feat: system prompt, streaming usage, stdin prompt & model updates#16
Open
smartchainark wants to merge 10 commits intoatalovesyou:mainfrom
Open
feat: system prompt, streaming usage, stdin prompt & model updates#16smartchainark wants to merge 10 commits intoatalovesyou:mainfrom
smartchainark wants to merge 10 commits intoatalovesyou:mainfrom
Conversation
The OpenAI Chat Completions API allows message content to be either a
plain string or an array of content parts (e.g.
[{type: "text", text: "..."}]). When an upstream client sends array
content, the previous code passed it directly to template literals,
resulting in "[object Object]" being sent to Claude CLI.
Add extractContent() helper that normalises both string and array
formats into a single string, and update the OpenAIChatMessage type to
reflect the actual API spec.
Adds support for passing system messages to Claude Code CLI via the --append-system-prompt flag. This enables OpenClaw agents to properly send their identity, backstory, and instructions through the proxy. Changes: - Support both "system" and "developer" message roles (OpenAI spec) - Extract system messages separately from conversation messages - Pass system prompt via --append-system-prompt CLI flag - Add optional tools parameter for future tool restriction support This fixes the issue where OpenClaw system messages (containing agent identity and configuration) were being ignored, causing Claude to respond with its default identity instead of the configured persona. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Add console.error logging for system prompt tracking - Add --dangerously-skip-permissions flag for service operations - Log assistant messages and results for debugging Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Replace all console.error calls with conditional debug() method - Debug logging disabled by default, enabled via DEBUG_SUBPROCESS=true - Add documentation in README troubleshooting section - Keeps production logs clean while allowing detailed debugging when needed Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Without usage data in the final streaming chunk, downstream consumers (e.g. OpenClaw compaction) cannot track token consumption, causing sessions to grow unbounded and eventually hit context overflow. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When Claude CLI returns a rate limit response, modelUsage can be an
empty object causing Object.keys({})[0] to return undefined, which
crashes normalizeModelName.
Credit: PR atalovesyou#7 by @wende
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Large prompts (conversation history + system prompt) can exceed the OS argument size limit, causing spawn() to fail with E2BIG. Passing via stdin has no size limitation. Credit: PR atalovesyou#12 by @kevinfealey Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add model mappings for claude-opus-4-6, claude-sonnet-4-5, and claude-max/* provider prefix used by OpenClaw. Also add opus-max and sonnet-max aliases. Credit: PR atalovesyou#10 by @jamshehan (model mapping portion) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Rebrand as actively maintained fork of atalovesyou/claude-max-api-proxy - Document new models (Opus 4.6, Sonnet 4.5) and model aliases - Add "What's Different" comparison table - Add OpenClaw, Continue.dev, and Python client configuration examples - Add LaunchAgent setup instructions - Update architecture notes for stdin-based prompt delivery - Credit community PR contributors Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Keep original README content intact, only add fork-specific changes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
bwiedmann
added a commit
to bwiedmann/claude-max-api-proxy
that referenced
this pull request
Feb 15, 2026
…e, stdin prompt & model updates # Conflicts: # README.md # src/adapter/openai-to-cli.ts # src/subprocess/manager.ts
wende
added a commit
to wende/claude-max-api-proxy
that referenced
this pull request
Feb 18, 2026
Triaged all 14 open PRs from atalovesyou/claude-max-api-proxy, implemented the valuable fixes, and added end-to-end test coverage. Changes: - Fix normalizeModelName crash on undefined model (atalovesyou#7 regression) - Pass prompt via stdin instead of CLI arg to avoid E2BIG (atalovesyou#12) - Increase subprocess timeout from 5 to 15 minutes (atalovesyou#20) - Add Claude 4.5/4.6 model IDs and claude-max/ prefix (atalovesyou#10, atalovesyou#20) - Include usage data in final streaming SSE chunk (atalovesyou#16) - Wrap subprocess logging with DEBUG_SUBPROCESS env check (atalovesyou#5, atalovesyou#16) - Strip CLAUDECODE env var from subprocesses (own fix) - Add e2e test suite (7 tests covering health, models, completions)
3 tasks
wende
added a commit
to wende/claude-max-api-proxy
that referenced
this pull request
Feb 18, 2026
Triaged all 14 open PRs from atalovesyou/claude-max-api-proxy, implemented the valuable fixes, and added end-to-end test coverage. Changes: - Fix normalizeModelName crash on undefined model (atalovesyou#7 regression) - Pass prompt via stdin instead of CLI arg to avoid E2BIG (atalovesyou#12) - Increase subprocess timeout from 5 to 15 minutes (atalovesyou#20) - Add Claude 4.5/4.6 model IDs and claude-max/ prefix (atalovesyou#10, atalovesyou#20) - Include usage data in final streaming SSE chunk (atalovesyou#16) - Wrap subprocess logging with DEBUG_SUBPROCESS env check (atalovesyou#5, atalovesyou#16) - Strip CLAUDECODE env var from subprocesses (own fix) - Add e2e test suite (7 tests covering health, models, completions) Co-Authored-By: kevinfealey <10552286+kevinfealey@users.noreply.github.com> Co-Authored-By: Max <257223904+Max-shipper@users.noreply.github.com> Co-Authored-By: James Hansen <1359077+jamshehan@users.noreply.github.com> Co-Authored-By: bitking <213560776+smartchainark@users.noreply.github.com> Co-Authored-By: Alex Rudloff's AI Agents <258647843+alexrudloffBot@users.noreply.github.com>
chrisle
added a commit
to chrisle/claude-max-api-proxy
that referenced
this pull request
Feb 27, 2026
Users can now send large prompts without E2BIG errors, use array-based content format for multimodal support, properly handle system prompts, and receive token usage data in streaming responses. Model names with any provider prefix are now correctly normalized. Changes: - Fix array content serialization preventing [object Object] bugs - Pass prompts via stdin to prevent E2BIG errors with large requests - Add proper system prompt support via --append-system-prompt flag - Include usage data in final streaming chunks - Support any provider prefix in model names (claude-max/, etc.) - Add CLAUDE_DANGEROUSLY_SKIP_PERMISSIONS env var for service mode - Update to Claude 4.5/4.6 model support Incorporates improvements from PRs atalovesyou#12, atalovesyou#13, atalovesyou#16, atalovesyou#24, atalovesyou#27 at https://github.com/atalovesyou/claude-max-api-proxy
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Consolidated improvements from our actively maintained fork. All changes are production-tested with OpenClaw.
Bug fixes
usage(prompt_tokens, completion_tokens, total_tokens). Downstream consumers relying on token counts (e.g. session compaction) could not track consumption.modelUsagecan be{}, causingObject.keys({})[0]to return undefined and crash. (Credit: PR fix: handle undefined model in normalizeModelName to prevent crash on rate limits #7 by @wende)contentas both string and array of content parts (OpenAI multi-part message format).New features
--append-system-promptflag instead of embedding in user prompt.claude-opus-4-6,claude-sonnet-4-5,claude-max/*prefix, andopus-max/sonnet-maxaliases. (Credit: PR feat: add support for Claude 4.5/4.6 models and claude-max prefix #10 by @jamshehan, model mapping portion)Docs
Test plan
🤖 Generated with Claude Code