fix(footer): show runtime provider, not configured primary, in OpenClaw footer

[madaxia]

Problem

The footer displayed model.provider from config instead of the provider that actually answered the turn. After a fallback switch this was wrong — e.g. showing Provider: kimi while the agent ran volces-agent.

Root cause

agent/turn_finalizer.py invoked the transform_llm_output hook without passing provider, so the runtime-footer plugin fell back to config.model.provider. gateway/run.py likewise did not propagate the runtime provider into the agent_result dicts consumed by the built-in footer path.

Changes (scoped to provider propagation only)

Per triage feedback, this PR is now scoped strictly to the provider-propagation bug. The broader OpenClaw footer formatting / field-behavior work stays in #66372.

• agent/turn_finalizer.py — pass provider=getattr(agent, "provider", None) and the agent id into the transform_llm_output hook call. (root fix)
• gateway/run.py — include "provider": getattr(_agent, "provider", None) in both agent_result dicts; pass provider to build_footer_line.
• gateway/runtime_footer.py — add a provider param to format_runtime_footer and build_footer_line; falls back to config.model.provider only when the runtime value is missing (legacy call sites stay populated).
• tests/gateway/test_runtime_footer.py — +4 cases: runtime provider wins over config, config fallback, omitted-when-both-missing.

Verification

• venv/bin/python -m pytest tests/gateway/test_runtime_footer.py -q → 29 passed
• End-to-end: with provider="volces-agent" propagated, the footer now shows volces-agent instead of kimi (the configured primary).

Relationship to #66372 / #62277

This PR is now orthogonal to both: it only fixes the provider-display root cause (the transform_llm_output hook not receiving the runtime provider). OpenClaw footer formatting and field-behavior enhancements remain in #66372. #62277 (CLI status-bar provider display) is a separate surface.

[alt-glitch]

This was generated by AI during triage.

Related to #66372 and #62277. Runtime-provider propagation fixes a distinct fallback-display bug, while this branch also adds broader OpenClaw footer formatting and field behavior; please choose or split the overlapping footer work.

[madaxia]

Thanks for the triage. Good call on the scope — I've narrowed this PR to only the runtime-provider propagation fix (the distinct fallback-display bug), and dropped the broader OpenClaw footer formatting / field-behavior changes so they stay in #66372.

What remains in this PR (b31091a, +93/-0):

• agent/turn_finalizer.py: pass provider (+ agent id) into the transform_llm_output hook — the root cause; the runtime-footer plugin was falling back to config.model.provider because the hook never received the runtime value.
• gateway/run.py: propagate provider into both agent_result dicts and pass it to build_footer_line.
• gateway/runtime_footer.py: add a provider param (with config.model.provider fallback for legacy call sites) — no style/duration/field-alias changes.
• tests/gateway/test_runtime_footer.py: +4 cases (runtime-wins-over-config, config fallback, omitted-when-both-missing).

What moved out (openclaw style, duration, ctx/cwd_label aliases, labelled Agent: | Model: | Provider: rendering) stays in #66372, which is the OpenClaw-footer PR.

This PR is now orthogonal to both #66372 and #62277. venv/bin/python -m pytest tests/gateway/test_runtime_footer.py -q → 29 passed.

[teknium1]

Thanks for splitting the provider-propagation work out of the broader footer changes. The core premise is real: fallback activation updates agent.provider in agent/chat_completion_helpers.py:1877-1879, while current main drops that value before the footer at gateway/run.py:23175, gateway/run.py:23302, and gateway/run.py:14710-14717.

Problems

• provider is newly accepted as a footer field, but website/docs/user-guide/configuration.md:1658-1665 still documents only model, context_pct, and cwd.
• The added tests call build_footer_line() directly. They do not cover either changed result-dict path or the final gateway call site that propagates the value.
• agent/turn_finalizer.py also adds an undocumented agent argument to the public transform_llm_output hook. The documented callback contract at website/docs/user-guide/features/hooks.md:1257-1275 does not include it, and it is outside the stated provider-only scope.

Suggested changes

• Document the new field and add a gateway-level fallback regression test.
• Remove the agent hook kwarg unless its contract is documented and independently covered.

Automated hermes-sweeper review.

[teknium1]

transform_llm_output is a public plugin hook, but its documented callback contract currently has no agent kwarg (website/docs/user-guide/features/hooks.md:1257-1275). Please remove this scope expansion or document and test the new hook payload separately.

[teknium1]

Please update the runtime-footer configuration reference: it currently declares only model, context_pct, and cwd as supported fields (website/docs/user-guide/configuration.md:1658-1665).

[jimmywu0816]

Independent provider-only implementation and verification

I reproduced and implemented the provider-propagation slice independently against the current upstream main (as of today).

Current main still drops the live runtime provider before the built-in footer render path: agent.provider is updated on fallback activation, but agent_result only carried model/tokens, so a footer configured with the provider never renders it. The minimal, byte-stable change is:

• gateway/runtime_footer.py — add an opt-in provider field (legacy default fields model, context_pct, cwd untouched; field skipped when the runtime value is missing).
• gateway/run.py — capture provider from the same live agent instance as model on both result paths and pass it to build_footer_line().
• tests/gateway/test_runtime_footer.py — field order, missing-provider omission, and the public build_footer_line() path.
• English + Simplified Chinese configuration docs — document the new field.

Validation on current upstream main + this minimal patch:

uv run --extra dev pytest -q tests/gateway/test_runtime_footer.py
40 passed in 2.02s
git diff --check
clean

Real-world Telegram output with the field enabled (default profile, fallback chain active in the second example):

default · deepseek · deepseek-v4-flash · 12%
default · openrouter · gpt-5.6-luna · 29%

This is intentionally narrower than the token/cost/quota expansion in #35427 / #18188. Since this PR already targets the same runtime-provider propagation problem (including the agent_result provider propagation), I did not open a duplicate PR — please consider folding the minimal documentation and field-order coverage into this branch or the canonical footer-expansion work.