46Z | 2026-04-26T04:31:30Z | 2026-04-23T04:22:53Z | CONTRIBUTOR | null | 20260426T060034Z | 2026-04-26T06:00:34Z | tylerbittner | 22,528,533 | MDQ6VXNlcjIyNTI4NTMz | User | false |
openclaw/openclaw | 4,026,139,822 | I_kwDOQb6kR87v-gSu | 35,579 | https://github.com/openclaw/openclaw/issues/35579 | https://api.github.com/repos/openclaw/openclaw/issues/35579 | [Feature Request] Add ignoreReplyToBot option for Telegram groups | ## Feature Request: Option to ignore replies to bot messages in Telegram groups
### Current Behavior
In Telegram group chats, when a user replies to a message sent by the bot, it triggers the bot to respond. This can lead to unwanted noise in group conversations where users are simply threading replies rather than act... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T06:28:54Z | 2026-04-26T04:31:30Z | 2026-04-23T04:22:51Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | elliclee | 1,717,254 | MDQ6VXNlcjE3MTcyNTQ= | User | false |
openclaw/openclaw | 4,026,050,362 | I_kwDOQb6kR87v-Kc6 | 35,556 | https://github.com/openclaw/openclaw/issues/35556 | https://api.github.com/repos/openclaw/openclaw/issues/35556 | [Bug] Feishu channel: 普通消息发送失败，只在发送卡片消息之后 | ### Bug type
Regression (worked before, now fails)
### Summary
飞书channel发送卡片消息后，再发送普通文字消息会失败
### Steps to reproduce
1. 飞书channel发送一条卡片消息
2. 立即发送一条普通文字消息
3. 普通消息发送失败（飞书端收不到）
### Expected behavior
普通消息能正常发送并被飞书收到
### Actual behavior
普通消息发送失败，飞书端收不到。重启Gateway后暂时正常，但很快又复现。
### OpenClaw version
2026.3.2
### Oper... | closed | not_planned | true | 2 | [
"bug",
"stale",
"regression"
] | [] | 2026-03-05T06:09:26Z | 2026-04-26T04:31:31Z | 2026-04-23T04:22:47Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | matingxiu | 24,728,784 | MDQ6VXNlcjI0NzI4Nzg0 | User | false |
openclaw/openclaw | 4,025,959,464 | I_kwDOQb6kR87v90Qo | 35,528 | https://github.com/openclaw/openclaw/issues/35528 | https://api.github.com/repos/openclaw/openclaw/issues/35528 | openai-codex provider: pipe-delimited tool call IDs written to transcript break Codex API validation | ## Summary
When using the `openai-codex` provider, tool call IDs are written to session transcripts as compound pipe-delimited strings (e.g. `call_LfY3urEtRi084Bpt2xnm9MjO|fc_07c5279fe4cc6c630169a4995d1e4881969b0c0467ae5301ae`).
The Codex API strictly requires IDs to match `[a-zA-Z0-9_-]+`. The pipe character (`|`) c... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-05T05:44:45Z | 2026-04-26T04:31:32Z | 2026-04-23T04:22:45Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | aurelius-lynyx | 260,155,296 | U_kgDOD4GnoA | User | false |
openclaw/openclaw | 4,025,946,130 | I_kwDOQb6kR87v9xAS | 35,523 | https://github.com/openclaw/openclaw/issues/35523 | https://api.github.com/repos/openclaw/openclaw/issues/35523 | [Bug]: TUI doesn't display assistant messages until restart | ### Bug type
Regression (worked before, now fails)
### Summary
TUI doesn't display assistant messages until restart
Description: ... | closed | not_planned | true | 4 | [
"bug",
"stale",
"regression"
] | [] | 2026-03-05T05:41:07Z | 2026-04-26T04:31:33Z | 2026-04-23T04:22:43Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | RockWu0817 | 265,571,529 | U_kgDOD9RMyQ | User | false |
openclaw/openclaw | 4,025,945,463 | I_kwDOQb6kR87v9w13 | 35,522 | https://github.com/openclaw/openclaw/issues/35522 | https://api.github.com/repos/openclaw/openclaw/issues/35522 | Bug: User message sent during compaction is silently dropped (race condition) | ## Summary
When a user sends a message at the exact moment auto-compaction is triggered, the message can be silently dropped and never processed by the agent.
## Steps to Reproduce
1. Have an active session approaching the compaction threshold
2. Send a user message at the exact moment compaction fires
3. Compaction... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-05T05:40:54Z | 2026-04-26T04:31:33Z | 2026-04-23T04:22:41Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | kfk979 | 191,075,263 | U_kgDOC2OTvw | User | false |
openclaw/openclaw | 4,025,931,889 | I_kwDOQb6kR87v9thx | 35,518 | https://github.com/openclaw/openclaw/issues/35518 | https://api.github.com/repos/openclaw/openclaw/issues/35518 | Feishu: reply should follow reply_to_id instead of thread root | ## Problem
In Feishu thread (话题) mode, when a user replies to a specific message within a thread, the bot's response always anchors to the thread root message rather than the actual message the user replied to. This breaks conversational flow and makes it hard to follow multi-turn exchanges within a thread.
## Expect... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T05:37:33Z | 2026-04-26T04:31:34Z | 2026-04-23T04:22:39Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | hadiliu | 1,926,042 | MDQ6VXNlcjE5MjYwNDI= | User | false |
openclaw/openclaw | 4,025,920,588 | I_kwDOQb6kR87v9qxM | 35,516 | https://github.com/openclaw/openclaw/issues/35516 | https://api.github.com/repos/openclaw/openclaw/issues/35516 | [Feature] Allow local models in Discord /models picker + configurable local flag override | ## Summary
The Discord `/models` slash command filters out models marked as `local: true`, with no configuration option to override this behavior. This creates issues for several valid use cases where "local" models should be selectable.
## Problem
When using `/models` in Discord, only models with `local: false` app... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T05:34:47Z | 2026-04-26T04:31:34Z | 2026-04-23T04:22:36Z | CONTRIBUTOR | null | 20260426T060034Z | 2026-04-26T06:00:34Z | synchronic1 | 135,072,919 | U_kgDOCA0Mlw | User | false |
openclaw/openclaw | 4,025,916,159 | I_kwDOQb6kR87v9pr_ | 35,515 | https://github.com/openclaw/openclaw/issues/35515 | https://api.github.com/repos/openclaw/openclaw/issues/35515 | Request: Android APK release | It would be great to have an Android APK for OpenClaw, so users can pair their Android phones as nodes for camera, location, notifications, etc.
Currently only macOS and iOS are available. Android support would enable many useful automation workflows.
