Part 6

Inside a Production Coding Agent

A source-led tour of a production coding agent: runtime, tools, memory, security, and extension points. Use the architecture as a set of questions for evaluating any coding agent you may rely on.

25 notesOpen reading · no login

Part 625notes

THE QUESTION THIS PAGE ANSWERS

ANSWER FIRST

What will the “Inside a Production Coding Agent” AI learning path help you do?

A source-led tour of a production coding agent: runtime, tools, memory, security, and extension points. Use the architecture as a set of questions for evaluating any coding agent you may rely on. The path contains 25 free notes, each centered on one question you can understand and test.

DECISION RULE

Core themes include System Map, Agent Core Loop, Tool System, Context & Memory.

TRY NEXT

Begin with “How 79 Workspace Members Form a Product,” then choose the next note by the task in front of you.

WATCH FOR

Do not optimize for finishing the list. Explaining one trade-off with your own example matters more than opening more titles.

What this route helps you practice

Open the first note

Each chapter follows a class of real decisions. Follow the sequence, or enter at the problem you are solving today.

25notes

01How 79 Workspace Members Form a ProductReconstructing the real Cargo Workspace layers by entry point, Agent runtime, tools, and infrastructureArchitecture3 min↗02Rust Technology Choice: Facts vs InferencesStarting from source-verifiable facts: analyzing engineering trade-offs from the type system, concurrency safety, and distributionConcept3 min↗03From the Real main() to the First SamplingTracing the full call chain: entry point, session creation, prompt rendering, model sampling, and streaming responseDeep Dive3 min↗04Session Actor: Threads, State & CancellationAnalyzing session state ownership, message routing, background tasks, and the CancellationToken interrupt pathDeep Dive4 min↗05Compaction: 85% Threshold & Optional Two-PassVerifying auto-compression thresholds, memory flush, two-pass, and timeout budget from real configInteractive3 min↗06PromptContext: Inspectable Rendering InputDeconstructing serializable context, TemplateOverride, and the TemplateRenderer rendering boundaryConcept3 min↗07Process-Level External Toolset Preset RegistryUnderstanding build functions, Public vs Internal visibility, and how late registration affects downstream resolutionArchitecture4 min↗08ToolKind Provides Default Read-Only SemanticsTracing read-only defaults and capability filtering boundaries from the enum and is_read_only()Design Pattern4 min↗09Implementation Families, Registry & Dynamic MCPDistinguishing built-in tool implementations, the static registry, and runtime-discovered MCP toolsDeep Dive4 min↗10Canonical Input Is a Stable ProjectionUsing CanonicalToolMeta and input projection to explain stable contracts across tool implementationsCase Study4 min↗11Estimation, Percentages & Strict ThresholdsDistinguishing Token estimation, usage-rate calculation, and the strict comparison semantics of exceeds_thresholdInteractive4 min↗12From File Changes to Hybrid RankingTracing the memory recall pipeline: FTS, vector retrieval, time decay, and MMR re-rankingDeep Dive5 min↗13Dream: The Real MechanismVerifying idle gating, DreamLock, background tidying, and the actual boundaries of memory write-backFrontier4 min↗14How AgentDefinition & Persona MergeDeconstructing Agent definitions, Persona overrides, and the merge order of final session behaviorDesign Pattern5 min↗15Four Isolation Dimensions of Sub-AgentsAnalyzing isolation boundaries across context source, recovery mode, worktree, and task stateDeep Dive5 min↗16Multi-Agent Organization PatternsComparing Agents, Personas, coordinators, and parallel tasks based on public evidenceCase Study5 min↗17Five Sandbox ProfilesComparing workspace, devbox, read-only, strict, off, and custom Profile boundariesSecurity5 min↗18From Tool Request to Restricted ExecutionTracing the full authorization chain through ToolKind, permission decisions, and platform sandboxingSecurity6 min↗19Hooks: Only Explicit Deny BlocksVerifying lifecycle events, matchers, PreToolUse blocking, and fail-open semantics on errorsHands-on6 min↗20MCP Connection, Discovery & RecoveryConfirming the client role; deconstructing OAuth, tool naming, capability discovery, state merging, and reconnectionDeep Dive6 min↗21Plugin Marketplace: Discovery & TrustDistinguishing catalog, installation, runtime discovery, enablement state, and plugin root trustArchitecture6 min↗22Grok Build vs Claude Code: Evidence-Based ComparisonMulti-dimensional comparison based on source code, repo docs, and public product behavior — preserving unknownsRecap7 min↗23Grok Build Engineering RetrospectiveReviewing engineering strengths and limitations through types, state machines, tests, and repo policiesFinale6 min↗24Coding Agent Design WorkbenchProducing architecture decisions, failure paths, verification methods, and course deliverables across nine system dimensionsFinale6 min↗25Dissecting Grok Build · 30 Tough QuestionsEach with intent, framework, and bonus points: runtime loop / Compaction / tool permissions / memory retrieval / sandbox security / MCP integrationQuiz17 min↗