DX feedback from a fresh setup against a non-Elixir workspace (umbrella / tracking issue)

[NikolaRHristov]

Summary

This issue originally captured a broad set of DX observations from running iksnerd/code-nexus:latest against a large polyglot workspace (Rust + TypeScript + Astro, ~3000+ files, vendored VS Code + Tauri trees) via the HTTP MCP at :3002 from Claude Code.

That feedback has now been split into more focused issues. This issue is kept as an umbrella tracker and narrative overview, and should no longer be used for new feature work.

Child issues

The concrete work items are now tracked in the following issues:

1. Setup & embedding backend – make Ollama / embedding backend requirements explicit and resilient.

   • Issue: Setup & onboarding: make Ollama / embedding backend requirements explicit and resilient #5
   • Focus: EMBEDDING_BACKEND, TF-IDF-only mode, Ollama requirements, clear errors when the embedding backend is unreachable.

2. Ignore configuration – make .gitignore / .nexusignore behavior and skip statistics first-class.

   • Issue: Ignore configuration: make .gitignore / .nexusignore behavior and skip statistics first-class #6
   • Focus: .nexusignore semantics, default deny-list, index skip stats, and a dry-run index_plan tool.

3. Indexing lifecycle – async reindex, progress reporting, incremental updates, and per-project collections.

   • Issue: Indexing lifecycle: async reindex, progress reporting, incremental updates, and per-project collections #7
   • Focus: job IDs, reindex_status, progress, watch / incremental modes, per-project Qdrant collections.

4. Indexing strategy – make indexing inclusive-first, using .gitignore / .nexusignore as the primary filter.

   • Issue: Make indexing inclusive-first (index all code files under the base path, using .gitignore/.nexusignore as the primary filter) #4
   • Focus: detect_indexable_dirs/1 semantics and interaction with ignore rules.

5. Project introspection – make indexed projects, current project, and index health discoverable.

   • Issue: Project introspection: make indexed projects, current project, and index health discoverable #8
   • Focus: project_path, index status tool, and clearer tool preconditions.

6. Errors, health, and observability – /health endpoint, backoff/log dedup, and structured MCP errors.

   • Issue: Errors, health, and observability: /health endpoint, backoff/log dedup, and structured MCP errors #9
   • Focus: container health, retry behavior, and structured error payloads for embedding / Qdrant failures.

7. Minor polish – tool naming consistency and language coverage in reindex results.

   • Issue: Minor polish: tool naming consistency and language coverage in reindex results #10
   • Focus: normalizing tool verbs and surfacing per-language file counts in indexing responses.

How to use this issue going forward

• Treat this issue as the top-level DX narrative linking out to the concrete issues above.
• For bugs or feature work, please comment on or open PRs against the specific child issues rather than this umbrella.
• If new DX themes emerge that don’t fit any of the above, they can either be added here and then spun out into new issues, or added directly as new focused issues and linked from this list.

[iksnerd]

Status update (v1.9.0) — tracking what shipped from this umbrella:

✅ Closed:

• Make indexing inclusive-first (index all code files under the base path, using .gitignore/.nexusignore as the primary filter) #4 — Inclusive-first indexing (index all code, use ignore files as filter)
• Ignore configuration: make .gitignore / .nexusignore behavior and skip statistics first-class #6 — Skip statistics in reindex response + .nexusignore / .gitignore glob support
• Project introspection: make indexed projects, current project, and index health discoverable #8 — get_status tool: current project, Qdrant health, file count, collections, workspace projects
• Minor polish: tool naming consistency and language coverage in reindex results #10 — Language coverage in reindex results (languages: [{lang, file_count}])
• bug: Vector Store "Lines" column renders \u2013 as literal text instead of en dash #11 — Vector Store Lines column rendering bug

✅ Also shipped (not originally tracked as sub-issues):

• EMBEDDING_BACKEND=tfidf — skip Ollama entirely for ~10× faster indexing (Setup & onboarding: make Ollama / embedding backend requirements explicit and resilient #5, Indexing lifecycle: async reindex, progress reporting, incremental updates, and per-project collections #7 workaround)
• /health endpoint (Errors, health, and observability: /health endpoint, backoff/log dedup, and structured MCP errors #9 partial)
• Single-project workspace auto-default
• .nexusignore support

❌ Still open:

• Indexing lifecycle: async reindex, progress reporting, incremental updates, and per-project collections #7 — Async reindex + progress reporting (highest priority)
• Errors, health, and observability: /health endpoint, backoff/log dedup, and structured MCP errors #9 — Structured MCP error payloads + correlation IDs
• .nexusignore relative path normalization (patterns like Documentation/ without **/)

[iksnerd]

All child issues are now resolved:

Issue	Status	Shipped
#4 Inclusive-first indexing	✅ Closed	Earlier
#5 Embedding backend onboarding	✅ Closed	Earlier
#6 .gitignore / .nexusignore	✅ Closed	Earlier
#7 Async reindex + progress	✅ Closed	v1.14.0
#8 Project introspection (get_status)	✅ Closed	Earlier
#9 Health endpoint + structured errors	✅ Closed	v1.13.0
#10 Minor polish	✅ Closed	Earlier
#12 Search debounce	✅ Closed	v1.13.0
#13 EMBEDDING_MODEL alias	✅ Closed	v1.13.0

Closing the umbrella.