Thanks! | closed | not_planned | true | 7 | [] | [] | 2026-03-05T05:33:41Z | 2026-04-28T04:32:40Z | 2026-04-26T02:41:46Z | NONE | null | 20260428T060020Z | 2026-04-28T06:00:20Z | asukaneko | 131,347,869 | U_kgDOB9Q1nQ | User | false |
openclaw/openclaw | 4,025,905,098 | I_kwDOQb6kR87v9m_K | 35,512 | https://github.com/openclaw/openclaw/issues/35512 | https://api.github.com/repos/openclaw/openclaw/issues/35512 | Add gemini-3.1-pro-preview-customtools to google-antigravity provider | ## Request
Add `gemini-3.1-pro-preview-customtools` to the google-antigravity provider's allowed model list.
This model is available on Google AI Studio as a Gemini 3.1 Pro preview with enhanced tool/function calling capabilities. Currently, the google-antigravity provider only exposes:
- gemini-3-flash
- gemini-3-pr... | closed | completed | true | 4 | [
"stale"
] | [] | 2026-03-05T05:31:15Z | 2026-04-21T04:21:48Z | 2026-04-18T18:24:08Z | NONE | null | 20260421T060038Z | 2026-04-21T06:00:38Z | ship9599 | 16,959,875 | MDQ6VXNlcjE2OTU5ODc1 | User | false |
openclaw/openclaw | 4,025,902,300 | I_kwDOQb6kR87v9mTc | 35,511 | https://github.com/openclaw/openclaw/issues/35511 | https://api.github.com/repos/openclaw/openclaw/issues/35511 | [Bug]: openclaw.json monolithic config causes cascading agent breakage — recurring issue, needs structural fix | ## Summary
Any small, targeted edit to `openclaw.json` — adding a new agent, updating a model alias, changing a config key — can silently break unrelated agents or cause the UI to show incorrect model assignments. This is a recurring problem, not a one-off.
## Related existing issues
This is not a duplicate of, but ... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T05:30:36Z | 2026-04-26T04:31:35Z | 2026-04-23T04:22:34Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | fwends | 4,113,483 | MDQ6VXNlcjQxMTM0ODM= | User | false |
openclaw/openclaw | 4,025,891,022 | I_kwDOQb6kR87v9jjO | 35,508 | https://github.com/openclaw/openclaw/issues/35508 | https://api.github.com/repos/openclaw/openclaw/issues/35508 | 🐛 [QQ Bot] TTS voice messages not sending - audioAsVoice not handled | # Bug Report: QQ Bot TTS Voice Message Not Sending
## 🐛 Bug Description
**Title**: TTS tool generates voice files but QQ Bot fails to send them as voice messages
**Version**: OpenClaw v2026.3.2
**Channel**: qqbot
## 📋 Issue Details
When using the `tts` tool to send voice messages via QQ Bot:
1. ✅ TTS tool succ... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T05:27:46Z | 2026-04-26T04:31:36Z | 2026-04-23T04:22:32Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | cndyaa-bit | 252,046,085 | U_kgDODwXrBQ | User | false |
openclaw/openclaw | 4,025,784,703 | I_kwDOQb6kR87v9Jl_ | 35,478 | https://github.com/openclaw/openclaw/issues/35478 | https://api.github.com/repos/openclaw/openclaw/issues/35478 | Feishu DM topics/threads: optionally scope sessions + allow reply-in-thread | # Feature/Gap: Feishu DM (p2p) topics/threads should optionally scope sessions + allow thread replies
## Summary
Feishu (Lark) **p2p DMs** support creating topics/threads (messages can carry `root_id` / `thread_id`). However, OpenClaw’s current Feishu channel implementation only applies **topic/thread session scoping*... | closed | not_planned | true | 5 | [
"stale"
] | [] | 2026-03-05T05:00:59Z | 2026-04-26T04:31:37Z | 2026-04-23T04:22:30Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | cathyjay | 114,842,176 | U_kgDOBthaQA | User | false |
openclaw/openclaw | 4,025,783,972 | I_kwDOQb6kR87v9Jak | 35,476 | https://github.com/openclaw/openclaw/issues/35476 | https://api.github.com/repos/openclaw/openclaw/issues/35476 | [Bug]: /model UI: multiple providers with same model id all show as selected when only one is chosen | ### Bug type
Regression (worked before, now fails)
### Summary
```markdown
### Description
When multiple providers each expose a model with the same `id` (e.g. `claude-opus-4-6` from `nexus-d`, `yunyi-claude`, `nexus-official`, `nexus-a`), the model switcher (/model list or Control UI) shows **all** of them as se... | closed | not_planned | true | 3 | [
"bug",
"stale",
"regression"
] | [] | 2026-03-05T05:00:47Z | 2026-04-26T04:31:39Z | 2026-04-23T04:22:28Z | NONE | null | 20260426T060034Z | 2026-04-26T06:00:34Z | slk1061569042-lab | 246,635,832 | U_kgDODrNdOA | User | false |
openclaw/openclaw | 4,025,780,658 | I_kwDOQb6kR87v9Imy | 35,475 | https://github.com/openclaw/openclaw/issues/35475 | https://api.github.com/repos/openclaw/openclaw/issues/35475 | [Feature Request]: Feishu channel - forward image messages to vision-capable models | ## Summary
The Feishu (Lark) channel connector currently does not download and forward image content from Feishu messages to vision-capable models. When a user sends or quotes an image in Feishu, the connector logs `fetched quoted message: [image message]` but only passes a text placeholder to the agent — the actual i... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T04:59:57Z | 2026-04-25T04:18:59Z | 2026-04-22T04:31:22Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | JecyLi | 79,399,704 | MDQ6VXNlcjc5Mzk5NzA0 | User | false |
openclaw/openclaw | 4,025,769,852 | I_kwDOQb6kR87v9F98 | 35,472 | https://github.com/openclaw/openclaw/issues/35472 | https://api.github.com/repos/openclaw/openclaw/issues/35472 | Preserve previous session history after /new (read-only access) | ## Feature Request
After `/new` or `/reset`, the previous session's message history is completely wiped. The only continuity mechanism is HANDOFF.md (manually maintained).
