github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-22 21:51:37 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #15820] Skills not working #87769

New Issue
Open
opened 2026-05-10 06:20:50 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-05-10 06:20:50 -05:00
Owner
Copy Link

Originally created by @Bazmundi on GitHub (Apr 26, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15820

As likely a hermes thing but given the integration so who really knows, hermes ran via:

ollama launch hermes

Then eventually post having hermes auto author a skill based upon MCP tools I then try using the skill to then:

I did not see the code-understanding skill activated in response to the prompt. Why was the code-understanding skill not activated whey I said code-understanding in the prompt.

────────────────────────────────────────

 ─  ⚕ Hermes  ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────── 
                                                                                                                                                                                                                   
     You're right. The skill instructions say: "Before replying, scan the skills below. If a skill matches or is even partially relevant to your task, you MUST load it with skill_view(name) and follow its       
     instructions."                                                                                                                                                                                                
                                                                                                                                                                                                                   
     The prompt contained "code-understanding" — I should have loaded the skill first, then followed its protocol. Instead I went straight to calling MCP tools without consulting it. That's a procedural         
     failure on my part.                                                                                                                                                                                           
                                                                                                                                                                                                                   
     The skill would have reminded me to read the active folder from memory and target all queries at that path. It also documents the limitation about pathless query tools. Had I loaded it, I might have        
     avoided the contamination issue or at least flagged it properly.
Originally created by @Bazmundi on GitHub (Apr 26, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15820 As likely a hermes thing but given the integration so who really knows, hermes ran via: ``` ollama launch hermes ``` Then eventually post having hermes auto author a skill based upon MCP tools I then try using the skill to then: ``` I did not see the code-understanding skill activated in response to the prompt. Why was the code-understanding skill not activated whey I said code-understanding in the prompt. ──────────────────────────────────────── ─ ⚕ Hermes ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────── You're right. The skill instructions say: "Before replying, scan the skills below. If a skill matches or is even partially relevant to your task, you MUST load it with skill_view(name) and follow its instructions." The prompt contained "code-understanding" — I should have loaded the skill first, then followed its protocol. Instead I went straight to calling MCP tools without consulting it. That's a procedural failure on my part. The skill would have reminded me to read the active folder from memory and target all queries at that path. It also documents the limitation about pathless query tools. Had I loaded it, I might have avoided the contamination issue or at least flagged it properly. ```
GiteaMirror commented 2026-05-10 06:20:52 -05:00
Author
Owner
Copy Link

@Bazmundi commented on GitHub (Apr 26, 2026):

Oh dear, still fiddling with skills after stopping and restarting.

Requesting continuation (1/3)...
 ─  ⚕ Hermes  ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────── 
                                                                                                                                                                                                                   
     技能 v6.0.0 完成。融合了来自 AGENT.md 的内容：                                                                                                                                                                
                                                                                                                                                                                                                   
     1.  决策树 — 根据你需要完成的任务，指明应使用哪个工具                                                                                                                                                         
     2.  推荐会话流程 — 路径调用优先，然后无路径调用，并明确指出第一步“设置活动 db”                                                                                                                                
     3.  路径调用与无路径调用规则 — 明确的假设：调用路径工具优先设置活动 db，然后无路径工具在该 db 上运行。注明确认这有待 MCP 作者确认。                                                                           
     4.  常见错误 — 来自 AGENT.md，包括“不要跳过学习检查”、“不要对概念使用文本搜索”、“不要不必要地强制重新学习”                                                                                                    
     5.  contribute_insights 内容结构 — 现在包括来自 AGENT.md 的 architectural_decision 类型                                                                                                                       
     6.  所有三个来源都已归功于顶部The full skill edit was applied successfully. Nothing was truncated — v6.0.0 is complete with all three source documents incorporated.

That seems in part associated with me burning so many of my precious cloud tokens getting hermes over the line writing a skill based upon sources for the mcp server.

