liveBook · Manning

Argus

As explained in Designing AI Agents

Argus is the code review agent built as the book's progressively developed reference system.

Argus is a code review agent and an evolving reference design: it starts with a single-pass review of one diff, then adds capabilities while preserving its PRA loop. Each extension is weighed against its outcomes, costs, and remaining failures. section 2.6

Table 2.8 Argus evolution across the book
Chapter
Module Added
What Argus Gains
Ch 2
Core PRA loop
Single-pass review of one diff
Ch 3
Perception
Context Triage: reads only relevant files, not everything
Ch 4
Memory
Retains selected review evidence and retrieves relevant prior findings
Ch 5
Reasoning
Complexity-Based Routing: deep analysis for complex code, quick pass for simple changes
Ch 6
Action
Tool use: runs tests, invokes linters, checks type errors
Ch 7
Reflection
Generator-Critic loop: self-checks for false positives before reporting
Ch 8
Collaboration
Fan-Out/Gather: parallel review of large PRs across multiple sub-agents
Ch 9
Governance
Approval gates, audit trails, read-only sandbox
Ch 10
Composition
Composes the modules and evaluates the end-to-end review system

Trace how Argus grows from a core review loop through perception, memory, reasoning, and further architectural modules. section 2.6

Table 5.9 Argus capabilities through chapter 5
Capability
Module
Chapter
Basic PRA loop
core.py
Ch2
Smart context loading
perception.py
Ch3
Perception traces
perception.py
Ch3
Cross-session memory
memory.py
Ch4
Progress tracking
memory.py
Ch4
Failure journals
memory.py
Ch4
Explicit claim-and-dependency chain
 
reasoning.py
Ch5
Complexity-based routing
reasoning.py
Ch5
Bounded bug hypotheses and test plans
reasoning.py
Ch5
Reasoning telemetry schema
reasoning_trace.py
Ch5

See the implemented capabilities through reasoning and the boundaries Argus still cannot cross. section 5.8

How Designing AI Agents covers it

How deep it goes
Tracks modules and implemented reasoning capabilities, but stops short of unrestricted changes to production state.
Examples
Uses single-diff review, complexity-based routing, claim-and-dependency chains, and bounded sandbox checks.
What you can do after
Build an agent architecture incrementally by adding capabilities where evidence justifies their costs.
How it connects
Connects the PRA loop with perception, memory, reasoning, action, reflection, collaboration, and governance.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Designing AI Agents. The section links point to the parts of the book it draws on.