# Agents Ebook > Design and implementation notes for Ikidna an agentic swarm for autonomous software delivery. Every page below links to its raw markdown source. The full corpus as a single file is at https://agents.ebooks.kiberonlabs.com/llms-full.txt. ## Overview - [Agentic Engineering](https://agents.ebooks.kiberonlabs.com/index.md): Software teams are learning to hand whole units of work to autonomous agents. Not to autocomplete a line, but to take an issue and return a reviewed, merged change. Doing that rel… - [Overview](https://agents.ebooks.kiberonlabs.com/overview.md): Kiberon labs is documenting an architecture for autonomous software delivery: a coordinated system of agents that moves work from issue through enrichment, planning, implementatio… ## Preface - [Preface](https://agents.ebooks.kiberonlabs.com/preface.md): This is a design book about building a system that writes software on its own. Not a code assistant that finishes your sentence, but a coordinated set of agents that takes an issu… ## Concepts - [Agent Architecture](https://agents.ebooks.kiberonlabs.com/concepts/agent-architecture.md): The diagram below captures the current high-level architecture for a Ikidna agent runtime. - [Agent-Computer Interface (ACI)](https://agents.ebooks.kiberonlabs.com/concepts/agent-computer-interface.md): The Agent-Computer Interface is the surface through which an agent perceives and acts on its environment: the file viewer, the edit command, the search tool, the shell wrapper, th… - [Agent Readiness](https://agents.ebooks.kiberonlabs.com/concepts/agent-readiness.md): Most attempts to improve an agent start with the words: "A better prompt, a longer CLAUDE.md, another skill". When results stay flat, the artifacts pile up: prompt collections nob… - [Agent-to-Agent Protocols](https://agents.ebooks.kiberonlabs.com/concepts/agent-to-agent-protocols.md): The Coordination Model explains when many agents should collaborate at all. This chapter covers the narrower interoperability question: what protocol do they use when they need to… - [The Agentic Swarm](https://agents.ebooks.kiberonlabs.com/concepts/agentic-swarm.md): Everything in this book is in service of the agentic swarm is a group of autonomous agents that collaborate toward a shared goal, rather than a set of isolated assistants each ans… - [Context](https://agents.ebooks.kiberonlabs.com/concepts/context.md): Context is everything an agent knows at the moment it acts. Unlike a human who accumulates background knowledge passively over years, an agent has no persistent world model betwee… - [Context Development Lifecycle](https://agents.ebooks.kiberonlabs.com/concepts/context/development-lifecycle.md): Context quality requires a continuous loop, not one-time authoring. - [Context Evaluation](https://agents.ebooks.kiberonlabs.com/concepts/context/evaluation.md): When context is changed, evaluate its effect with both quantitative and qualitative measures. - [Context Topics](https://agents.ebooks.kiberonlabs.com/concepts/context/index.md): This section groups context scope, lifecycle, and evaluation into one navigable area. - [Context Layers](https://agents.ebooks.kiberonlabs.com/concepts/context/layers.md): Context does not arrive at an agent in one undifferentiated block. It has natural scope boundaries that determine how broadly it applies, how frequently it changes, and who is res… - [Runtime Context Management](https://agents.ebooks.kiberonlabs.com/concepts/context/runtime-management.md): The foundational Context model is about authoring quality density, hierarchy, curation, evolution. This page covers the complementary problem: keeping context coherent and high-si… - [Controls & Autonomy](https://agents.ebooks.kiberonlabs.com/concepts/controls.md): Every agent system answers one question, whether its designers ask it or not: how much behaviour is delegated to agents, and how much is constrained by structure, guardrails, and… - [The Coordination Model](https://agents.ebooks.kiberonlabs.com/concepts/coordination-model.md): The Agentic Swarm names the target system as many agents collaborating toward a shared goal. This chapter addresses the harder question the name leaves open: how they collaborate.