<!-- gh-comment-id:4321400943 --> @Bazmundi commented on GitHub (Apr 26, 2026): Oh dear, still fiddling with skills after stopping and restarting. ``` Requesting continuation (1/3)... ─ ⚕ Hermes ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────── 技能 v6.0.0 完成。融合了来自 AGENT.md 的内容： 1. 决策树 — 根据你需要完成的任务，指明应使用哪个工具 2. 推荐会话流程 — 路径调用优先，然后无路径调用，并明确指出第一步“设置活动 db” 3. 路径调用与无路径调用规则 — 明确的假设：调用路径工具优先设置活动 db，然后无路径工具在该 db 上运行。注明确认这有待 MCP 作者确认。 4. 常见错误 — 来自 AGENT.md，包括“不要跳过学习检查”、“不要对概念使用文本搜索”、“不要不必要地强制重新学习” 5. contribute_insights 内容结构 — 现在包括来自 AGENT.md 的 architectural_decision 类型 6. 所有三个来源都已归功于顶部The full skill edit was applied successfully. Nothing was truncated — v6.0.0 is complete with all three source documents incorporated. ``` That seems in part associated with me burning so many of my precious cloud tokens getting hermes over the line writing a skill based upon sources for the mcp server.
GiteaMirror commented 2026-05-10 06:20:55 -05:00
Author
Owner
Copy Link

@Bazmundi commented on GitHub (Apr 26, 2026):

Given the skill structure hermes supposedly uses at:
https://www.mintlify.com/NousResearch/hermes-agent/user-guide/features/skills

And providing information to the autoskill create a skill based upon:
https://github.com/pi22by7/In-Memoria/blob/main/AGENT.md
https://github.com/pi22by7/In-Memoria/tree/main/.github/chatmodes
https://github.com/pi22by7/In-Memoria/blob/main/.github/copilot-instructions.md

The skill it creates is curiously flat and dumb (even with the information given). I appear to not just need to provide the links but instruct it to read the links and probably needed to instruct it to build the skill based upon the info in the links. Yes, I am aware of deontic logic and so vibe coding of skill even with a skill for authoring skills needs more than less deontics.

Last chance, I gave it the following so it understood I understood it should have understood how to develop its own skills:
https://www.mintlify.com/NousResearch/hermes-agent/user-guide/features/skills

May yes be simply as I am using glm-5.1:cloud under ollama launch hermes.

Half a lazy sunday lost and I've still to get to the place hermes can cope with calling the mcp since it (or the llm) cannot as yet, hermes still wants to go straight out to grep or other skills I have not specified.

Is this a "pick the right model to write skills problem"?

<!-- gh-comment-id:4321425199 --> @Bazmundi commented on GitHub (Apr 26, 2026): Given the skill structure hermes supposedly uses at: https://www.mintlify.com/NousResearch/hermes-agent/user-guide/features/skills And providing information to the autoskill create a skill based upon: https://github.com/pi22by7/In-Memoria/blob/main/AGENT.md https://github.com/pi22by7/In-Memoria/tree/main/.github/chatmodes https://github.com/pi22by7/In-Memoria/blob/main/.github/copilot-instructions.md The skill it creates is curiously flat and dumb (even with the information given). I appear to not just need to provide the links but instruct it to read the links and probably needed to instruct it to build the skill based upon the info in the links. Yes, I am aware of deontic logic and so vibe coding of skill even with a skill for authoring skills needs more than less deontics. Last chance, I gave it the following so it understood I understood it should have understood how to develop its own skills: https://www.mintlify.com/NousResearch/hermes-agent/user-guide/features/skills May yes be simply as I am using glm-5.1:cloud under ollama launch hermes. Half a lazy sunday lost and I've still to get to the place hermes can cope with calling the mcp since it (or the llm) cannot as yet, hermes still wants to go straight out to grep or other skills I have not specified. Is this a "pick the right model to write skills problem"?
GiteaMirror commented 2026-05-10 06:20:56 -05:00
Author
Owner
Copy Link

@ghmer commented on GitHub (Apr 27, 2026):

They guy that created open claw suggested Claude Opus for a reason. That being said, I had some okayish results using qwen- or the minimax models. "Okayish" being emphasized.

<!-- gh-comment-id:4323465981 --> @ghmer commented on GitHub (Apr 27, 2026): They guy that created open claw suggested Claude Opus for a reason. That being said, I had some okayish results using qwen- or the minimax models. "Okayish" being emphasized.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#87769
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?