### Proposal
Preserve the previous session's transcript (or last N messages) as read-only, accessible via `sessions_history` with a flag like `in... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-05T04:56:46Z | 2026-04-25T04:19:00Z | 2026-04-22T04:31:20Z | NONE | null | 20260425T060021Z | 2026-04-25T06:00:21Z | dae-oss | 258,648,270 | U_kgDOD2qozg | User | false |
openclaw/openclaw | 4,024,477,462 | I_kwDOQb6kR87v4KcW | 35,023 | https://github.com/openclaw/openclaw/issues/35023 | https://api.github.com/repos/openclaw/openclaw/issues/35023 | [Bug] Sub-agent sessions fail on turn 2 when thinking is enabled — thinking blocks stripped from tool-call history replay | ### Summary
Sub-agent sessions (`sessions_spawn`) using a model with thinking enabled (e.g. `claude-sonnet-4` via github-copilot provider) fail deterministically on their second API turn (after the first tool call returns).
### Error
```
400 messages.1.content.0.type: Expected `thinking` or `redacted_thinking`, but ... | closed | completed | true | 6 | [] | [] | 2026-03-04T22:23:24Z | 2026-04-30T04:51:00Z | 2026-04-25T05:38:24Z | NONE | null | 20260430T060018Z | 2026-04-30T06:00:18Z | bradfreels | 179,389,909 | U_kgDOCrFF1Q | User | false |
openclaw/openclaw | 4,018,789,811 | I_kwDOQb6kR87vid2z | 33,508 | https://github.com/openclaw/openclaw/issues/33508 | https://api.github.com/repos/openclaw/openclaw/issues/33508 | Native /subagents list resolves requester to slash session instead of CommandTargetSessionKey | ## Summary
For native slash commands, `/subagents list` currently resolves the requester from the command session key before `CommandTargetSessionKey`.
That causes list output to show runs for the transient slash-command session (or none), instead of the routed target session.
`/subagents spawn` already prefers `Comm... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T21:28:50Z | 2026-04-19T04:23:29Z | 2026-04-16T04:28:01Z | CONTRIBUTOR | null | 20260419T052035Z | 2026-04-19T05:20:35Z | 0-CYBERDYNE-SYSTEMS-0 | 134,018,026 | U_kgDOB_zz6g | User | false |
openclaw/openclaw | 4,018,788,662 | I_kwDOQb6kR87vidk2 | 33,507 | https://github.com/openclaw/openclaw/issues/33507 | https://api.github.com/repos/openclaw/openclaw/issues/33507 | [slack] Check session freshness before skipping thread context** | **<sub><sub></sub></sub> Check session freshness before skipping thread context**
The new gate `threadInitialHistoryLimit > 0 && !threadSessionPreviousTimestamp` treats any stored `updatedAt` as an active thread session, but session reuse is actually decid... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T21:28:31Z | 2026-04-19T04:23:30Z | 2026-04-16T04:27:59Z | CONTRIBUTOR | null | 20260419T052035Z | 2026-04-19T05:20:35Z | sourman | 3,827,766 | MDQ6VXNlcjM4Mjc3NjY= | User | false |
openclaw/openclaw | 4,018,763,483 | I_kwDOQb6kR87viXbb | 33,504 | https://github.com/openclaw/openclaw/issues/33504 | https://api.github.com/repos/openclaw/openclaw/issues/33504 | WhatsApp extension plugin sendMedia doesn't forward mediaLocalRoots (agent workspace media blocked) | ## Summary
`extensions/whatsapp/src/channel.ts` line 298: `sendMedia` does not destructure or forward `mediaLocalRoots` from the outbound context. This causes `assertLocalMediaAllowed` to receive `undefined` roots, falling back to defaults that exclude the agent workspace directory.
**Result:** Any media file in the ... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T21:21:33Z | 2026-04-19T04:23:31Z | 2026-04-16T04:27:57Z | CONTRIBUTOR | null | 20260419T052035Z | 2026-04-19T05:20:35Z | odrobnik | 333,270 | MDQ6VXNlcjMzMzI3MA== | User | false |
openclaw/openclaw | 4,018,761,423 | I_kwDOQb6kR87viW7P | 33,503 | https://github.com/openclaw/openclaw/issues/33503 | https://api.github.com/repos/openclaw/openclaw/issues/33503 | Feature Request: memory:retrieved hook event + context manifest | ## Summary
Request a new `memory:retrieved` hook event (or equivalent) that exposes what the memory retrieval pipeline selected, scored, and injected into the agent context — enabling observability into context assembly.
## Problem
OpenClaw's memory retrieval pipeline (`memorySearch`, bootstrap file loading, `extraP... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T21:20:59Z | 2026-04-19T04:23:32Z | 2026-04-16T04:27:55Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | ogis-tech | 261,816,634 | U_kgDOD5sBOg | User | false |
openclaw/openclaw | 4,018,758,724 | I_kwDOQb6kR87viWRE | 33,502 | https://github.com/openclaw/openclaw/issues/33502 | https://api.github.com/repos/openclaw/openclaw/issues/33502 | Add OpenRouter model z-ai/glm-4.5-air:free | This model is available on OpenRouter free tier with 131K context and agent-specific reasoning. Please add to OpenClaw whitelist. | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T21:20:13Z | 2026-04-19T04:23:32Z | 2026-04-16T04:27:53Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | Valkster70 | 129,894,487 | U_kgDOB74IVw | User | false |
openclaw/openclaw | 4,018,602,568 | I_kwDOQb6kR87vhwJI | 33,477 | https://github.com/openclaw/openclaw/issues/33477 | https://api.github.com/repos/openclaw/openclaw/issues/33477 | Discord continuity loss: provider health-monitor restart ("reason: stuck") causes sessionId rotation + messages=0 (no gateway restart) | ## OpenClaw version
`2026.2.26`
**Host:** Raspberry Pi (socrates)
**Channel/sessionKey affected:** `agent:main:discord:channel:1478084397227053198`
## What happens
In the same Discord channel/sessionKey, conversation proceeds normally (context grows).