… - [The Evaluation Harness](https://agents.ebooks.kiberonlabs.com/concepts/evaluation-harness.md): Skill Evaluation measures a single packaged skill. Context Evaluation measures a context layer. This chapter covers the level above both: evaluating the whole system. Does the end… - [Execution Ledger](https://agents.ebooks.kiberonlabs.com/concepts/execution-ledger.md): The Execution Ledger is a unified system of record for all agent activity. It binds together invocations, actors, artifacts, tool calls, cost signals, and arbitrary metadata into… - [The Harness](https://agents.ebooks.kiberonlabs.com/concepts/harness.md): Part I ended with a formula: an agent is a model plus a harness. The model is rented; every team at the frontier has the same one. The harness is the execution environment wrapped… - [Knowledge Base](https://agents.ebooks.kiberonlabs.com/concepts/knowledge-base.md): A knowledge base is any structured or semi-structured information surface an agent can query to support reasoning, planning, and execution. - [Adoption Sequence](https://agents.ebooks.kiberonlabs.com/concepts/knowledge-base/adoption-sequence.md): Adopt retrieval-reasoning architecture incrementally. Most systems should begin with strong retrieval, then add routing, then add exhaustive reasoning where retrieval alone fails. - [Knowledge base](https://agents.ebooks.kiberonlabs.com/concepts/knowledge-base/index.md) - [RAG-RLM Hybrid Architecture](https://agents.ebooks.kiberonlabs.com/concepts/knowledge-base/rag-rlm-hybrid-architecture.md): RAG and RLM solve different problems in the same system. - [Retrieval-Reasoning Spectrum](https://agents.ebooks.kiberonlabs.com/concepts/knowledge-base/retrieval-reasoning-spectrum.md): Knowledge base queries are not uniform. They lie on a spectrum from focused retrieval to exhaustive reasoning. - [Routing Heuristics](https://agents.ebooks.kiberonlabs.com/concepts/knowledge-base/routing-heuristics.md): A router should classify query intent before retrieval begins. The objective is to separate focused lookups from exhaustive synthesis and route each to the cheapest path that pres… - [MCP](https://agents.ebooks.kiberonlabs.com/concepts/mcp.md): MCP (Model Context Protocol) is the interface layer that lets agents discover and use external capabilities in a structured way. It provides a common contract between an agent run… - [Agent Pool](https://agents.ebooks.kiberonlabs.com/concepts/primitives/agent-pool.md): An agent pool is the static definition and factory layer for agents. It contains the code and configuration needed to scaffold agents, including supporting concerns such as memory… - [Agent](https://agents.ebooks.kiberonlabs.com/concepts/primitives/agent.md): An agent is the runtime system that autonomously performs work using a workflow plus an available tool harness. In the current repo model, agents are not standalone definitions; t… - [Agent Cohort](https://agents.ebooks.kiberonlabs.com/concepts/primitives/cohort.md): While Agent Pools are factories that define individual agents, cohorts define relationships between agents and organizational units. A cohort is a higher abstraction than a pool:… - [Foundational AI Model](https://agents.ebooks.kiberonlabs.com/concepts/primitives/foundational-ai-model.md): A foundational AI model is a large, general-purpose LLM model trained on broad datasets so it can perform many tasks without being built for only one narrow domain. - [Orchestrator](https://agents.ebooks.kiberonlabs.com/concepts/primitives/orchestrator.md): An Orchestrator is the runtime primitive responsible for receiving work, maintaining a durable task registry, and dispatching that work to the correct agent or agent factory. It i… - [Prior Art & Lessons](https://agents.ebooks.kiberonlabs.com/concepts/prior-art.md): This page anchors the external sources behind the wiki's design decisions and maps each project's hard-won lesson to the Ikidna concept it informs. It exists so that claims elsewh… - [Security & Prompt Injection](https://agents.ebooks.kiberonlabs.com/concepts/security.md): An autonomous coding fleet is a high-value, high-blast-radius attack surface: it holds credentials, reads untrusted input, and can write code, open PRs, and reach the network. The… - [Skill Gateway Comparison with tech-leads-club/agent-skills](https://agents.ebooks.kiberonlabs.com/concepts/skill-gateway-prior-art.md): tech-leads-club/agent-skills (TLC) is an open-source skill registry for AI coding agents. It has approximately 4,500 stars, 57 releases, and a structured catalog of skills across… - [Skill Gateway](https://agents.ebooks.kiberonlabs.com/concepts/skill-gateway.md): Skills in the current system are static files SKILL.md and supporting assets committed to a repository and loaded by an agent harness at startup. This works well for small, stable… - [Skills](https://agents.ebooks.kiberonlabs.com/concepts/skills.md): A skill is a reusable capability package that tells an AI agent how to perform a specific class of work with consistent quality. Instead of relying only on