Repository navigation
docs(proxy): verify four CLI feasibility rows against a live proxy - #1459
Conversation
The audit left Gemini, Amp, Hermes and Grok unresolved, three of them on documentation rather than evidence. Cursor is the reason that matters: it read as the cleanest win in the matrix until someone pointed it at a real proxy and found the env vars were inert. Same method here — start a proxy, point the documented override at it, run one command, see what arrives. Gemini CLI's override is real. Pointed at the proxy it reached it and failed with ModelNotFoundError: 404, which is the correct answer from a proxy that has no generateContent route. The redirect works; the door is the whole remaining job, so §4's list is complete rather than optimistic. Two traps recorded for whoever builds it: the CLI refuses to run outside a trusted directory, and it issues a generateJson call at startup, so the door must answer more than the user's turn. Amp's AMP_URL is honoured too — it built its login URL against the proxy rather than ignoring the variable as Cursor did. But that login URL is the finding: Amp expects an OAuth-style CLI flow at its own endpoint before any API traffic, so onboarding it means implementing an auth surface, not writing a config file. That moves it out of the config-writer class and in beside Gemini. Its bundle vendors Google's GenAI SDK, so generateContent strings inside it describe a dependency, not Amp's wire — noted so the next person grepping does not conclude otherwise. Hermes and Grok cannot be verified here at all: neither is installed, and the "easy" verdict for Hermes came from documentation. The blocker is availability rather than difficulty, and recording that is more useful than leaving the rows looking merely undecided.
|
Warning Review limit reached
Next review available in: 38 minutes Limit details: You’ve used all 2 included reviews currently available. You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. How can I continue?Wait for the limit to reset, then comment An organization admin can change what happens after included review limits in Billing. How do review limits work?CodeRabbit enforces per-developer PR review limits within each organization. For paid Pro and Pro+ reviews, CodeRabbit uses a developer's included PR review attempts over the past 7 days to set the current hourly allowance. At typical activity levels, the full plan allowance applies. Higher sustained activity can lower the allowance until earlier attempts leave the 7-day window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (1)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
✅ Single Commit Policy - COMPLIANTStatus: Policy requirements met • 1 commit • Valid format • Ready for merge 📊 View validation details📝 Commit Details
✅ Validation Results
🤖 Automated validation by NeuroLink Single Commit Enforcement |
🤖 AI Review & Build Compliance ✅Status: AI analysis complete • Build rules validated • Ready for review 📊 View detailed analysis results🛡️ Analysis Complete
📋 Ready for Merge When
🤖 AI analysis complete - check individual code comments for specific feedback |
Documentation Validation Results🚀 Documentation validation passed!
📦 Build artifact uploaded successfully. Ready for deployment preview. Commit: |
Review SummaryDecision: ✅ APPROVED Findings
Impact on Existing Code
Review ScopeThis PR adds empirical verification of CLI feasibility for four tools (Gemini CLI, Amp, Hermes Agent, Grok CLI) against a live proxy. The documentation:
The content accurately reflects the testing methodology and results, improving the documentation with concrete evidence rather than speculation. VerdictAPPROVED - Documentation improvement with accurate empirical data, no security or correctness concerns. |
|
🎉 This PR is included in version 11.17.1 🎉 The release is available on: Your semantic-release bot 📦🚀 |
The audit left Gemini, Amp, Hermes and Grok unresolved — three of them on documentation rather than evidence. Cursor is why that matters: it read as the cleanest win in the entire matrix until someone pointed it at a real proxy and found the env vars were inert.
Same method here: start
neurolink proxy start --port 9911, point the documented override at it, run one command, see what arrives.GOOGLE_GEMINI_BASE_URLAMP_URLGemini — the override is real, the door is the whole job
That is the correct answer from a proxy with no
generateContentroute: the redirect worked and there was nothing to answer it. So §4's new-wire-format list is complete rather than optimistic — no vendor cooperation needed, no override to negotiate.Two traps recorded for whoever builds it: the CLI refuses to run outside a trusted directory (
--skip-trust/GEMINI_CLI_TRUST_WORKSPACE=truefor headless), and it issues agenerateJsoncall at startup, so the door must answer more than the user's turn.Amp — honoured, but it brings its own front door
AMP_URLis genuinely live, unlike Cursor's inert variables. Pointed at the proxy, Amp built its login URL against it:That login URL is the finding. Amp expects an OAuth-style CLI flow at its own endpoint before any API traffic — so onboarding it means implementing an auth surface, not writing a config file. That moves Amp out of the config-writer class and in beside Gemini.
Its bundle vendors Google's GenAI SDK, so
generateContentstrings inside it describe a dependency rather than Amp's own wire. Noted explicitly so the next person to grep does not conclude otherwise.Hermes and Grok — unverifiable, and that is the finding
Neither is installed: no binary on
PATH, no~/.hermesor~/.config/hermes, nothing under any package root. Hermes's "easy" verdict came from documentation, never from a bundle. The blocker is availability, not difficulty, and the first step is installing one — not writing a configurator against a guessed config surface.Docs only; no code, no test changes.