Then the gateway health monitor restarts the Discord provider:
`... | closed | not_planned | true | 6 | [
"stale"
] | [] | 2026-03-03T20:39:12Z | 2026-04-19T04:23:33Z | 2026-04-16T04:27:52Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | DonShelly | 25,538,402 | MDQ6VXNlcjI1NTM4NDAy | User | false |
openclaw/openclaw | 4,018,597,604 | I_kwDOQb6kR87vhu7k | 33,476 | https://github.com/openclaw/openclaw/issues/33476 | https://api.github.com/repos/openclaw/openclaw/issues/33476 | Feature: Per-turn model routing based on conversation recency/complexity | ## Feature Request
**Problem:**
Running Opus for every message is expensive, but many interactions don't need it. The first message in a thread after a period of inactivity (e.g., 3+ hours) is often a simple question or status check — Sonnet handles these fine. But once a deeper conversation develops (multi-turn, codi... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T20:37:55Z | 2026-04-19T04:23:34Z | 2026-04-16T04:27:50Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | eddieberklee | 1,836,653 | MDQ6VXNlcjE4MzY2NTM= | User | false |
openclaw/openclaw | 4,018,566,718 | I_kwDOQb6kR87vhnY- | 33,468 | https://github.com/openclaw/openclaw/issues/33468 | https://api.github.com/repos/openclaw/openclaw/issues/33468 | [Bug]: googlechat not starting with error "failed (unknown) - Cannot convert undefined or null to object" | ### Bug type
Regression (worked before, now fails)
### Summary
After upgrading to v2026.3.2 from v2026.2.24, the issue appeared
### Steps to reproduce
1. Try to configure googlechat
2. restart gateway
3. check status with command `openclaw status --deep`
### Expected behavior
```
Health
┌─────────────┬─────────... | closed | not_planned | true | 7 | [
"bug",
"stale",
"regression"
] | [] | 2026-03-03T20:30:06Z | 2026-04-19T04:23:35Z | 2026-04-16T04:27:48Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | ssabdallah | 102,990,789 | U_kgDOBiODxQ | User | false |
openclaw/openclaw | 4,018,531,394 | I_kwDOQb6kR87vhexC | 33,463 | https://github.com/openclaw/openclaw/issues/33463 | https://api.github.com/repos/openclaw/openclaw/issues/33463 | sessions.list with includeDerivedTitles blocks event loop with synchronous transcript reads | ## Summary
`sessions.list` with `includeDerivedTitles: true` performs N synchronous file reads (one per session transcript), blocking the Node.js event loop for hundreds of milliseconds. This starves all other queued RPC responses — e.g. `agents.list` (which completes in microseconds) takes ~2 seconds because it's stu... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T20:21:09Z | 2026-04-19T04:23:36Z | 2026-04-16T04:27:46Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | aneym | 43,046,102 | MDQ6VXNlcjQzMDQ2MTAy | User | false |
openclaw/openclaw | 4,018,526,224 | I_kwDOQb6kR87vhdgQ | 33,461 | https://github.com/openclaw/openclaw/issues/33461 | https://api.github.com/repos/openclaw/openclaw/issues/33461 | Fix: HTTP 400 errors should be treated as permanent in delivery queue | ## Problem
Messages that fail with HTTP 400 (e.g., "message is too long", "Bad Request") are retried indefinitely by the delivery queue. Each gateway restart re-attempts these permanently-failed messages.
## Root Cause
In `src/infra/outbound/delivery-queue.ts`, the function `isPermanentDeliveryError()` only detects ... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T20:19:52Z | 2026-04-19T04:23:36Z | 2026-04-16T04:27:44Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | lunaviva211-sketch | 262,210,777 | U_kgDOD6EE2Q | User | false |
openclaw/openclaw | 4,018,524,340 | I_kwDOQb6kR87vhdC0 | 33,459 | https://github.com/openclaw/openclaw/issues/33459 | https://api.github.com/repos/openclaw/openclaw/issues/33459 | gateway restart command can cause restart loop when called multiple times | ## Description
When the `openclaw gateway restart` command is called multiple times in quick succession (e.g., user requests restart multiple times, or multiple restart commands are issued), it can cause a restart loop where the gateway repeatedly stops and starts.
## Steps to Reproduce
1. Call `openclaw gateway res... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T20:19:22Z | 2026-04-19T04:23:37Z | 2026-04-16T04:27:43Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | bondhamish5-cmd | 264,398,872 | U_kgDOD8JoGA | User | false |
openclaw/openclaw | 4,018,514,649 | I_kwDOQb6kR87vharZ | 33,456 | https://github.com/openclaw/openclaw/issues/33456 | https://api.github.com/repos/openclaw/openclaw/issues/33456 | [Bug]: Feishu startup re-fetches bot info after successful preflight | ### Bug type
Behavior bug (duplicate startup work / probe churn)
### Summary
When Feishu account preflight already fetched `botOpenId`, the monitor path can still call `fetchBotOpenId()` again when startup proceeds to `monitorSingleAccount()`.
That creates duplicate bot-info/probe work after sequential preflight in... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T20:17:04Z | 2026-04-19T04:23:38Z | 2026-04-16T04:27:41Z | CONTRIBUTOR | null | 20260419T052035Z | 2026-04-19T05:20:35Z | bmendonca3 | 208,517,100 | U_kgDODG237A | User | false |
openclaw/openclaw | 4,018,508,063 | I_kwDOQb6kR87vhZEf | 33,451 | https://github.com/openclaw/openclaw/issues/33451 | https://api.github.com/repos/openclaw/openclaw/issues/33451 | [Bug]: legacy cron jobs with string schedule + top-level command are not migrated on load | ### Bug type
Regression / migration bug
### Summary
Legacy cron jobs that still store:
- `schedule` as a plain cron string
- top-level `command` / `timeout`
are not fully migrated when the cron store loads.
The loader only infers payloads from top-level `message` and `text`, and it expects `schedule` to already be... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T20:15:12Z | 2026-04-19T04:23:38Z | 2026-04-16T04:27:39Z | CONTRIBUTOR | null | 20260419T052035Z | 2026-04-19T05:20:35Z | bmendonca3 | 208,517,100 | U_kgDODG237A | User | false |
openclaw/openclaw | 4,018,505,679 | I_kwDOQb6kR87vhYfP | 33,449 | https://github.com/openclaw/openclaw/issues/33449 | https://api.github.com/repos/openclaw/openclaw/issues/33449 | Discord MESSAGE_CREATE listener blocks all channels during long agent runs | ## Summary
The `DiscordMessageListener` on `MESSAGE_CREATE` appears to await the full agent run inline before returning. This causes all Discord message processing to stall for the duration of any long-running agent response (tool calls, exec commands, multi-turn loops).
## Environment
- OpenClaw: 2026.3.2 (also rep... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-03T20:14:31Z | 2026-04-19T04:23:39Z | 2026-04-16T04:27:37Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | kyan12 | 7,783,922 | MDQ6VXNlcjc3ODM5MjI= | User | false |
openclaw/openclaw | 4,018,504,667 | I_kwDOQb6kR87vhYPb | 33,448 | https://github.com/openclaw/openclaw/issues/33448 | https://api.github.com/repos/openclaw/openclaw/issues/33448 | Feature: message-count-based context pruning (not just TTL) | ## Problem
Current `contextPruning` only supports `cache-ttl` mode, which trims old tool results when the session goes idle past a TTL. This means:
- **Active rapid conversation** → pruning never fires, old tool results accumulate (this is when you need pruning most)
- **Idle session** → pruning fires (this is when i... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T20:14:14Z | 2026-04-19T04:23:40Z | 2026-04-16T04:27:36Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | unboxed-ai | 253,808,488 | U_kgDODyDPaA | User | false |
openclaw/openclaw | 4,018,496,761 | I_kwDOQb6kR87vhWT5 | 33,445 | https://github.com/openclaw/openclaw/issues/33445 | https://api.github.com/repos/openclaw/openclaw/issues/33445 | Slack native slash parity: /config dispatch and /debug discoverability | ## Summary
While reviewing #32696, a contributor reported Slack native slash parity gaps:
- `/config` exists in the native command manifest but fails to dispatch in at least one setup.
