Thread Types

Every Claude Code session is classified into one of seven thread types. Classification follows a priority-based evaluation — the first matching rule wins, from most specific to least.

The Seven Types

Priority	Type	Name	Description
1	Z-thread	Zero-touch	Minimal human input, maximum autonomous work
2	B-thread	Big	Nested sub-agents (agents spawning agents)
3	L-thread	Long	Extended autonomous stretch with high tool usage
4	F-thread	Fusion	Similar tasks distributed to multiple agents
5	P-thread	Parallel	2+ concurrent agent execution paths
6	C-thread	Chained	Multiple human checkpoints with work between each
7	Base	Default	Standard conversation

Classification Rules

Z-thread (Zero-touch)

human_messages <= 1 AND total_tool_calls >= 10

The holy grail of agentic coding. You give one command, Claude handles everything autonomously. Requires significant work output (10+ tool calls) to avoid classifying trivial one-shot queries.

B-thread (Big)

max_sub_agent_depth >= 2

Agents spawning agents. The primary agent delegates tasks to sub-agents, and those sub-agents spawn their own sub-agents. This indicates deep, hierarchical task decomposition.

L-thread (Long)

longest_autonomous_stretch >= 30 minutes AND total_tool_calls >= 50

Claude works for 30+ minutes without human intervention, executing 50+ tool calls. These are deep, focused work sessions where Claude has enough context and permissions to make sustained progress.

F-thread (Fusion)

2+ sub-agents with Jaccard word similarity >= 0.7 in their prompts

Multiple agents receive substantially similar prompts — a pattern of distributing related tasks across parallel workers. OMAS detects this using Jaccard similarity on sub-agent prompt text.

P-thread (Parallel)

concurrent_sessions >= 2

Two or more Claude Code sessions running simultaneously (cross-session). Detected using the sweep-line algorithm across all session time ranges to find peak concurrent count.

C-thread (Chained)

human_messages >= 3 AND each gap between human messages has >= 3 tool calls

A conversational pattern with multiple human checkpoints, but with meaningful work between each. This represents iterative development with human review at each stage.

Base

(default — no other rule matched)

Standard Claude Code conversation without significant agentic patterns.

Evolution Roadmap

Progressing through thread types represents increasing mastery of agentic coding:

Base → C-thread    Use 3+ conversation turns with work between each
C    → P-thread    Request parallel tasks ("analyze these 3 files simultaneously")
P    → L-thread    Give detailed instructions, let Claude work 30+ minutes
L    → B-thread    Use teams/worktrees for deep sub-agent hierarchies
B    → Z-thread    One command, full feature implementation, auto-approve

Tips for Advancing

Transition	Strategy
Base → C	Have multi-turn conversations with real work in between
C → P	Ask Claude to work on multiple things simultaneously
P → L	Write a detailed CLAUDE.md, give complete requirements upfront
L → B	Use the Agent tool, teams, and worktrees for hierarchical work
B → Z	Enable auto-approve, provide comprehensive instructions in one prompt

Threshold Configuration

All classification thresholds are defined in src/omas/config.py and can be viewed in the Configuration Reference.