Skip to content

RFC (draft): OAST multi-instance model — manage several interaction servers (spawn local + connect remote) from the workbench #488

Description

@thomas-stegemann

Status: DRAFT / RFC — needs refinement before implementation. Captures a direction from the #486 discussion; the design sketch and open questions below are starting points, not decisions.

Motivation

Today Bowire binds to exactly one OAST interaction server: a single Bowire:Oast:Server config value → one InteractshClient session per workbench process (see OastWorkbenchSession, #486). Real use wants several at once:

Two usage variants surfaced in discussion, and we likely want both:

  1. Connect to a remote / standalone server — already possible via --oast-server (single). Generalise to several selectable servers.
  2. Spawn + stop LOCAL catcher instances from the workbench UI — start/stop a bowire oast serve catcher without leaving the UI. There is a direct precedent: the reverse-proxy / mock servers are started/stopped/listed from the workbench via ReverseProxyRegistry + the tools endpoints (/api/tools/reverse-proxy/*). An OAST instance registry would mirror that.

Design sketch (to refine)

  • Config: single Bowire:Oast:Server → a named-instance list (e.g. Bowire:Oast:Instances:[] with { name, url, token }), plus a default.
  • Registry: an OastInstanceRegistry singleton holding N sessions (remote) and N spawned local catchers, modelled on ReverseProxyRegistry (lifetime tied to the host; spawned catchers die with the process).
  • Endpoints: /api/security/oast/instances (list), .../instances POST (add remote / spawn local), .../instances/{id} DELETE (remove / stop), and the existing allocate/poll become per-instance (.../instances/{id}/allocate|poll).
  • UI: an instance picker in the OAST panel; "spawn local catcher" + "add remote server" actions; per-instance payload list + live feed.
  • Scanner: --oast-server stays the single-instance shortcut; the workbench selects an active instance for manual work.

Constraints / gotchas

Open questions (the "needs refinement" part)

  • Per-workspace instances vs one global set?
  • Persist configured instances across restarts, or session-only like today?
  • Default-instance selection + what the scanner uses when several exist.
  • Does this same instance surface double as a rail-agnostic OOB inspector (requestbin / webhook.site style)? OAST's primitive is cross-cutting (it lives in the neutral Kuestenlogik.Bowire.Oast package, like Bowire.Telemetry / Bowire.Map), so a non-security use — egress/firewall testing, webhook debugging, DNS-resolver testing — could share the registry + feed. If so, the endpoints may belong at /api/oast/* (not /api/security/oast/*) with the security rail merely embedding the panel via the __bowireOast global seam.

Related

Metadata

Metadata

Labels

No labels
No labels

Projects

  • Status
    Backlog

Relationships

None yet

Development

No branches or pull requests

Issue actions