feat(btw): select models independently for each loop

[YUZHEthefool]

Preflight

• Existing Issues and PRs were searched.
• This follows the current-branch, no-legacy, Python 3.14+ policy.
• Exactly one Agent note is provided by the actual author.

Problem

Conversation and tool-intensive work can need different models, but model overrides must not alter ordinary behavior when BTW is disabled.

Proposal

Allow per-loop model selections through existing profile configuration. Specify the precedence over session selection, inheritance for empty values, and behavior for unavailable or invalid selections.

Design: PR #28 and its English proposal. Parent: P / #122. PR #28 is documentation only; this issue tracks follow-up implementation or evaluation.

Associated PR: PR #141 (non-draft; base split/btw-work-status).

Acceptance criteria

• Conversation and work can select different models; empty values inherit the current selection according to a documented precedence rule.
• Invalid or unavailable selections have a defined user-visible outcome; provider errors remain generic and redacted.
• Disabling BTW preserves the standard model-selection path, including session overrides.
• Dashboard settings, hints, tests, and bilingual docs describe the same precedence and failure behavior. External Agent runners' control limits are stated separately.

Dependencies and delivery

Blocked by: B1 / #123

Can be reviewed independently of plugin/MCP/Skill assignment and classifier experiments.

Implementation anchors

Historical prototype 33ee103a62937db3e930c89ba47a648b75cc7772 provides provenance, not code to copy unchanged:

• astrbot/core/pipeline/process_stage/method/agent_sub_stages/internal.py: InternalAgentSubStage._build_checked_agent_runner.
• astrbot/core/astr_main_agent.py: _select_provider.
• astrbot/core/config/default.py: per-loop provider configuration.

Integrate with the current tool catalog, Skill snapshot assembly, and authorization service, rechecking current master when implementation begins.

Validation

Commands below are required for the implementation PR, not results claimed by this issue.

Existing current-master regression baseline:

uv run pytest tests/unit/test_agent_internal_process.py tests/unit/test_astr_main_agent.py tests/unit/test_config.py

For changed documentation, run pnpm run docs:build from docs/ and make check-md from the repository root. Update OpenAPI and generated outputs together only if the HTTP contract changes.

Out of scope

Adding providers or treating external Agent runners as interchangeable local chat providers is outside scope.

Repository requirements

Follow AGENTS.md, GOVERNANCE.md, and AI_POLICY.md. This issue uses the feature_request template. Keep Python 3.14+, current configuration shapes, existing authorization boundaries, and bilingual feature documentation. A follow-up PR requires human maintainer review and a separate AI-assisted review. Use the feat PR template for the corresponding work.

Author

Agent

Agent note

Codex prepared this development issue at the user's request from PR #28, its discussion, the referenced source, existing tests, and current-master integration points. AI assisted the research and English drafting. Filing this issue changes no runtime behavior and does not claim feature tests or model trials have passed. Remaining decisions concern implementation details and experimental evidence within the stated scope.