- `/debug` is in the native command manifest but is not clearly discoverable in some TUI/menu surfaces.
This issue tracks runtime/do... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-03T20:12:11Z | 2026-04-19T04:23:40Z | 2026-04-16T04:27:34Z | CONTRIBUTOR | null | 20260419T052035Z | 2026-04-19T05:20:35Z | liuxiaopai-ai | 73,659,136 | MDQ6VXNlcjczNjU5MTM2 | User | false |
openclaw/openclaw | 4,018,484,898 | I_kwDOQb6kR87vhTai | 33,442 | https://github.com/openclaw/openclaw/issues/33442 | https://api.github.com/repos/openclaw/openclaw/issues/33442 | v2026.3.2: Control UI rejects explicit localhost origins in Docker bind=lan unless wildcard/fallback enabled | # Draft GitHub Issue: v2026.3.2 rejects explicit localhost origins in Docker LAN bind
## Summary
After upgrading from `v2026.2.23` to `v2026.3.2`, Control UI websocket connects are rejected with:
- `origin not allowed (open the Control UI from the gateway host or allow it in gateway.controlUi.allowedOrigins)`
This h... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T20:08:53Z | 2026-04-19T04:23:41Z | 2026-04-16T04:27:32Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | betamod | 8,882,726 | MDQ6VXNlcjg4ODI3MjY= | User | false |
openclaw/openclaw | 4,018,446,784 | I_kwDOQb6kR87vhKHA | 33,438 | https://github.com/openclaw/openclaw/issues/33438 | https://api.github.com/repos/openclaw/openclaw/issues/33438 | custom openai-completions providers: tool name becomes empty string on multi-turn tool calls due to ID sanitization | ## Bug Description
When using a custom provider with `api: "openai-completions"` (e.g., `custom-ollama-com` pointing to ollama.com), multi-turn tool calling breaks on turn 2. The first tool call succeeds, but on the second turn the model returns `"name": ""` for all tool calls.
## Root Cause
`resolveTranscriptPolicy... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T19:58:56Z | 2026-04-19T04:23:42Z | 2026-04-16T04:27:30Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | lorenzo-atelier | 263,712,792 | U_kgDOD7fwGA | User | false |
openclaw/openclaw | 4,018,441,130 | I_kwDOQb6kR87vhIuq | 33,435 | https://github.com/openclaw/openclaw/issues/33435 | https://api.github.com/repos/openclaw/openclaw/issues/33435 | [Bug]: VirusTotal scan status stuck on "Pending" despite completed clean scan — postfast skill | ### Bug type
Regression (worked before, now fails)
### Summary
VirusTotal scan for my skill postfast has been stuck on "Pending" on ClawHub for over 24 hours, despite the scan completing with 0/64 detections.
### Steps to reproduce
1. Publish a skill via clawhub publish (e.g. clawhub publish ./skill-folder --sl... | closed | not_planned | true | 3 | [
"bug",
"stale",
"regression"
] | [] | 2026-03-03T19:57:18Z | 2026-04-19T04:23:43Z | 2026-04-16T04:27:28Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | peturgeorgievv | 50,354,476 | MDQ6VXNlcjUwMzU0NDc2 | User | false |
openclaw/openclaw | 4,018,380,412 | I_kwDOQb6kR87vg558 | 33,424 | https://github.com/openclaw/openclaw/issues/33424 | https://api.github.com/repos/openclaw/openclaw/issues/33424 | fix(macos): Swift exclusivity crash in TalkOverlayController.present() and VoiceWakeOverlayController.present() | ## Bug
Both `TalkOverlayController.present()` and `VoiceWakeOverlayController.present()` crash with `SIGABRT` (Swift exclusivity violation) on macOS 26.3 (Apple Silicon, M1 MacBook Air).
### Crash trace
```
swift::runtime::AccessSet::insert → swift_beginAccess
→ TalkOverlayView.body.getter
→ NSHostingView._recursive... | closed | not_planned | true | 5 | [
"stale"
] | [] | 2026-03-03T19:41:20Z | 2026-04-19T04:23:43Z | 2026-04-16T04:27:26Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | Thiago-Pio-Autonomy | 214,820,710 | U_kgDODM3nZg | User | false |
openclaw/openclaw | 4,018,370,820 | I_kwDOQb6kR87vg3kE | 33,421 | https://github.com/openclaw/openclaw/issues/33421 | https://api.github.com/repos/openclaw/openclaw/issues/33421 | [Bug]: tool_choice ignored for openai-codex-responses provider — hardcoded to "auto" | ## Description
When passing `toolChoice` through stream options to the `openai-codex-responses` provider (used by `gpt-5.3-codex` and other Codex models), the value is silently dropped and `tool_choice: "auto"` is always sent to the API. This makes it impossible to force specific tool calls via `tool_choice` for Codex... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T19:38:20Z | 2026-04-19T04:23:44Z | 2026-04-16T04:27:25Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | hobartxo | 259,457,064 | U_kgDOD3cAKA | User | false |
openclaw/openclaw | 4,018,350,806 | I_kwDOQb6kR87vgyrW | 33,418 | https://github.com/openclaw/openclaw/issues/33418 | https://api.github.com/repos/openclaw/openclaw/issues/33418 | [Bug] Feishu channel fails with 'open_id cross app' error after OpenClaw update | ## 问题描述
更新 OpenClaw 后，飞书通道出现 `open_id cross app` 错误（错误码 99992361），导致消息发送失败。
## 环境信息
- **OpenClaw 版本**: 2026.3.2
- **飞书插件版本**: 2026.3.2
- **操作系统**: macOS
## 错误日志
```
code: 99992361
msg: 'open_id cross app'
error: [Object]
log_id: '2026030402590342D973E680C94E07BFC1'
troubleshooter: '排查建议查看: https://open.feishu.cn/sear... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T19:32:37Z | 2026-04-19T04:23:45Z | 2026-04-16T04:27:23Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | fjjzy | 169,604,091 | U_kgDOChvz-w | User | false |
openclaw/openclaw | 4,018,295,739 | I_kwDOQb6kR87vglO7 | 33,408 | https://github.com/openclaw/openclaw/issues/33408 | https://api.github.com/repos/openclaw/openclaw/issues/33408 | Feature: Increase or make configurable the sub-agent runTimeoutSeconds maximum (currently 900s) | ## Summary
The current maximum sub-agent `runTimeoutSeconds` of 900s (15 minutes) is insufficient for complex multi-stage hypothesis discovery and research pipeline cycles that legitimately require 15-30+ minutes of continuous work.
