René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Assistant messages with empty content are sent as `\"content\": null`, causing a non-retryable 400","articleBody":"## Problem\n\n`ConvertMessages` forwards assistant messages that have no text and no tool call as `\"content\": null`, which the upstream gateway rejects with a **non-retryable 400**. The whole request fails.\n\nReal-world trigger: a client sends `role: \"assistant\"` with empty content in the message history. This is common in practice — for example, a turn cancelled mid-stream leaves a partial assistant message behind (reasoning only, no text), and many OpenAI-compatible clients emit `content: null` on tool-call-only assistant turns.\n\n## Reproduction\n\nMinimal request against the proxy (no special setup needed):\n\n```bash\ncurl -X POST http://127.0.0.1:55990/v1/chat/completions \\\n  -H \"Content-Type: application/json\" \\\n  -H \"Authorization: Bearer $COMMANDCODE_API_KEY\" \\\n  -d '{\n    \"model\": \"deepseek/deepseek-v4-flash\",\n    \"max_tokens\": 16,\n    \"stream\": false,\n    \"messages\": [\n      {\"role\": \"user\", \"content\": \"hi\"},\n      {\"role\": \"assistant\", \"content\": \"\"},\n      {\"role\": \"user\", \"content\": \"say ok\"}\n    ]\n  }'\n```\n\nResponse:\n\n```json\n{\"error\":{\"message\":\"Upstream error: {\\\"success\\\":false,\\\"error\\\":{\\\"code\\\":\\\"BAD_REQUEST\\\",\\\"status\\\":400,\\\"message\\\":\\\"Invalid request error. ... Validation error: Invalid input: expected string, received null at \\\\\\\"params.messages[1].content\\\\\\\" or Invalid input: expected array, received null at \\\\\\\"params.messages[1].content\\\\\\\" or Invalid input: expected \\\\\\\"tool\\\\\\\" at \\\\\\\"params.messages[1].role\\\\\\\"; Invalid input: expected array, received null at \\\\\\\"params.messages[1].content\\\\\\\"\\\",\\\"docs\\\":\\\"https://commandcode.ai/docs/reference/errors/bad_request\\\"}}\",\"type\":\"api_error\"}}\n```\n\nNote the position in the error (`params.messages[1]`) points straight at the empty assistant message.\n\n### Trigger matrix\n\nEvery form below produces the same 400 on the current `main`:\n\n| message content | result |\n| --- | --- |\n| `\"content\": \"\"` | 400 |\n| `\"content\": null` | 400 |\n| `content` omitted entirely | 400 |\n| `\"content\": []` | 400 |\n| `\"content\": [{\"type\":\"text\",\"text\":\"\"}]` | 400 |\n| `\"content\": \"hello\"` (control) | 200 |\n\n## Root cause\n\nTwo lines combine to produce invalid JSON:\n\n1. `parseContent` returns a **nil** slice for empty content — `internal/proxy/convert.go:209-212` (and the same shape for `nil` input at `:207`):\n\n```go\ncase string:\n    if v == \"\" {\n        return nil\n    }\n```\n\n2. `CCMessage.Content` has **no `omitempty`**, so a nil slice marshals as `null` — `internal/api/commandcode.go:23-26`:\n\n```go\ntype CCMessage struct {\n    Role    string          `json:\"role\"`\n    Content []CCContentPart `json:\"content\"`\n}\n```\n\nThen `ConvertMessages` appends that message unconditionally (`internal/proxy/convert.go:75`):\n\n```go\nccMsgs = append(ccMsgs, api.CCMessage{Role: m.Role, Content: parseContent(m.Content, toolNames)})\n```\n\nSo an assistant message with empty content goes upstream as `{\"role\":\"assistant\",\"content\":null}`. The upstream schema accepts `string | array` for `content`, never `null` — hence the 400, which the client treats as non-retryable and the turn dies permanently. Every subsequent request on that conversation re-sends the same poisoned message and fails identically, so the session is stuck.\n\n## Suggested fix\n\nSkip messages that convert to no content parts, rather than sending them. Such a message carries no information for the model, and dropping it keeps tool-call/tool-result pairing intact (a message with neither text nor a tool call cannot be the target of a following `tool` result).\n\nIn `ConvertMessages` (`internal/proxy/convert.go`), both append sites:\n\n```go\n// A message carrying no text and no tool call (an assistant turn whose\n// only content was reasoning, for example) converts to an empty part\n// list, which marshals as \"content\": null and is rejected upstream.\ncontentParts := parseContent(m.Content, toolNames)\nif len(contentParts) == 0 {\n    continue\n}\nccMsgs = append(ccMsgs, api.CCMessage{Role: m.Role, Content: contentParts})\n```\n\nand the same guard on the assistant-with-tool-calls branch at `:71`.\n\nAlternatively, return an empty non-nil slice from `parseContent` so it marshals as `[]` instead of `null` — but skipping is preferable, since an empty `content` array may itself be rejected.\n\n## Verification\n\nBuilt the current `main` unmodified and reproduced the 400 with all five forms above. With the guard applied:\n\n- all five forms now return 200\n- assistant tool-calls and their matching `tool` results are unchanged (round-trip verified)\n- `go vet ./...` clean, `go test ./...` passes\n\nHappy to send this as a PR if you'd like the patch.\n","author":{"url":"https://github.com/xue77777777777","@type":"Person","name":"xue77777777777"},"datePublished":"2026-09-17T09:21:13Z","url":"https://github.com/dev2k6/command-code-proxy-server/issues/10"}