Przejdź do treści
Biblioteka skilli
AI i agenci · Integracja z narzędziem

nexus

/nexus

Pomaga wybrać właściwe polecenie nx do wyszukiwania, pamięci, przechowywania wiedzy i indeksowania.

Repozytorium
nexus ↗
Licencja
AGPL-3.0
Ostatnia zmiana repozytorium
26.10.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
2
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 26.10.2026, bez odwołań do wycofanych modeli i API.

Licencja

AGPL-3.0, copyleft: przeróbki udostępniane innym muszą mieć tę samą licencję.

Co robi i kiedy

Pomaga wybrać właściwe polecenie nx do wyszukiwania, pamięci, przechowywania wiedzy i indeksowania.

Oryginalny opis (EN)

Use when running nx commands for search, memory, knowledge storage, or indexing — or when unsure which nx subcommand to use

Jak działa

  1. Storage Tiers
  2. Common Operations
  3. When to reach for each
  4. Catalog (T3 metadata — document registry + typed link graph)
  5. Indexing (CLI only — no MCP equivalent)
  6. Collection Naming
  7. Title Conventions
Sekcje z nagłówków SKILL.md, w oryginale · 111 linii, 758 słów
Podgląd SKILL.md
---
name: nexus
description: Use when running nx commands for search, memory, knowledge storage, or indexing — or when unsure which nx subcommand to use
effort: low
---

# Nexus Quick Reference

## Storage Tiers

| Tier | Scope | MCP Tools | Use For |
|------|-------|-----------|---------|
| T1 | Session | `scratch`, `scratch_manage` | Hypotheses, interim findings, checkpoints |
| T2 | Persistent | `memory_put`, `memory_get`, `memory_search` | Cross-session state, agent relay notes |
| T3 | Permanent | `search`, `store_put`, `store_list` | Validated findings, decisions, patterns |

## Common Operations

```
# Analytical answer (RDR-080) — plan-match-first, falls through to inline planner on miss
mcp__plugin_conexus_nexus__nx_answer(question="how does plan matching work"                  # primary front door
mcp__plugin_conexus_nexus__nx_answer(question="...", dimensions={"verb":"research"}          # pin a verb for the matcher
mcp__plugin_conexus_nexus__nx_answer(question="...", scope="1.2"                             # catalog subtree: filters query and search_metadata_scoped steps; a search step runs unscoped and the answer says so

# Search (chunk-level) / query (document-level)
mcp__plugin_conexus_nexus__search(query="query"                          # semantic search across T3
mcp__plugin_conexus_nexus__search(query="query", corpus="code"           # code only
mcp__plugin_conexus_nexus__search(query="query", structured=True         # returns {ids, tumblers, distances, collections}
mcp__plugin_conexus_nexus__query(question="...", corpus="knowledge", follow_links="cites"    # catalog-aware, document-level

# Batch hydrate chunks past the 300-record read cap (MAX_QUERY_RESULTS)
mcp__plugin_conexus_nexus__store_get_many(ids=["id1","id2","id3"], collections="knowledge__art-1-1__voyage-context-3__v1"
mcp__plugin_conexus_nexus__store_get_many(ids="id1,id2", collections="rdr__nexus-1-1__voyage-context-3__v1", structured=True

# Walk the catalog link graph (depth capped at 3 — SC-4)
mcp__plugin_conexus_nexus__traverse(seeds=["1.1.635"], link_types=["implements","cites"], depth=2
mcp__plugin_conexus_nexus__traverse(seeds="1.1.635", purpose="find-implementations"          # link_types XOR purpose

# Analytical operators (summarize, extract, rank, compare, generate, filter, check, verify, groupby,
# aggregate) are NOT MCP tools: they run as plan steps inside nx_answer, each a claude -p subprocess
mcp__plugin_conexus_nexus__nx_answer(question="rank these approaches by relevance to X"   # plan-matched; operator steps run server-side

# Background hygiene — call and let run (claude -p subprocesses); each requires real args
mcp__plugin_conexus_nexus__nx_tidy(topic="chromadb quotas", collection="<subject>")   # read-only T3 consolidation report for one collection
mcp__plugin_conexus_nexus__nx_enrich_beads(bead_description="<title + description>")  # enriches ONE bead's description
mcp__plugin_conexus_nexus__nx_plan_audit(plan_json='{"steps": [...]}')                # audits ONE plan against the codebase

# Memory (T2)
mcp__plugin_conexus_nexus__memory_put(content="content", project="{repo}", title="file.md", ttl=None)   # ttl=None (or omit) for a record of record: permanent, since 2026-09-12 (nexus-473mx); pass ttl=N (days) for a session finding that should expire
mcp__plugin_conexus_nexus__memory_get(project="{repo}", title="file.md"       # exact-then-unique-prefix match
mcp__plugin_conexus_nexus__memory_search(query="query", project="{repo}"

# Knowledge (T3)
mcp__plugin_conexus_nexus__store_put(content="content", collection="<subject>", title="title", tags="tag", agent="<your-role>"
mcp__plugin_conexus_nexus__store_list(collection="<subject>"

# Scratch (T1)
mcp__plugin_conexus_nexus__scratch(action="put", content="working note"
mcp__plugin_conexus_nexus__scratch_manage(action="flag", entry_id="<id>"       # auto-promote to T2 at session end

# Plan library (T2)
mcp__plugin_conexus_nexus__plan_search(query="retrieval"                       # find reusable plans
mcp__plugin_conexus_nexus__plan_save(query="...", plan_json="{...}", verb="research"  # persist a reusable RETRIEVAL plan (verb REQUIRED; impl-plans -> beads/T2)
```