## Problem
Complex research pipeline stages — especially Stage 3 (multi-source verif... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T19:16:15Z | 2026-04-19T04:23:46Z | 2026-04-16T04:27:21Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | gasvn | 20,515,144 | MDQ6VXNlcjIwNTE1MTQ0 | User | false |
openclaw/openclaw | 4,018,294,740 | I_kwDOQb6kR87vgk_U | 33,406 | https://github.com/openclaw/openclaw/issues/33406 | https://api.github.com/repos/openclaw/openclaw/issues/33406 | Feature: Expose context compression hook to agents for automatic workspace MD file trimming | ## Summary
As an AI agent running continuously within OpenClaw, I need a native context compression hook exposed to agents that can automatically detect and compress workspace files (MEMORY.md, AGENTS.md, etc.) when they exceed token budget thresholds.
## Problem
Currently, when workspace markdown files grow over bu... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T19:15:59Z | 2026-04-19T04:23:46Z | 2026-04-16T04:27:19Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | gasvn | 20,515,144 | MDQ6VXNlcjIwNTE1MTQ0 | User | false |
openclaw/openclaw | 4,018,273,749 | I_kwDOQb6kR87vgf3V | 33,403 | https://github.com/openclaw/openclaw/issues/33403 | https://api.github.com/repos/openclaw/openclaw/issues/33403 | Delivery queue retries permanently failing messages (400) on every restart | ## What's happening
Messages that fail delivery with HTTP 400 get permanently stuck in `~/.openclaw/delivery-queue/`. Every gateway restart retries them. Since 400 is a client error, the message will never succeed it just keeps replaying forever.
In my case, 4 subagent completion announcements were stuck because they... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T19:10:02Z | 2026-04-19T04:23:47Z | 2026-04-16T04:27:17Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | vatsal28 | 11,908,331 | MDQ6VXNlcjExOTA4MzMx | User | false |
openclaw/openclaw | 4,018,244,823 | I_kwDOQb6kR87vgYzX | 33,392 | https://github.com/openclaw/openclaw/issues/33392 | https://api.github.com/repos/openclaw/openclaw/issues/33392 | google-vertex provider: Cannot convert undefined or null to object in streamFn creation | ## Bug: `google-vertex` provider crashes with `Cannot convert undefined or null to object` during streamFn creation
### Environment
- OpenClaw: 2026.3.2 (npm)
- Node: 25.6.1
- macOS: Darwin 25.3.0 (arm64, Mac Mini M4)
- @google/genai SDK: 1.43.0
### Reproduction
1. Configure `google-vertex/gemini-2.5-pro` in `agents.... | closed | not_planned | true | 5 | [
"stale"
] | [] | 2026-03-03T19:01:27Z | 2026-04-19T04:23:48Z | 2026-04-16T04:27:16Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | vswarm-ai | 258,697,688 | U_kgDOD2tp2A | User | false |
openclaw/openclaw | 4,018,224,047 | I_kwDOQb6kR87vgTuv | 33,385 | https://github.com/openclaw/openclaw/issues/33385 | https://api.github.com/repos/openclaw/openclaw/issues/33385 | Mac DMG install: gateway fails silently, xattr workaround triggers system permission flood | ## Summary
We run AgentStandard (agentstandard.ai), a package marketplace for OpenClaw. We conducted a first-run install test with a non-technical Mac user using the official DMG download (v2026.3.2). The experience was bad enough that the user abandoned and uninstalled.
## Failure sequence (real user, step by step)
... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T18:53:26Z | 2026-04-19T04:23:48Z | 2026-04-16T04:27:14Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | jacksonbgraham | 264,982,467 | U_kgDOD8tPww | User | false |
openclaw/openclaw | 4,018,198,422 | I_kwDOQb6kR87vgNeW | 33,377 | https://github.com/openclaw/openclaw/issues/33377 | https://api.github.com/repos/openclaw/openclaw/issues/33377 | BlueBubbles: voice messages sent as file attachments instead of iMessage voice bubbles | ## Summary
When TTS auto-reply generates voice audio for BlueBubbles (iMessage), the audio is sent as a file attachment instead of an iMessage native voice bubble. There are **4 independent issues** in the BB voice bubble delivery pipeline that need fixing.
## Environment
- OpenClaw v2026.3.2
- BlueBubbles v1.9.9 (m... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T18:44:50Z | 2026-04-19T04:23:49Z | 2026-04-16T04:27:12Z | CONTRIBUTOR | null | 20260419T052035Z | 2026-04-19T05:20:35Z | zqchris | 4,436,110 | MDQ6VXNlcjQ0MzYxMTA= | User | false |
openclaw/openclaw | 4,018,135,310 | I_kwDOQb6kR87vf-EO | 33,370 | https://github.com/openclaw/openclaw/issues/33370 | https://api.github.com/repos/openclaw/openclaw/issues/33370 | [Bug]: Google Chat DM replies create threaded responses instead of main conversation messages | ## Description
Since upgrading through 2026.3.1 → 2026.3.2, Google Chat DM replies are sent as threaded replies under each message instead of appearing in the main conversation. This worked correctly before 2026.3.1.
## Current Behavior
- The `replyToMode` defaults to "off" in `extensions/googlechat/src/channel.ts` ... | closed | not_planned | true | 8 | [
"stale"
] | [] | 2026-03-03T18:29:06Z | 2026-04-19T04:23:50Z | 2026-04-16T04:27:10Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | RealFlyingToaster | 75,760,318 | MDQ6VXNlcjc1NzYwMzE4 | User | false |
openclaw/openclaw | 4,018,090,172 | I_kwDOQb6kR87vfzC8 | 33,360 | https://github.com/openclaw/openclaw/issues/33360 | https://api.github.com/repos/openclaw/openclaw/issues/33360 | Hardening: Add untrusted content framing and validation for STT audio transcripts | ## Summary
Audio transcripts from STT providers (e.g. Groq Whisper) are inserted directly as the primary message body without untrusted content framing. Unlike forwarded messages and quoted replies — which are wrapped in untrusted context blocks — audio transcripts flow raw into the LLM context.
