| name | okf-impact |
| description | Compute the transitive impact (blast radius / ripple) of changing a concept in an OKF bundle. Use when the user asks what depends on a concept, what breaks if something changes, needs a ranked update list, or wants blast-radius / ripple analysis before editing agents, workflows, or knowledge nodes. |
OKF Impact Analysis
Goal
Given a concept ID/path (or a proposed change), return the full affected subgraph plus a practical update order.
Process
- Resolve the target concept (relative path, absolute path, title, or ID). Locate the bundle root (
.okf/, knowledge/, sample-okf/).
- Use deterministic tools first:
- Build both:
- Outbound closure — what the concept depends on / routes to
- Inbound closure — who cites / depends on / routes through it
- Direct typed edges —
direct_edges in impact JSON (routes_to, depends_on, …)
- Classify nodes by type and trust/lifecycle (
verified, status, stale_after).
- Produce two outputs:
- Human report — ranked list (critical → low), with reasons and suggested actions
- Structured JSON (optional):
{ target, inbound, outbound, direct_edges, suggested_order }
Ranking heuristics
| Signal | Effect on rank |
|---|
Type AgentNode, Workflow, Harness, SharedState | Higher impact |
verified: false on high-type nodes | Escalate to critical |
status: deprecated | Flag for cleanup, lower urgency for feature work |
| Shallow hop depth with many dependents | Higher priority |
| Leaf knowledge nodes | Usually lower |
Report template
# Impact: <title> (`<path>`)
## Summary
- Inbound dependents: N
- Outbound dependencies: M
- Critical / unverified highlights: ...
## Ranked dependents (update order)
1. **[Name](path)** — type, why it matters, suggested action
2. ...
## Outbound dependencies to review
- ...
## Recommended next steps
- [ ] ...
Rules
- Never invent links. Only report edges found by tools or careful Markdown crawl.
- If graph tools are unavailable, state the limitation and use link crawl.
- Prefer high-trust paths when ranking recommendations.
- After structural edits, re-run impact on the same target.
Related
- Progressive disclosure packs: use
okf-query / okf-graph.py subgraph
- Validation before big refactors:
okf-validate
