# Okf Query

> Multi-hop query and subgraph extraction over OKF bundles for progressive disclosure. Use when packing minimal context for long-running agents, exploring the neighborhood of a concept, or answering what relates to X within N hops.

- Skill: `spillwavesolutions/okf-query` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add spillwavesolutions/okf-query`
- Raw SKILL.md: https://api.skillmd.com/api/skills/spillwavesolutions/okf-query/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: spillwavesolutions (https://skillmd.com/u/spillwavesolutions)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/spillwavesolutions/okf-query

---


# OKF Query & Progressive Disclosure

## Goal

Answer graph questions and emit **minimal context packs** (subgraphs) instead of dumping entire directories.

This is the **read-path optimizer**. Defaults exist so a long-running agent does not eat the whole second brain.

## Query types

| Intent | Approach |
|--------|----------|
| Neighbors of X | 1-hop inbound + outbound |
| Context pack for agent run | 2-hop **outbound** subgraph around the entry concept |
| Path A → B | BFS / path search on typed edges |
| By type | Filter concepts where frontmatter `type` matches |
| By tag | Filter on `tags` |
| Trust slice | Prefer `verified: true` and non-stale nodes |

## Process

1. Resolve bundle root and query target(s).
2. Prefer deterministic tools:
   - `okf graph`, search, or list commands if available
   - Fallback (defaults: **2 hops**, **20 nodes**, **outbound-only**):
     ```bash
     python3 "${CLAUDE_PLUGIN_ROOT}/scripts/okf-graph.py" pack <bundle> <concept> --hops 2 --max-nodes 20
     python3 "${CLAUDE_PLUGIN_ROOT}/scripts/okf-graph.py" subgraph <bundle> <concept> --hops 2
     python3 "${CLAUDE_PLUGIN_ROOT}/scripts/okf-graph.py" backlinks <bundle> <concept>
     python3 "${CLAUDE_PLUGIN_ROOT}/scripts/okf-graph.py" edges <bundle> --rel depends_on
     ```
3. Shape results for the consumer:
   - **Human / agent pack**: use `pack` → `markdown` field (preferred)
   - **JSON**: nodes + typed edges for downstream steps
4. Cap pack size. Default hops = 2, max-nodes = 20. Increase only if the user needs deeper context.
5. Annotate trust: mark unverified or draft nodes so consumers can deprioritize them. Do not silently omit schema-declared high-impact unverified nodes — flag them.

Deep reference: `references/progressive-disclosure.md`.

## Progressive disclosure pack format

```markdown
# Context pack: <root concept>
Hops: 2 | Nodes: N | Generated: <timestamp>

## Entry
- path, type, one-line description

## Included concepts (read order)
1. ...
2. ...

## Graph (Mermaid)
```mermaid
graph LR
  A --> B
```

## Excluded (available on request)
- siblings / deeper hops summary
```

## Rules

- Outbound-only by default. `--undirected` floods through Catalog hubs; prefer `impact` for “what would break?”.
- Prefer verified, active nodes when trimming packs under a size budget.
- Do not silently omit high-impact unverified nodes — flag them (`x-impact` from the owning plugin).
- Absolute paths in links inside packs so agents can open files reliably.
- Never fabricate nodes missing from the bundle.

## Done when

- Query answered with explicit hop depth and node count
- Pack is small enough for agent context (or user approved a larger extract)