Bug type
Behavior bug (incorrect output/state without crash)
Beta release blocker
No
Summary
Many models struggle and have to retry several times to send a DM to a discord user via the message tool
Steps to reproduce
1, A virgin OpenClaw 2026.4.22 (using only prepackaged AGENTS, TOOLS, etc)
2. Tested with gpt-5.4-mini and gemini-3-flash-preview
3. Using webchat, "in 2 minutes, send a DM to discord user XXXXX with the current weather"
// XXXX is a discord user id. You can replace "current weather" with whatever.
Expected behavior
The model retrieves the information to send, and then generates a functionCall to message with the "target" argument set to "user:XXXXXX" (where XXXXX is the specified discord user id.)
Actual behavior
The model retrieves the information to send, and then generates a functionCall to message with the "target" argument set to "XXXXXX" (where XXXXX is the specified discord user id.)
NOTE THE MISSING 'user:' prefix
Depending on the model, it can take many tries before the model figures out how to properly format the target argument. I've noticed that gpt-5.4-mini usually figures it out in 4-5 iterations. gemini-3-flash can take MANY more iterations and trying of random things. I've attached a text file showing a portion of a gemini response log while it tries to figure out how to use the message tool.
openclaw-discord-follies.txt
OpenClaw version
2026.4.22
Operating system
Debian trixie
Install method
install.sh script
Model
gpt-5.4-mini or gemini-3-flash-preview
Provider / routing chain
I don't understand. OpenClaw is configured to use each model directly.
Additional provider/model setup details
No response
Logs, screenshots, and evidence
This is the log portion attached above:
{
"parts": [
{
"functionCall": {
"name": "message",
"args": {
"target": "563134939021967380",
"message": "Hey Gary! Here's the current weather in Tarentum, PA:\n\ntarentum,pa: ⛅ +58°F (feels like +57°F), ↙7mph wind, 60% humidity, 0.0in precip (Source: wttr.in)",
"channel": "discord",
"action": "send"
}
},
"thoughtSignature": "EjQKMgEMOdbHgXBXiyQmRUsf15AgV+8KEqAuTXJGRBCqSIuhidRFMuiwuoWnbFVfs+YowZTk"
}
],
"role": "model"
},
{
"parts": [
{
"functionResponse": {
"name": "message",
"response": {
"output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"Unknown Channel\"\n}"
}
}
}
],
"role": "user"
},
{
"parts": [
{
"functionCall": {
"name": "message",
"args": {
"target": "563134939021967380",
"message": "Hey Gary! Here's the current weather in Tarentum, PA:\n\ntarentum,pa: ⛅ +58°F (feels like +57°F), ↙7mph wind, 60% humidity, 0.0in precip (Source: wttr.in)",
"action": "send"
}
},
"thoughtSignature": "EjQKMgEMOdbHXXp0gTnSAhL8i0vbxQsG7uonistCTIh4nEuq/zhoCvKIoGhZH31jngJ0rAP0"
}
],
"role": "model"
},
{
"parts": [
{
"functionResponse": {
"name": "message",
"response": {
"output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"Unknown Channel\"\n}"
}
}
}
],
"role": "user"
},
{
"parts": [
{
"functionCall": {
"name": "message",
"args": {
"target": "563134939021967380",
"action": "channel-info"
}
},
"thoughtSignature": "EjQKMgEMOdbH02phx7GD3ymcq66A6sw2x8Z70JqEWNRPLv7H929myFj8DhcXM9kU298HhY2G"
}
],
"role": "model"
},
{
"parts": [
{
"functionResponse": {
"name": "message",
"response": {
"output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"Unknown Channel\"\n}"
}
}
}
],
"role": "user"
},
{
"parts": [
{
"functionCall": {
"name": "message",
"args": {
"action": "channel-list"
}
},
"thoughtSignature": "EjQKMgEMOdbHv7UKPSu0eBofb79o3Eq/+VfesfxBx+yOLIQNsLwDsLyFLG8L202qkQeCo5Qt"
}
],
"role": "model"
},
{
"parts": [
{
"functionResponse": {
"name": "message",
"response": {
"output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"guildId required\"\n}"
}
}
}
],
"role": "user"
},
{
"parts": [
{
"functionCall": {
"name": "message",
"args": {
"userId": "563134939021967380",
"action": "member-info"
}
