nav-graph_skill

This skill queries and manages the unified project knowledge graph to surface relevant knowledge and memories across tasks, SOPs, and concepts.
  • Python

142

GitHub Stars

1

Bundled Files

2 months ago

Catalog Refreshed

4 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill alekspetrov/navigator --skill nav-graph

  • SKILL.md10.4 KB

Overview

This skill queries and manages a unified project knowledge graph that surfaces tasks, SOPs, memories, and concepts. It helps answer questions like “What do we know about X?” and captures project-specific patterns, pitfalls, decisions, and learnings. It keeps relevant context compact so sessions stay focused and efficient.

How this skill works

The skill reads a compact graph file (.agent/knowledge/graph.json) and supports actions: query, add-memory, initialize, stats, and related traversal. Queries extract a normalized concept from user input, return matched tasks, SOPs, files, and memories, and allow drilling into details. Capture actions parse the user sentence to infer memory type (pattern, pitfall, decision, learning), attach concepts and confidence, and add a memory node to the graph.

When to use it

  • Ask “What do we know about X?” to surface project knowledge across tasks, SOPs, files, and memories.
  • Ask “Show everything related to X” or “What’s related to TASK-29?” to traverse relationships and find linked items.
  • Use “Any pitfalls for X?” or “Any decisions about X?” to get focused memory types.
  • Say “Remember this pattern/pitfall/decision/learning: …” to capture a new memory tied to concepts.
  • Run “Initialize knowledge graph” or “Rebuild knowledge graph” when setting up or refreshing project state.

Best practices

  • Scope queries to a specific concept or node (e.g., ‘auth’, ‘testing’, or ‘TASK-29’) to keep results actionable.
  • When capturing memories include clear context and rationale (who, when, why) so confidence and relationships are meaningful.
  • Avoid storing large code blobs in memories; reference file paths instead and keep summaries concise.
  • Run graph health, stale detection, and pruning periodically to keep the graph under token budget.
  • Enable auto-capture for corrections and decisions cautiously to prevent noise.

Example use cases

  • Developer asks “What do we know about authentication?” and gets tasks, SOPs, and recent pitfalls to avoid.
  • After a postmortem, capture a pitfall: “Remember this pitfall: auth changes often break session tests.”
  • On session start, surface a few high-confidence memories relevant to active concepts.
  • During a planning meeting, traverse relationships for TASK-29 to find related tasks, SOPs, and decisions.
  • Rebuild the graph after major documentation changes to refresh concepts and relationships.

FAQ

The graph file is kept compact (~1–2k tokens for moderate projects). Full graphs can grow; queries aim to load only the relevant subset.

How does confidence work for memories?

Memories have base confidence (correction-based ~0.8, explicit capture ~0.9), decay weekly, and gain boosts with use. Thresholds mark candidates for pruning.

Built by
VeilStrat
AI signals for GTM teams
© 2026 VeilStrat. All rights reserved.All systems operational