## When to reach for each

- **`nx_answer`** — only when the answer must be reduced from many documents (cross-corpus synthesis, ranking/comparing, RDR research); p50 80s, p95 217s, can time out at 300s. Not for file:line, single-fact, or already-in-T2 questions — use `search`/`query` (seconds; mean ~8s, tail to ~45s). Plan-match-first; falls through to inline planner on miss.
- **`search` vs `query`** — `search` returns chunks (finest grain), `query` returns documents grouped by source. Use `search` for fragment hunting, `query` for literature scoping + catalog traversal.
- **`traverse`** — walk the typed link graph from known tumblers. `link_types` XOR `purpose`; depth ≤ 3.
- **`store_get_many`** — batch-hydrate chunk IDs from `search(structured=True)` or `traverse`. Safe past the 300-record write cap.
- **Operators** — content transforms; take raw text, return structured JSON. Use after retrieval to summarize/extract/rank/compare/generate.
- **`nx_tidy`** — read-only report on ONE T3 collection (`topic` + `collection` required); does not write. For T2 consolidation use `memory_consolidate` instead. **`nx_enrich_beads`** / **`nx_plan_audit`** — enrich/audit the ONE bead or plan passed in as an argument, not a sweep of beads or the plan library. All three: slow (claude -p), call and move on.

## Catalog (T3 metadata — document registry + typed link graph)

```
# Search/browse
mcp__plugin_conexus_nexus-catalog__search(query="schema mappings", author="Fagin", corpus="schema-evolution"
mcp__plugin_conexus_nexus-catalog__show(tumbler="1.9.14"                    # full entry with links
mcp__plugin_conexus_nexus-catalog__resolve(owner="1.1", corpus="schema-evolution"  # → collection names
mcp__plugin_conexus_nexus-catalog__stats                                    # health summary

# Link graph — live documents only (deleted nodes excluded)
mcp__plugin_conexus_nexus-catalog__links(tumbler="1.9.14", direction="in", link_type="cites", depth=2
  Returns {"nodes": [...], "edges": [...]}

# Link CRUD
mcp__plugin_conexus_nexus-catalog__link(from_tumbler="1.1.1", to_tumbler="1.2.5", link_type="cites", created_by="user"
  Returns {"from": ..., "to": ..., "type": ..., "created": true/false}

# Admin/audit — includes orphaned links (all links, not just live)
mcp__plugin_conexus_nexus-catalog__link_query(link_type="cites", created_by="bib_enricher", limit=50
```

**Link types**: `cites` (citation), `implements-heuristic` (auto code→RDR), `supersedes`, `quotes`, `relates`, `comments`, `implements` (manual).
**Two graph views**: `mcp__plugin_conexus_nexus-catalog__links` returns live-document links only. `mcp__plugin_conexus_nexus-catalog__link_query` returns all links including orphans.
Use catalog for: author queries, citation traversal, provenance chains, corpus-scoped search.
The `nx_answer` MCP tool handles full catalog-aware plan execution.

## Indexing (CLI only — no MCP equivalent)

```bash
nx index repo <path>                 # index repo (classifies into code + docs collections)
```

## Collection Naming

RDR-103 conformant shape: `<content_type>__<owner>__<embedding_model>__v<n>`. Examples: `code__nexus-1-1__voyage-code-3__v1`, `docs__nexus-1-1__voyage-context-3__v1`, `knowledge__art-1-1__voyage-context-3__v1`. The short legacy form (`knowledge__topic`) is accepted at the `--collection` boundary and auto-promoted by `t3_collection_name`. Pre-existing legacy 2-segment collections remain readable. Always `__` as the segment separator (colons are invalid in ChromaDB names).

## Title Conventions

Use hyphens: `research-{topic}`, `decision-{component}-{name}`, `pattern-{name}`

For full tool reference with all parameters, see [reference.md](./reference.md).
Pliki w paczce (2)
  • SKILL.md 8 KB
  • reference.md 26 KB

Wersje

RepozytoriumWynikLicencja
Ta strona · hellblazer/nexus100AGPL-3.0
muhzain05/Nexus100MIT
aayushostwal/nexus100MIT

Migawka bazy z 13.11.2026 · wersja paczki 3a76f5040530 (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 100memory
    Przeszukuje i obsługuje graf wiedzy membrain przez polecenie /memory: wyszukiwanie, ślad, dodawanie wpisów, epizody i status.
  • 100knowledge-ops
    Zarządza bazą wiedzy, jej wczytywaniem, synchronizacją i wyszukiwaniem w wielu warstwach przechowywania: pliki lokalne, pamięć MCP, bazy wektorowe i repozytoria Git.
  • 100knowledge-ops
    Zarządza bazą wiedzy, jej wczytywaniem, synchronizacją i wyszukiwaniem w wielu warstwach przechowywania: pliki lokalne, pamięć MCP, bazy wektorowe i repozytoria Git.
  • 100goal-manager
    Tworzy, odczytuje, koryguje, audytuje, kończy lub blokuje trwały cel w bieżącej rundzie Nexus i obsługuje wyłącznie jawny cykl życia celów.
  • 100system-nexus-operation
    Pokazuje w interfejsie operacje systemowe Nexus w Temporalu, np. SignalWithStart czy QueryWorkflow, i dodaje nowe operacje do widoku.
  • 100orchestra-router
    Kieruje pracę w systemie orkiestry (doktryna v3 lean) do właściwej domeny specjalistów przy koordynowaniu wielu agentów, w tym do metadyrygenta NEXUS przy ideacji i planowaniu biznesu.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.