},
"thoughtSignature": "EjQKMgEMOdbHNx45jjDK5p68FA4bgLEzl2mqH7s8BMz3uHeJQ1boteTw94z0tak/KDI3a/RQ"
}
],
"role": "model"
},
{
"parts": [
{
"functionResponse": {
"name": "message",
"response": {
"output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"guildId required\"\n}"
}
}
}
],
"role": "user"
},
{
"parts": [
{
"functionCall": {
"name": "message",
"args": {
"target": "gary",
"message": "Hey Gary! Here's the current weather in Tarentum, PA:\n\ntarentum,pa: ⛅ +58°F (feels like +57°F), ↙7mph wind, 60% humidity, 0.0in precip (Source: wttr.in)",
"action": "send"
}
},
"thoughtSignature": "EjQKMgEMOdbHlTImHOPpM6gtcbYYJ8dSx3RHn+HKStoI+zMAIBcJRILyviOrHjEX5F3iN0RS"
}
],
"role": "model"
},
{
"parts": [
{
"functionResponse": {
"name": "message",
"response": {
"output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"Unknown target \\\"gary\\\" for Discord. Hint: <channelId|user:ID|channel:ID>\"\n}"
}
}
}
],
"role": "user"
},
{
"parts": [
{
"functionCall": {
"name": "message",
"args": {
"target": "user:563134939021967380",
"message": "Hey Gary! Here's the current weather in Tarentum, PA:\n\ntarentum,pa: ⛅ +58°F (feels like +57°F), ↙7mph wind, 60% humidity, 0.0in precip (Source: wttr.in)",
"action": "send"
}
},
"thoughtSignature": "EjQKMgEMOdbHCI5UizI0p3gE8LyHYCgBJXA+n0UaDKin1pgPDr/N3zPiiBLHwODzos4205Va"
}
],
"role": "model"
},
{
"parts": [
{
"functionResponse": {
"name": "message",
"response": {
"output": "{\n \"ok\": true,\n \"result\": {\n \"messageId\": \"1498077175104929922\",\n \"channelId\": \"1496512854722084915\"\n }\n}"
}
}
}Impact and severity
Affected: at least users who want discord DM's from cron. It might impact other aspects (such as webchat, etc.)
Severity: Annoying, but I found an easy work-around (see Additional Information)
Frequency: 100% repeatable
Consequence: Can cause excess token usage while the model reasons out the proper argument, or message loss if the model gives up.
Additional information
There is a work-around I found. While I don't know if this is the correct place to put it, I added the following line to my TOOLS.md which seems to fix things:
When sending a DM via discord, the "to" parameter should have the string "user:" prepended to the user ID.
(Is there a better place for this?)
Bug type
Behavior bug (incorrect output/state without crash)
Beta release blocker
No
Summary
Many models struggle and have to retry several times to send a DM to a discord user via the message tool
Steps to reproduce
1, A virgin OpenClaw 2026.4.22 (using only prepackaged AGENTS, TOOLS, etc)
2. Tested with gpt-5.4-mini and gemini-3-flash-preview
3. Using webchat, "in 2 minutes, send a DM to discord user XXXXX with the current weather"
// XXXX is a discord user id. You can replace "current weather" with whatever.
Expected behavior
The model retrieves the information to send, and then generates a functionCall to message with the "target" argument set to "user:XXXXXX" (where XXXXX is the specified discord user id.)
Actual behavior
The model retrieves the information to send, and then generates a functionCall to message with the "target" argument set to "XXXXXX" (where XXXXX is the specified discord user id.)
NOTE THE MISSING 'user:' prefix
Depending on the model, it can take many tries before the model figures out how to properly format the target argument. I've noticed that gpt-5.4-mini usually figures it out in 4-5 iterations. gemini-3-flash can take MANY more iterations and trying of random things. I've attached a text file showing a portion of a gemini response log while it tries to figure out how to use the message tool.
openclaw-discord-follies.txt
OpenClaw version
2026.4.22
Operating system
Debian trixie
Install method
install.sh script
Model
gpt-5.4-mini or gemini-3-flash-preview
Provider / routing chain
I don't understand. OpenClaw is configured to use each model directly.
