pal-mcp-server · Issues· 145 open

Large non-streaming requests to reasoning models deterministically 504 at ~600s

consensus sends no max_tokens — OpenRouter default (131072) exceeds kimi-k2-thinking's actual backend caps (400)

LICENSE body is not the canonical Apache 2.0 text — §2 omits the sublicense, public display and public performance grants

MCP spec conformance: 10 requirement(s) violated (via @hasmcp/mcp-spec-test)

Crash on install: unpinned mcp>=1.0.0 resolves to mcp 2.0.0 which removed @server.list_tools()

Bug: ConsensusTool concurrent callers overwrite each other's proposal state (singleton mutable state)

Add Antigravity (agy) CLI as a clink client

OpenRouter reasoning models: chat-completions path never sends the `reasoning` param and drops `reasoning_details`

clink always injects “Gemini CLI agent” guidance text regardless of selected cli_name

Bug: chat continuation history is rebuilt twice when server injects CONTINUATION header

PAL-BUG-3 (NEW W23): `qwen/qwen-2.5-coder-32b-instruct` returns content but OR `/v1/generation/{id}` returns all-null fields

PAL-BUG-2: `pal__analyze` intermittently returns `empty_response` after long body-read

PAL-BUG-1: `pal__planner` is orchestration-only — no upstream model call is made

[Bug] clink "codex: command not found" when Codex CLI is installed globally via nvm + npm

Z.AI API Integration and GLM Model Support