a generic system prompt… - [Skill Format](https://agents.ebooks.kiberonlabs.com/concepts/skills/format.md): The skill format defines how a skill is structured on disk, what files it contains, and which of those files is the AI interface versus system metadata. The format is an extension… - [Software Development](https://agents.ebooks.kiberonlabs.com/concepts/software-development.md): Fully agentic software development should not be understood as "let an LLM write code." The stronger model is an operating system for delivery: humans and agents share issue intak… - [Collaborative Development Structure](https://agents.ebooks.kiberonlabs.com/concepts/software-development/collaborative-development-structure.md): Agentic software development should preserve the strong coordination patterns of human teams instead of bypassing them. The right target is not fewer engineering rituals, but more… - [Economics and Routing](https://agents.ebooks.kiberonlabs.com/concepts/software-development/economics-and-routing.md): Fully agentic software development only scales if model spend, provider concentration risk, and throughput are treated as first-class design concerns. Cost control is not separate… - [Execution and Feedback](https://agents.ebooks.kiberonlabs.com/concepts/software-development/execution-and-feedback.md): The central question in agentic software development is not only which model plans the work, but where the work actually runs and how the system learns whether it succeeded. A cre… - [Operations and Governance](https://agents.ebooks.kiberonlabs.com/concepts/software-development/operations-and-governance.md): An agentic development system becomes real only when it can be operated, inspected, and governed after day 1. The challenge is not merely generating code, but running a persistent… - [Verification & the Outer Loop](https://agents.ebooks.kiberonlabs.com/concepts/verification.md): Verification is the mechanism by which the system distinguishes done from looks done. It is the single most load-bearing concept in autonomous software delivery, because an agent… ## Processes - [End-to-End Orchestration](https://agents.ebooks.kiberonlabs.com/processes/end-to-end-orchestration.md): End-to-end orchestration is the delivery path that moves work from issue creation through enrichment, planning, implementation, review, and rework. It is a sequential pipeline wit… - [Execution](https://agents.ebooks.kiberonlabs.com/processes/execution.md): Execution is the runtime phase where a planned task is implemented by an agent. This page covers the operational concerns of running work chiefly the parallelism decision, which i… - [Guardrails](https://agents.ebooks.kiberonlabs.com/processes/guardrails.md): Guardrails are the operational limits placed on a single agent run to bound its cost, duration, and blast radius. They are the run-level complement to the architectural controls i… - [Model Usage](https://agents.ebooks.kiberonlabs.com/processes/model-usage.md): Model choice is a per-role and per-task decision, not a global one. The three roles in the delivery loop demand different model profiles, and selecting deliberately for each is on… - [Smart Routing](https://agents.ebooks.kiberonlabs.com/processes/optimizations/smart-routing.md): Smart routing is the repository's first explicit optimization strategy: match work with the cheapest model and execution path that preserves answer quality, escalating to expensiv… - [Orchestration Triggers](https://agents.ebooks.kiberonlabs.com/processes/orchestration-triggers.md): Orchestration triggers are events that cause the system to invoke, resume, or re-invoke an agent. They span the full range of system operation from work intake to autonomous recov… - [Planning & Execution](https://agents.ebooks.kiberonlabs.com/processes/planning-and-execution.md): The naive way to hand work to an agent is to describe the task and let it run. It is also the way that fails most expensively. A single-shot agent gathers whatever context it happ… - [Skill Evaluation](https://agents.ebooks.kiberonlabs.com/processes/skill-evaluation.md): Skill evaluation is the process of measuring whether a skill package is doing its job: producing better, more consistent agent behavior for the task class it targets. Without eval… - [Swarm Operations](https://agents.ebooks.kiberonlabs.com/processes/swarm-operations.md): The earlier chapters of this part establish what a swarm is and how its agents coordinate. This chapter is about keeping one running: the operational patterns that appear, in syst… - [Ticket Enrichment](https://agents.ebooks.kiberonlabs.com/processes/ticket-enrichment.md): Ticket