next_task

Returns frontier items (candidate next actions) with graph context attached.

Read-only: No (triggers frontier computation)

Description

The deterministic layer has already filtered out:

• Out-of-scope targets
• Duplicate/already-tested actions
• Actions exceeding OPSEC hard noise limits
• Dead hosts

Everything else passes through for the LLM's analysis. Each item includes graph metrics (hops to objective, fan-out estimate, node degree, confidence, OPSEC noise rating).

The LLM's job is to:

1. Score and rank these by overall value
2. Spot multi-step attack chains across items
3. Consider sequencing (what should happen first)
4. Assess likely defenses and risks
5. Recommend specific actions for the top items

Parameters

Parameter	Type	Default	Description
max_items	integer	20	Maximum frontier items to return (1–50)
include_filtered	boolean	false	Also return items that were filtered out, with reasons

Returns

Field	Type	Description
candidate_count	number	Total candidates after filtering
candidates	FrontierItem[]	Frontier items with graph metrics
filtered_count	number	Number filtered out (if include_filtered)
filtered	array	Filtered items with reasons (if include_filtered)

FrontierItem

Field	Type	Description
id	string	Unique frontier item ID
type	string	incomplete_node, untested_edge, or inferred_edge
description	string	Human-readable description
graph_metrics.hops_to_objective	number\|null	Shortest path to any objective
graph_metrics.fan_out_estimate	number	How many new nodes this could expose
graph_metrics.node_degree	number	Connected edges
graph_metrics.confidence	number	Current confidence
opsec_noise	number	Estimated noise (0.0–1.0)
staleness_seconds	number	Time since last update
community_id	number	Louvain community this node belongs to
community_unexplored_count	number	Unexplored frontier items in the same community

Usage Notes

• Call this to decide what to do next — the core of the engagement loop
• Set include_filtered: true to understand what was rejected and why
• Pair with validate_action before executing any chosen task