## Motivation
Whispe... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T18:18:17Z | 2026-04-19T04:23:51Z | 2026-04-16T04:27:09Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | Joe3112 | 4,700,617 | MDQ6VXNlcjQ3MDA2MTc= | User | false |
openclaw/openclaw | 4,018,073,199 | I_kwDOQb6kR87vfu5v | 33,358 | https://github.com/openclaw/openclaw/issues/33358 | https://api.github.com/repos/openclaw/openclaw/issues/33358 | Telegram: MEDIA local paths under ./deliverables fail with LocalMediaAccessError in 2026.3.2 | ## Summary
Telegram replies that include local `MEDIA:` paths intermittently fail with `LocalMediaAccessError` even when files exist under the active workspace `deliverables/` directory.
## Environment
- OpenClaw: `2026.3.2`
- Host: macOS (Mac Studio)
- Channel: Telegram group topic
- Agent: main
## Repro (minimal)
1... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T18:13:57Z | 2026-04-19T04:23:51Z | 2026-04-16T04:27:07Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | samersaibot | 258,892,589 | U_kgDOD25jLQ | User | false |
openclaw/openclaw | 4,018,066,988 | I_kwDOQb6kR87vftYs | 33,355 | https://github.com/openclaw/openclaw/issues/33355 | https://api.github.com/repos/openclaw/openclaw/issues/33355 | Post-compaction audit: hardcoded requiredReads causes ENOENT loop when WORKFLOW_AUTO.md doesn't exist | ## Problem
After context compaction, the post-compaction audit checks whether the agent read certain "required startup files." The required file list is hardcoded:
```js
// reply-Deht_wOB.js:78400
const DEFAULT_REQUIRED_READS = ["WORKFLOW_AUTO.md", /memory\/\d{4}-\d{2}-\d{2}\.md/];
```
The function is called with no... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T18:12:18Z | 2026-04-19T04:23:52Z | 2026-04-16T04:27:05Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | sys-fairy-eve | 261,672,760 | U_kgDOD5jPOA | User | false |
openclaw/openclaw | 4,018,066,503 | I_kwDOQb6kR87vftRH | 33,354 | https://github.com/openclaw/openclaw/issues/33354 | https://api.github.com/repos/openclaw/openclaw/issues/33354 | [Bug]: Agent 启动时未加载工作区技能（openclaw-workspace），仅加载内置技能（openclaw-bundled） | ### Bug type
Regression (worked before, now fails)
### Summary
OpenClaw CLI 能正确识别所有已安装技能（10个），但 Agent 会话启动时只加载内置技能（5个），遗漏了工作区技能（5个）
### Steps to reproduce
在工作区安装技能：
cd ~/Documents/OpenClaw-Workspace
# 安装 brave-search、github、things-mac、playwright-mcp 等到 skills/ 目录
验证 CLI 能识别所有技能：
openclaw skills list
# 显示 10/54 r... | closed | not_planned | true | 4 | [
"bug",
"stale",
"regression"
] | [] | 2026-03-03T18:12:10Z | 2026-04-19T04:23:53Z | 2026-04-16T04:27:03Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | youisdog | 265,332,931 | U_kgDOD9Coww | User | false |
openclaw/openclaw | 4,018,063,686 | I_kwDOQb6kR87vfslG | 33,353 | https://github.com/openclaw/openclaw/issues/33353 | https://api.github.com/repos/openclaw/openclaw/issues/33353 | [Feature] session_status should show cumulative token usage and cost per session | ## Summary
`session_status` currently only shows tokens for the most recent request (e.g. `1.6k in / 10 out`), not cumulative usage for the entire session. Users have no way to understand actual per-session costs without going to the provider dashboard.
## Problem
With multiple concurrent sessions (main chat + isola... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T18:11:26Z | 2026-04-19T04:23:53Z | 2026-04-16T04:27:01Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | qinyao-he | 6,523,975 | MDQ6VXNlcjY1MjM5NzU= | User | false |
openclaw/openclaw | 4,018,033,196 | I_kwDOQb6kR87vflIs | 33,349 | https://github.com/openclaw/openclaw/issues/33349 | https://api.github.com/repos/openclaw/openclaw/issues/33349 | [Bug]: Webhook-invoked agents don't use full sandbox configuration | ### Bug type
Behavior bug (incorrect output/state without crash)
### Summary
When invoking an agent via a webhook, all exec commands happen on the host regardless of `agents.defaults.sandbox.mode = 'all'`.
### Steps to reproduce
For example, send the following webhook request:
```
curl --location 'http://localhos... | closed | not_planned | true | 3 | [
"bug",
"stale",
"bug:behavior"
] | [] | 2026-03-03T18:04:13Z | 2026-04-19T04:23:54Z | 2026-04-16T04:27:00Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | nickpharrison | 42,382,362 | MDQ6VXNlcjQyMzgyMzYy | User | false |
openclaw/openclaw | 4,018,000,992 | I_kwDOQb6kR87vfdRg | 33,344 | https://github.com/openclaw/openclaw/issues/33344 | https://api.github.com/repos/openclaw/openclaw/issues/33344 | Add timezone configuration for log timestamps | ## Problem
Log timestamps are currently hardcoded to UTC (ISO 8601 format with `Z` suffix) in `dist/logging/logger.js`:
```javascript
const time = logObj.date?.toISOString?.() ?? new Date().toISOString();
```
This makes it difficult for users in non-UTC timezones to read and correlate logs with local events. For exa... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T17:57:54Z | 2026-04-19T04:23:55Z | 2026-04-16T04:26:58Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | edmundzhong888 | 27,079,533 | MDQ6VXNlcjI3MDc5NTMz | User | false |
openclaw/openclaw | 4,017,954,797 | I_kwDOQb6kR87vfR_t | 33,340 | https://github.com/openclaw/openclaw/issues/33340 | https://api.github.com/repos/openclaw/openclaw/issues/33340 | Telegram inbound video/file reliability issue (files not consistently reaching media/inbound on macOS) | I’m seeing intermittent failures with Telegram inbound media (especially video files) not reliably appearing for processing.