enrichment is the intake mechanism that converts a high-level human outline into executable, context-rich work. It is the first stage of the delivery pipeline and a control… ## Software - [Promptfoo](https://agents.ebooks.kiberonlabs.com/software/context/evaluation/promptfoo.md): Promptfoo is an open source CLI and library for evaluating and red-teaming LLM applications, RAG systems, and agents. It was founded by Ian Webster and Michael D'Angelo and built… - [Tessl](https://agents.ebooks.kiberonlabs.com/software/context/evaluation/tessl.md): Tessl positions itself as a package manager and registry for agent skills and context. Its focus is not only storing reusable skills, docs, and rules, but also evaluating how well… - [GitNexus](https://agents.ebooks.kiberonlabs.com/software/context/gitnexus.md): GitNexus indexes any codebase into a persistent knowledge graph capturing dependencies, call chains, execution flows, and functional clusters then exposes that graph to AI agents… - [Graphiti / Zep](https://agents.ebooks.kiberonlabs.com/software/context/graphiti.md): Graphiti is an open-source framework for building temporal context graphs for AI agents. Unlike static knowledge graphs or RAG pipelines, Graphiti tracks how facts change over tim… - [Context and Knowledge Engines](https://agents.ebooks.kiberonlabs.com/software/context/index.md): A context engine is the layer responsible for transforming raw repository and project data into structured, queryable knowledge that agents can consume. Where a harness manages ho… - [Mem0](https://agents.ebooks.kiberonlabs.com/software/context/mem0.md): Mem0 is a memory infrastructure layer for AI agents and applications. It focuses on persistent memory: ingesting interactions and other inputs, extracting what matters, and retrie… - [Current](https://agents.ebooks.kiberonlabs.com/software/current.md) - [ForgeCode](https://agents.ebooks.kiberonlabs.com/software/harnesses/forgecode.md): ForgeCode (Forge) is an open source agentic coding harness built in Rust. It uses a multi-agent architecture with distinct built-in roles for planning, research, and execution, ea… - [Harnesses Index](https://agents.ebooks.kiberonlabs.com/software/harnesses/index.md): Harnesses are the execution environments that wrap an LLM and turn it into an agent. See the Harness concept page for the full definition and Ikidna's evaluation criteria. - [OpenCode](https://agents.ebooks.kiberonlabs.com/software/harnesses/opencode.md): OpenCode is an open source agentic coding harness built for the terminal, desktop, and IDE. It runs a working loop with full tool access against any LLM 75+ providers including lo… - [Pi](https://agents.ebooks.kiberonlabs.com/software/harnesses/pi.md): Pi is a terminal-first coding harness positioned as a minimal core that is extended through skills, extensions, prompt templates, themes, and installable packages. - [Software Index](https://agents.ebooks.kiberonlabs.com/software/index.md): The software included here should be things that are actually considered for implementation, not a never-ending repository of possible software. - [AWS Bedrock](https://agents.ebooks.kiberonlabs.com/software/model-providers/bedrock.md): AWS Bedrock is a managed AWS service for accessing and operating foundation models through AWS-native infrastructure and controls. - [Model Providers Index](https://agents.ebooks.kiberonlabs.com/software/model-providers/index.md): Reference index for model providers, ownership, and core strengths. - [OpenRouter](https://agents.ebooks.kiberonlabs.com/software/model-providers/openrouter.md): OpenRouter is a multi-provider API gateway that exposes many model families behind a unified interface. - [Self-Hosted](https://agents.ebooks.kiberonlabs.com/software/model-providers/self-hosted.md): Self-hosted model providers are organization-operated inference stacks where model serving, scaling, and access control are managed directly by the organization. - [Claude](https://agents.ebooks.kiberonlabs.com/software/models/claude.md): Claude is a frontier model family commonly used for high-reliability reasoning, coding, and tool-augmented workflows. - [DeepSeek](https://agents.ebooks.kiberonlabs.com/software/models/deepseek.md): DeepSeek is a model family frequently positioned for strong price-performance in coding and reasoning workloads. - [Models Index](https://agents.ebooks.kiberonlabs.com/software/models/index.md): Reference index for model families, ownership, and known strengths. - [Kimi K2](https://agents.ebooks.kiberonlabs.com/software/models/kimik2.md): Kimi K2 is a model in the Kimi family, typically