fix: normalize non-standard errors from anthropic-format 3rd-party APIs

[brotherlang8888]

Summary

• Routes anthropic-format providers through the local proxy (passthrough mode) instead of bypassing it
• Normalizes non-standard upstream error responses to proper Anthropic format
• Prevents conversation history corruption and cascade failures

Problem

When using third-party Anthropic-compatible APIs like step-router-v1, any API error (image reading, deferred tools, unrecognized tool calls) returns non-standard error shapes:

{"error":{"message":"The input you provided is invalid","type":"input_invalid"}}

Since anthropic-format providers bypassed the proxy, these raw errors went straight to cc-haha's error handler. The handler produced a malformed AssistantMessage that got injected into conversation history. On subsequent requests, the corrupted history was re-sent to the API → another error → infinite cascade. Only /clear could recover.

Fix

• handler.ts: Remove the "proxy not needed" rejection for anthropic format. Add handleAnthropicPassthrough that forwards requests/responses through unchanged, but intercepts error responses and normalizes them via normalizeAnthropicError() to:

  {"type":"error","error":{"type":"invalid_request_error","message":"..."}}

• providerService.ts: Route anthropic-format providers through the proxy (change needsProxy condition from apiFormat != null && apiFormat !== 'anthropic' to apiFormat != null)

Test plan

• 41 proxy transform + streaming tests pass
• Test with step-router-v1: trigger image error → text follow-up should work (no cascade)
• Test with step-router-v1: trigger ToolSearch/deferred tool error → text follow-up should work

Additional requests

1. Skills not loading after restart: After restarting the app, user-installed skills are no longer loaded. This may be a separate issue worth investigating.

2. OpenAI format support: Would it be possible to add an OpenAI-format proxy mode that converts Anthropic messages ↔ OpenAI Chat Completions format? This would help providers that only support OpenAI-compatible APIs.

🤖 Generated with Claude Code

[github-actions]

PR quality triage

Changed areas: area:server

CLI core policy: No CLI-core policy block detected.

Missing-test policy: Blocked by policy until a maintainer applies allow-missing-tests or matching tests are added.

Coverage baseline policy: No coverage-baseline policy block detected.

CLI core files:

• none

Coverage policy files:

• none

Expected checks:

• change-policy
• desktop-checks
• server-checks
• desktop-native-checks
• coverage-checks

Test coverage signals:

• BLOCKING unless allow-missing-tests is applied: Server product files changed without a server test file in the PR.

Risk notes:

• Provider/search behavior changed: PR gate uses mock tests; live-provider tests stay maintainer-only.

Hard merge gates still come from GitHub Actions, not AI review.

Dosu handoff: Dosu can be used as the AI reviewer for risk explanation, missing-test prompts, and maintainer Q&A. If it does not comment automatically from the PR template, ask:

@dosubot review this PR for changed-area risk, missing tests, docs impact, desktop startup risk, and CLI core impact.

[brotherlang8888]

这是来自QQ邮箱的假期自动回复邮件。收到了你的邮件，请确认你的发送时间