Request
Stage-2 LLM analysis currently requires a metered HTTP API key. Many users already have an authenticated claude or codex CLI; SkillSpector could use that session directly — no separate API key or billing.
Proposed fix
Add claude_cli / codex_cli providers routing LLM calls through the local CLI via one hardened subprocess chokepoint (no shell, stdin-only untrusted content, capability stripping, env scrub, bounded streamed output, fail-closed) plus a ChatOpenAI-compatible adapter so the analyzers work unchanged.
Opened to follow CONTRIBUTING.md (issue-first); fix is in #52.
🤖 Generated with Claude Code
Request
Stage-2 LLM analysis currently requires a metered HTTP API key. Many users already have an authenticated
claudeorcodexCLI; SkillSpector could use that session directly — no separate API key or billing.Proposed fix
Add
claude_cli/codex_cliproviders routing LLM calls through the local CLI via one hardened subprocess chokepoint (no shell, stdin-only untrusted content, capability stripping, env scrub, bounded streamed output, fail-closed) plus aChatOpenAI-compatible adapter so the analyzers work unchanged.Opened to follow CONTRIBUTING.md (issue-first); fix is in #52.
🤖 Generated with Claude Code