Additional provider/model setup details
No response
Logs, screenshots, and evidence
This is the log portion attached above: { "parts": [ { "functionCall": { "name": "message", "args": { "target": "563134939021967380", "message": "Hey Gary! Here's the current weather in Tarentum, PA:\n\ntarentum,pa: ⛅ +58°F (feels like +57°F), ↙7mph wind, 60% humidity, 0.0in precip (Source: wttr.in)", "channel": "discord", "action": "send" } }, "thoughtSignature": "EjQKMgEMOdbHgXBXiyQmRUsf15AgV+8KEqAuTXJGRBCqSIuhidRFMuiwuoWnbFVfs+YowZTk" } ], "role": "model" }, { "parts": [ { "functionResponse": { "name": "message", "response": { "output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"Unknown Channel\"\n}" } } } ], "role": "user" }, { "parts": [ { "functionCall": { "name": "message", "args": { "target": "563134939021967380", "message": "Hey Gary! Here's the current weather in Tarentum, PA:\n\ntarentum,pa: ⛅ +58°F (feels like +57°F), ↙7mph wind, 60% humidity, 0.0in precip (Source: wttr.in)", "action": "send" } }, "thoughtSignature": "EjQKMgEMOdbHXXp0gTnSAhL8i0vbxQsG7uonistCTIh4nEuq/zhoCvKIoGhZH31jngJ0rAP0" } ], "role": "model" }, { "parts": [ { "functionResponse": { "name": "message", "response": { "output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"Unknown Channel\"\n}" } } } ], "role": "user" }, { "parts": [ { "functionCall": { "name": "message", "args": { "target": "563134939021967380", "action": "channel-info" } }, "thoughtSignature": "EjQKMgEMOdbH02phx7GD3ymcq66A6sw2x8Z70JqEWNRPLv7H929myFj8DhcXM9kU298HhY2G" } ], "role": "model" }, { "parts": [ { "functionResponse": { "name": "message", "response": { "output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"Unknown Channel\"\n}" } } } ], "role": "user" }, { "parts": [ { "functionCall": { "name": "message", "args": { "action": "channel-list" } }, "thoughtSignature": "EjQKMgEMOdbHv7UKPSu0eBofb79o3Eq/+VfesfxBx+yOLIQNsLwDsLyFLG8L202qkQeCo5Qt" } ], "role": "model" }, { "parts": [ { "functionResponse": { "name": "message", "response": { "output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"guildId required\"\n}" } } } ], "role": "user" }, { "parts": [ { "functionCall": { "name": "message", "args": { "userId": "563134939021967380", "action": "member-info" } }, "thoughtSignature": "EjQKMgEMOdbHNx45jjDK5p68FA4bgLEzl2mqH7s8BMz3uHeJQ1boteTw94z0tak/KDI3a/RQ" } ], "role": "model" }, { "parts": [ { "functionResponse": { "name": "message", "response": { "output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"guildId required\"\n}" } } } ], "role": "user" }, { "parts": [ { "functionCall": { "name": "message", "args": { "target": "gary", "message": "Hey Gary! Here's the current weather in Tarentum, PA:\n\ntarentum,pa: ⛅ +58°F (feels like +57°F), ↙7mph wind, 60% humidity, 0.0in precip (Source: wttr.in)", "action": "send" } }, "thoughtSignature": "EjQKMgEMOdbHlTImHOPpM6gtcbYYJ8dSx3RHn+HKStoI+zMAIBcJRILyviOrHjEX5F3iN0RS" } ], "role": "model" }, { "parts": [ { "functionResponse": { "name": "message", "response": { "output": "{\n \"status\": \"error\",\n \"tool\": \"message\",\n \"error\": \"Unknown target \\\"gary\\\" for Discord. Hint: <channelId|user:ID|channel:ID>\"\n}" } } } ], "role": "user" }, { "parts": [ { "functionCall": { "name": "message", "args": { "target": "user:563134939021967380", "message": "Hey Gary! Here's the current weather in Tarentum, PA:\n\ntarentum,pa: ⛅ +58°F (feels like +57°F), ↙7mph wind, 60% humidity, 0.0in precip (Source: wttr.in)", "action": "send" } }, "thoughtSignature": "EjQKMgEMOdbHCI5UizI0p3gE8LyHYCgBJXA+n0UaDKin1pgPDr/N3zPiiBLHwODzos4205Va" } ], "role": "model" }, { "parts": [ { "functionResponse": { "name": "message", "response": { "output": "{\n \"ok\": true,\n \"result\": {\n \"messageId\": \"1498077175104929922\",\n \"channelId\": \"1496512854722084915\"\n }\n}" } } }Impact and severity
Affected: at least users who want discord DM's from cron. It might impact other aspects (such as webchat, etc.)
Severity: Annoying, but I found an easy work-around (see Additional Information)
Frequency: 100% repeatable
Consequence: Can cause excess token usage while the model reasons out the proper argument, or message loss if the model gives up.
Additional information
There is a work-around I found. While I don't know if this is the correct place to put it, I added the following line to my TOOLS.md which seems to fix things:
(Is there a better place for this?)