associated with long-context and document-centric analysis workloads. - [Minimax](https://agents.ebooks.kiberonlabs.com/software/models/minimax.md): Minimax is a series of language models developed by MiniMax, a Chinese AI company. The models are designed for a variety of natural language processing tasks, including chat, reas… - [Paperclip](https://agents.ebooks.kiberonlabs.com/software/orchestration/paperclip.md): https://paperclip.ing/ · GitHub · Docs · MIT license, self-hosted - [Coder](https://agents.ebooks.kiberonlabs.com/software/tools/mcp/coder.md): Internally at Kiberon Labs we use Coder to provide containers that can scaffold and run VS Code-enabled workspaces to develop code. - [Flux MCP](https://agents.ebooks.kiberonlabs.com/software/tools/mcp/flux.md): Flux is the GitOps provider in the Ikidna architecture (the flux-system namespace): it continuously reconciles the cluster against a Git source of truth, so deployment is a git op… - [Grafana MCP](https://agents.ebooks.kiberonlabs.com/software/tools/mcp/grafana.md): Grafana is the observability surface in the Ikidna architecture (the monitoring namespace), visualizing metrics from Prometheus and traces from Tempo / the OTel collector. The Gra… - [Port MCP](https://agents.ebooks.kiberonlabs.com/software/tools/mcp/port.md): Port is an internal developer platform (IDP) a software catalog and self-service action layer sitting in the External block of the Ikidna architecture. It models the organisation'… ## Ikidna - [Agent Container Runtime Interface (ACRI)](https://agents.ebooks.kiberonlabs.com/ikidna/acri.md): The Agent Container Runtime Interface (ACRI) is the standard interface contract between the Orchestrator and any agent harness. It defines how an agent is created, started, observ… - [Architecture](https://agents.ebooks.kiberonlabs.com/ikidna/architecture.md): ```mermaid --- - [Axioms](https://agents.ebooks.kiberonlabs.com/ikidna/axioms.md): The design of ikidna is informed by a few axioms chosen to guide the architectural principles needed for the system. - [Minimal Goals](https://agents.ebooks.kiberonlabs.com/ikidna/bootstrap/minimal-goals.md): Minimal goals are the smallest durable set of objectives an autonomous system needs in order to direct itself over time. They are the goal layer behind minimal-bootstrap: if the s… - [Goals](https://agents.ebooks.kiberonlabs.com/ikidna/goals.md): Ikidna is being developed in two intentional phases. The near-term phase optimizes for immediate usage and value generation, while the longer-term phase optimizes for interoperabi… - [Harnesses](https://agents.ebooks.kiberonlabs.com/ikidna/harness.md): Harnesses are obviously a core part of how the Ikidna system is intended to operate as they form the core runtime/ - [Minimal Bootstrap](https://agents.ebooks.kiberonlabs.com/ikidna/minimal-bootstrap.md): A minimal bootstrap is the smallest coherent set of conditions under which an AI system becomes capable of self-directed, ongoing improvement. Below this threshold the system can… ## foundations - [The Agent Loop](https://agents.ebooks.kiberonlabs.com/foundations/agent-loop.md): Strip any agent to its skeleton and you find the same cycle: think, act, observe, repeat. The model reasons about what to do, takes one action through a tool, reads the result, an… - [What an Agent Is](https://agents.ebooks.kiberonlabs.com/foundations/agents.md): A language model answers questions. An agent takes actions: it reads files, runs commands, calls services, and keeps going, deciding step by step what to do next until the work is… - [Part I: One Agent](https://agents.ebooks.kiberonlabs.com/foundations/introduction.md): Every system in this book, is built out of one repeating unit: a single agent doing a single task well. Teams that skip this level and jump straight to orchestration inherit every… - [Working With Models](https://agents.ebooks.kiberonlabs.com/foundations/models.md): Agentic engineering relies on the model as the underlying basis, which is inherently a probabilistic system. Engineering with such a component is a different discipline from ordin… - [Instructing an Agent](https://agents.ebooks.kiberonlabs.com/foundations/prompting.md): A prompt is not the sentence you type. By the time a model acts, its instruction is the union of a system prompt, your message, injected files and tool outputs, and the structural… - [Giving an Agent Tools](https://agents.ebooks.kiberonlabs.com/foundations/tools.md): An agent without tools is a chatbot. Tools are where reasoning becomes action, and a surprising share of total agent capability lives in how they are designed rather than in the m…