Raven Scott
f5e9473f7b
Add second-wave agent capabilities for initd/journal/network/IPC diagnostics, allowlisted maintenance + contract-check automation, and policy-gated HRPC bridge actions with emergency-stop controls. Harden xAI provider support by applying provider-aware request defaults, parsing reasoning-summary/tool-call streaming variants, and updating seeded docs/skills/tests to reflect max-autonomy safeguards and trace-first observability.
38 lines
1.4 KiB
Markdown
38 lines
1.4 KiB
Markdown
---
|
|
name: xai-compat
|
|
version: 1.0.0
|
|
description: Keep agent requests and tool loops compatible with xAI chat/responses semantics, reasoning summaries, and streaming behavior.
|
|
tags: [xai, compatibility, streaming, tools, reasoning]
|
|
requires: [edit_agent_config, run_command]
|
|
---
|
|
|
|
# xai-compat
|
|
|
|
Use this skill when provider is `xai` or when debugging xAI tool/reasoning behavior.
|
|
|
|
## Baseline configuration
|
|
|
|
1. Ensure:
|
|
- `provider = "xai"`
|
|
- `rest_base_url = "https://api.x.ai/v1"`
|
|
2. Prefer `reasoning_mode = "trace"` for observability when summaries are sparse.
|
|
3. Keep `stream = true` for tool-call visibility and progressive diagnostics.
|
|
|
|
## Tool calling expectations
|
|
|
|
- xAI supports OpenAI-compatible function tools on chat completions.
|
|
- `parallel_tool_calls` can be enabled; process all returned tool calls before continuing.
|
|
- Keep tool schemas explicit and bounded.
|
|
|
|
## Reasoning expectations
|
|
|
|
- Reasoning models may expose `reasoning_content` summaries, but availability can vary by model/endpoint.
|
|
- In trace mode, rely on local process lines even when no reasoning summary deltas arrive.
|
|
|
|
## Debug checklist
|
|
|
|
1. Verify provider/base URL in `~/.agent/config.json`.
|
|
2. Run with `reasoning_mode=trace`.
|
|
3. Check for tool-call chunks and usage tokens in stream output.
|
|
4. If reasoning deltas are absent, treat this as provider/model behavior and continue with trace-level progress lines.
|