[BUG-05] --lsp-status has two entry points with different payload shapes

[Wolfvin]

Summary

codelens --lsp-status (top-level flag) and codelens lsp-status (subcommand) return different payload shapes, despite the documentation treating them as the same operation.

Evidence

Path A — top-level flag (scripts/codelens.py:829-837):

# Handle --lsp-status as a special top-level flag (not a subcommand)
if "--lsp-status" in sys.argv:
    try:
        from hybrid_engine import get_lsp_status
        status = get_lsp_status()
        print(format_output(status, _default_format, "lsp-status"))
        sys.exit(0)
    except Exception as e:
        ...

Returns: hybrid_engine.get_lsp_status() payload.

Path B — subcommand (scripts/commands/lsp_status.py, auto-registered as "lsp-status"):
Calls lsp_client.detect_available_servers() and returns its payload.

The two functions (hybrid_engine.get_lsp_status vs lsp_client.detect_available_servers) are defined in different modules and emit different JSON shapes.

Impact

1. User confusion: codelens --lsp-status and codelens lsp-status look interchangeable in docs but produce different output. AI agents that parse the response schema have to handle two shapes.
2. MCP exposure: The MCP server exposes only the subcommand path (codelens_lsp_status -> lsp_client). CLI users typing --lsp-status get a different result than MCP clients calling the tool with the same name.
3. Documentation drift: SKILL-QUICK.md mentions only lsp-status; the top-level --lsp-status flag is undocumented in the trigger map but intercepted in code.

Repro

diff <(python3 scripts/codelens.py --lsp-status --format json | python3 -m json.tool) \
     <(python3 scripts/codelens.py lsp-status --format json | python3 -m json.tool)
# Output differs (different top-level keys, different per-server fields).

Suggested fix

Pick one of:

• (A) Consolidate on the subcommand. Remove the top-level --lsp-status interception in codelens.py:829-837. Update commands/lsp_status.py to call hybrid_engine.get_lsp_status() (the richer payload) instead of lsp_client.detect_available_servers(). This makes CLI and MCP behavior identical.
• (B) Keep both but unify the payload. Make lsp_client.detect_available_servers() and hybrid_engine.get_lsp_status() return the same schema, and document both entry points explicitly in SKILL-QUICK.md trigger map.

Either way, add a test that asserts both invocations produce structurally-equal JSON (same top-level keys).

Files

• scripts/codelens.py (L829-837)
• scripts/commands/lsp_status.py
• scripts/hybrid_engine.py (get_lsp_status)
• scripts/lsp_client.py (detect_available_servers)
• SKILL-QUICK.md (trigger map)

[Wolfvin]

Multi-worker analysis: this bug is a symptom of a larger gap

The dual-entry-point bug (#33) is real and must be fixed, but 4 worker reports argue the underlying problem is larger: CodeLens has no native LSP server, only an lsp_client.py that wraps external LSPs for --deep analysis. The --lsp-status flag was a stopgap that accidentally grew into a second entry point.

Worker	Source file	What they add
Opengrep	update!/CodeLens_Opengrep_Upgrade_Analysis.md #39	LSP server (codelens lsp) implementing LSP 3.17 over stdio: textDocument/publishDiagnostics, textDocument/codeAction (suggest autofix), textDocument/hover (rule doc + taint path), file operation filters. Use pylsp or pygls framework.
Semgrep	update!/CodeLens_Upgrade_Issues_from_Semgrep.md CL-008	LSP server native (codelens lsp) — initialize, didOpen, didChange, hover, definition, references, codeLens, codeAction + custom commands codelens/query, codelens/impact, codelens/guard. Enables Neovim, Emacs (lsp-mode), Helix, Zed — not just VS Code. Target <50ms for hover/definition.
Serena	update!/CodeLens_vs_Serena_Upgrade_Analysis.md S1	Vendor Serena's solidlsp module for LSP integration with 50+ language servers (Pyright, TypeScript LS, gopls, rust-analyzer, clangd, jdtls, OmniSharp) covering 40+ languages at IDE-level accuracy. Auto-detect language, lazy-start servers, cache at .codelens/lsp_cache/.
Serena	same file S13	LSP diagnostics integration — codelens diagnostics [workspace] [--file FILE] [--severity ...] using textDocument/publishDiagnostics for real-time type/syntax/undefined-reference errors. Integrate with check and smell.

Recommendation

This bug (#33) should be fixed quickly (consolidate on the subcommand path, ~1 hour). But the larger LSP server native feature deserves its own issue with phased scope:

1. Phase 1: Native LSP server via pygls (Opengrep [BUG-09] Command auto-import silently drops broken modules — registry missing commands with no CI signal #39 + Semgrep CL-008) — initialize + didOpen + didChange + hover + definition. ~3-4 weeks.
2. Phase 2: Vendor solidlsp for multi-language LSP client (Serena S1) — enables --deep without requiring user-installed language servers. ~3-4 weeks, depends on Phase 1.
3. Phase 3: Diagnostics integration (Serena S13) — push findings to editor Problems panel via publishDiagnostics. ~1-2 weeks, depends on Phase 1.

Filing the LSP server native issue separately is recommended; this comment is to record the worker consensus.

Aggregated from update!/ worker analysis reports by multi-worker triage.