Knowledge Graph | 知識圖

SkillDocs & knowledge

[UDS] Trace impact chains across specs, decisions, and code via a knowledge graph, with a Markdown fallback when no engine is present. Use when: asking what a spec or decision affects, finding which code implements an artifact, tracing dependencies between specs, ADRs, and modules. Not for: plain text search with no spec or decision anchor — use Grep; authoring the spec itself — use /sdd. Keywords: knowledge graph, impact chain, traceability, spec impact, decision graph, 知識圖, 影響鏈, 規格追蹤.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Knowledge Graph | 知識圖 skill

What this skill tells your AI

The instructions your AI receives, as published by asiaostrich/universal-dev-standards in skills/knowledge-graph/SKILL.md and read by ahel’s review.

Language: English | 繁體中文

Answer structural questions across specs, decisions, and code — "what is the full impact chain of XSPEC-205?" — using the Knowledge Graph Memory relationship schema. Works with or without a graph engine.

回答橫跨規格、決策與程式碼的結構性問題——「XSPEC-205 的完整影響鏈是什麼?」——依據知識圖記憶標準的關係 schema。有無圖引擎皆可運作。

Implements: XSPEC-237 Phase 5 — knowledge-graph skill (EngramGraph opt-in)

Mode Selection | 模式選擇

Detect which mode to use before answering:

ConditionMode
ENGRAM_URL set, or a local graph engine responds on /healthService mode (engine)
OtherwiseDegraded mode (Markdown)

Workflow | 工作流程

  1. Resolve the target — normalise the argument to a canonical id (XSPEC-205, DEC-062, a function name).
  2. Choose mode — probe for a graph engine (service) else fall back (degraded).
  3. Service mode (AC-5b) — issue a single multi-hop query and present the returned chain, including cross-domain links (code → spec → decision):
    curl -s -X POST "$ENGRAM_URL/graph/impact-analysis" \
      -H 'content-type: application/json' \
      -d '{"nodeId":"XSPEC-205","maxHops":3}'
    
  4. Degraded mode (AC-5a) — with no engine, read the target document, follow its impacts/impacted_by/supersedes/related front-matter and inline [[ref]] links by reading the linked files, and assemble the chain manually (bounded by reading depth).
  5. Present the chain — list the connected Specs and Decisions, the edge type for each hop, and (if present) each node's confidence, highest first.
  6. State the mode used — always say whether the answer came from the engine or Markdown fallback, so completeness is clear.

Relationship Schema | 關係 schema

See knowledge-graph-memory. Front-matter fields: related, impacts, impacted_by, supersedes, implements. Edge derivation: Decision impacts Spec → IMPACTS; Decision supersedes Decision → SUPERSEDES.

關係欄位與邊推導見知識圖記憶標準

Next Steps Guidance | 下一步引導

  • If degraded mode hit a reading-depth limit, tell the user a graph engine (e.g. EngramGraph) would give a complete chain, and how to set ENGRAM_URL.
  • If a referenced id was not found, surface it as a dangling reference to fix.
  • Offer to add missing impacts/impacted_by front-matter to the documents you traversed.

Reference | 參考

Signals

GitHub stars
73
Forks
13
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
knowledge-graph
Source
github.com/asiaostrich/universal-dev-standards