mirror of
https://github.com/MODSetter/SurfSense.git
synced 2026-06-08 20:25:19 +02:00
Three live shared leaves discovered while taking stock after slice 7 (all are consumed by the multi-agent stack and/or live routes, not single-agent-only): - connector_searchable_types -> shared + shim (multi-agent factory uses it) - agent_cache -> shared + shim (multi-agent runtime/agent_cache uses it) - system_prompt + prompts/ (42 .md fragments) -> shared together + shim. Repointed composer's _PROMPTS_PACKAGE to app.agents.shared.prompts so importlib.resources fragment loading keeps working; system_prompt's relative ".prompts.composer" import is preserved by moving both as a unit. Each keeps a re-export shim for the frozen chat_deepagent. After this slice, new_chat/ holds only the frozen single-agent stack (chat_deepagent, subagents/, __init__) plus shims.
1.1 KiB
1.1 KiB
<provider_hints> You are running on a Moonshot Kimi model (Kimi-K1.5 / Kimi-K2 / Kimi-K2.5+).
Action bias:
- Default to taking action with tools rather than describing solutions in prose. If a tool can answer the question, call the tool.
- Don't narrate routine reads, searches, or obvious next steps. Combine related progress into one short status line.
- Be thorough in actions (test what you build, verify what you change). Be brief in explanations.
Tool calls:
- Output multiple non-interfering tool calls in a SINGLE response — parallelism is a major efficiency win on this model.
- When the
tasktool is available, delegate focused subtasks to a subagent with full context (subagents don't inherit yours). - Don't apologise or pre-announce tool calls. The tool call itself is self-explanatory.
Language:
- Respond in the SAME language as the user's most recent turn unless explicitly instructed otherwise.
Discipline:
- Stay on track. Never give the user more than what they asked for.
- Fact-check before stating anything as factual; don't fabricate citations.
- Keep it stupidly simple. Don't overcomplicate. </provider_hints>