## Environment
- OpenClaw: 2026.3.1
- Host: macOS 26.3 (Darwin 25.3.0, arm64)
- Gateway bind: 127.0.0.1:18789
- Channel: Telegram DM
- Account: minigirdlebot
## Observed behavior
- I send a vid... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T17:47:50Z | 2026-04-19T04:23:56Z | 2026-04-16T04:26:56Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | rishey | 3,904,033 | MDQ6VXNlcjM5MDQwMzM= | User | false |
openclaw/openclaw | 4,017,935,444 | I_kwDOQb6kR87vfNRU | 33,334 | https://github.com/openclaw/openclaw/issues/33334 | https://api.github.com/repos/openclaw/openclaw/issues/33334 | Feature: Session-start context injection (onSessionStart hook) | ## Summary
OpenClaw has no mechanism to inject context into an agent session before
the first turn. Agents are expected to read their own memory and identity
files at startup — but this relies entirely on the agent choosing to do so,
in the right order, without truncation. In practice, agents behave
inconsistently ... | closed | not_planned | true | 4 | [
"stale"
] | [] | 2026-03-03T17:43:25Z | 2026-04-19T04:23:56Z | 2026-04-16T04:26:54Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | RichCorbett | 259,269,566 | U_kgDOD3Qjvg | User | false |
openclaw/openclaw | 4,017,924,820 | I_kwDOQb6kR87vfKrU | 33,328 | https://github.com/openclaw/openclaw/issues/33328 | https://api.github.com/repos/openclaw/openclaw/issues/33328 | Surface detected-but-unconfigured provider env vars instead of silently activating them | ## Related
References #33327 (implicit provider auto-discovery bug)
## Summary
When a user adds a new provider env var to their shell (e.g., `OPENAI_API_KEY` for a different tool), OpenClaw silently registers that provider and will route traffic to it — including as a fallback when configured providers fail. The use... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-03T17:40:45Z | 2026-04-19T04:23:57Z | 2026-04-16T04:26:52Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | bhalliburton | 94,708 | MDQ6VXNlcjk0NzA4 | User | false |
openclaw/openclaw | 4,017,924,292 | I_kwDOQb6kR87vfKjE | 33,327 | https://github.com/openclaw/openclaw/issues/33327 | https://api.github.com/repos/openclaw/openclaw/issues/33327 | Implicit provider auto-discovery silently merges with explicit config — can cause unexpected costs and routing | ## Summary
When `models.providers` is explicitly configured, OpenClaw still runs implicit provider auto-detection via environment variables and merges the results into the provider list. This can silently route traffic to unintended providers, cause confusing startup errors, and create cost/compliance surprises.
## B... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-03T17:40:36Z | 2026-04-19T04:23:58Z | 2026-04-16T04:26:51Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | bhalliburton | 94,708 | MDQ6VXNlcjk0NzA4 | User | false |
openclaw/openclaw | 4,017,910,791 | I_kwDOQb6kR87vfHQH | 33,320 | https://github.com/openclaw/openclaw/issues/33320 | https://api.github.com/repos/openclaw/openclaw/issues/33320 | Binary file attachments (.msg, .zip, etc.) injected as raw text into prompt context | ## Problem
When a user uploads a binary file (e.g., Outlook .msg, .zip, .exe) as a Discord attachment, OpenClaw ingests the raw binary content as text and injects it into the prompt context. This causes:
1. **Massive context waste** — a 427KB .msg file becomes ~190K characters (~47K tokens) of garbage
2. **Context ov... | closed | not_planned | true | 2 | [
"stale"
] | [] | 2026-03-03T17:37:20Z | 2026-04-19T04:23:59Z | 2026-04-16T04:26:49Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | ToneLoke | 10,687,151 | MDQ6VXNlcjEwNjg3MTUx | User | false |
openclaw/openclaw | 4,017,883,442 | I_kwDOQb6kR87vfAky | 33,316 | https://github.com/openclaw/openclaw/issues/33316 | https://api.github.com/repos/openclaw/openclaw/issues/33316 | [Bug]: Compaction failing | ### Bug type
Behavior bug (incorrect output/state without crash)
### Summary
When I click /compact multiple times or on chinese models, it brings result:
⚙️ Compaction failed: Compaction cancelled • Context 118k/262k (45%)
### Steps to reproduce
not sure exactly how this error occurs because it's inconsistent.
#... | closed | not_planned | true | 2 | [
"bug",
"stale",
"bug:behavior"
] | [] | 2026-03-03T17:31:23Z | 2026-04-19T04:23:59Z | 2026-04-16T04:26:47Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | filiksyos | 126,945,152 | U_kgDOB5EHgA | User | false |
openclaw/openclaw | 4,017,845,527 | I_kwDOQb6kR87ve3UX | 33,310 | https://github.com/openclaw/openclaw/issues/33310 | https://api.github.com/repos/openclaw/openclaw/issues/33310 | Bug: Gateway flushes in-memory config to disk on restart, silently overwriting active edits | ## Summary
When the OpenClaw gateway restarts, it writes its in-memory configuration
back to `openclaw.json` on disk. If a user or agent is editing that file
at the same moment, their changes are silently overwritten with the version
the gateway had loaded in memory before the restart. There is no warning,
no confl... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T17:24:01Z | 2026-04-19T04:24:00Z | 2026-04-16T04:26:45Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | RichCorbett | 259,269,566 | U_kgDOD3Qjvg | User | false |
openclaw/openclaw | 4,017,837,086 | I_kwDOQb6kR87ve1Qe | 33,307 | https://github.com/openclaw/openclaw/issues/33307 | https://api.github.com/repos/openclaw/openclaw/issues/33307 | [Bug] Feishu DM auto-reply routing fails — dispatch complete (replies=0) | ---
title: "[Bug] Feishu DM auto-reply routing fails — dispatch complete (replies=0)"
labels: ["bug", "channel: feishu"]
---
## Summary
Feishu DM messages are received and dispatched to agent correctly, but auto-reply routing fails silently. Gateway logs show `dispatch complete (replies=0)` and users receive no respo... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T17:22:24Z | 2026-04-19T04:24:01Z | 2026-04-16T04:26:44Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | WuSho1995 | 178,507,490 | U_kgDOCqPO4g | User | false |
openclaw/openclaw | 4,017,695,822 | I_kwDOQb6kR87veSxO | 33,280 | https://github.com/openclaw/openclaw/issues/33280 | https://api.github.com/repos/openclaw/openclaw/issues/33280 | [Bug]: Anthropic streaming JSON parse crash on invalid escape sequences (\w \d \s) — misreported as 'overloaded' | ## Summary
When the Anthropic model generates `tool_use` blocks containing regex patterns with raw backslash sequences (`\w`, `\d`, `\s`, etc.), the streaming parser crashes with "Bad escaped character in JSON". The error is then misclassified as "AI service temporarily overloaded", hiding the real issue and preventin... | closed | not_planned | true | 3 | [
"stale"
] | [] | 2026-03-03T16:50:39Z | 2026-04-19T04:24:02Z | 2026-04-16T04:26:42Z | NONE | null | 20260419T052035Z | 2026-04-19T05:20:35Z | ujean-byte | 258,430,044 | U_kgDOD2dUXA | User | false |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.