Feedback: MCP tools not called by DeepSeek V4, reasoning_content on tool calls, jj colocated
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Newbie friendliness
- 43/100
- Issue type
- Feature
- Clarity
- Mostly clear
- Activity status
- Active
- Tech stack
- rust
- Domain
- ai-infra-agents, tooling
Research direction
Start by reading the MCP wrapper and system prompt construction, then inspect the provider adapter’s handling of DeepSeek V4 tool-call history and retry behavior. The issue suggests exposing tool names and descriptions or documenting a steering hint, and preserving or guarding against missing reasoning_content. Done means MCP tools can be called reliably and thinking-mode tool calls do not cause repeated HTTP 400 retries.
Written by the indexing model from the issue text.
Description
Feedback from a week of using infinity with DeepSeek V4 and jj. One small website task, not a benchmark.
- MCP tools were not called by the model. Through the MCP wrapper the model only sees
<server>_list_toolsand<server>_invoke_tool. Without a hint in the system prompt, DeepSeek V4 did not call them. We worked around it by appending an addendum to the system prompt on each request. Exposing tool names and descriptions directly, or a documented steering hint, would avoid this. - DeepSeek V4 in thinking mode. Once tool calls are in the history, the API expects all earlier
reasoning_contentback. Otherwise it returns HTTP 400 and the agent retries in a loop. We disabled thinking in our provider adapter, which costs quality on hard tasks. A note in the provider docs or a guard against the retry loop would help. - jj colocated with Git worked well (
jj git init --colocate, bookmarks for the sandbox result,jj workspace addto inspect it).git logand GitHub kept working.
- Dominant language
- Rust
- Stars
- 716
- Forks
- 81
- Avg merge
- 1d 18h
- Merged PRs (30d)
- 11
Getting set up
This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from hydro-project/infinity
-
Difficulty 2/5 1-3 hours Newbie friendliness 64/100
hydro-project/infinity#189 ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 84/100
hydro-project/infinity#187 ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
hydro-project/infinity#174 · 1 comment ·
Maintainers usually reply within 1 day
-
Difficulty 4/5 3-5 days Newbie friendliness 45/100
hydro-project/infinity#190 ·
Maintainers usually reply within 1 day
-
Difficulty 4/5 3-5 days Newbie friendliness 55/100
hydro-project/infinity#177 ·
Maintainers usually reply within 1 day
All issues in hydro-project/infinity
Similar issues
-
binding:python bug
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
Maintainers usually reply within 1 day
-
enhancement
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
coollabsio/orbit#72 ·
Maintainers usually reply within 1 day
-
Synth suite fails unless run from the repo root: signal_flow's default machine path is cwd-relativeOpenbug
Difficulty 2/5 1-3 hours Newbie friendliness 80/100
Nimblesite/osprey#370 ·
Maintainers usually reply within 1 day
-
enhancement
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
-
bug order-flow studies ui
Difficulty 2/5 1-3 hours Newbie friendliness 88/100
userFRM/egui-charts#2417 ·
Maintainers usually reply within 1 day