agentleFS
JuankCadavidJuankCadavid/akili-specs/.claude/skills/ai-agent-development/SKILL.md

ai-agent-development

Skill12 starsChanged 13 days agoMITView on GitHub

Trigger: agent development, agent architecture, LangGraph, LangChain, Deep Agents, Bedrock AgentCore, Claude Agent SDK, multi-agent, agent evals, framework selection. Decision layer for building AI agents: framework and runtime selection, agent architecture, and the mapping from agent behavior to AKILI spec documents; defers runtime detail to the skills and docs that own it.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If the three answers do not converge on one option — or the task names a framework that must be justified against alternatives — open …

    line 38

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

AI Agent Development — Decision Layer

This skill decides what to build an agent with and how to shape it. It operates nothing: framework APIs, SDK calls, and deployment mechanics belong to the skills and docs named in the deferral table below.

This file is the router. Read it, answer the three routing questions, then load only the references/ file(s) the task actually needs.

Scope & Composition

Owns: framework and runtime selection; agent architecture (state, tools, memory, human-in-the-loop points, multi-agent topology); the hosting choice for agent workloads; and the bridge from agent behavior to requirements.md, design.md, and tasks.md.

Does not own: API signatures, SDK usage, service configuration, deployment implementation.

Composition, not exclusion. This skill composes with the skills that own operation — it never displaces them. Their triggers legitimately overlap on agent-shaped tasks, so precedence goes by question type:

  • Decision question — "which framework", "how should this agent be structured", "where should it run" → load this skill first.
  • Operation question — "how do I call this API", "how do I configure this service" → load the runtime skill first.
  • Both may be loaded in one task. A task that chooses and builds needs both, in that order.

Routing Heuristic — three questions

Answer these before opening the matrix; they shortlist the option in most cases.

  1. Where does it run? Your own process or container, a managed agent runtime, or inside an existing application. This is usually the hardest constraint, so it eliminates options first.
  2. How much orchestration control do you need? A prebuilt agent loop (tool-calling until done) versus an explicit graph you author (branching, parallelism, custom control flow). Control costs authoring effort — buy the loop when the behavior fits it.
  3. What must survive restarts, and where do humans intervene? Durable state across process death, resumable runs, and approval checkpoints are the requirements that most often force a framework instead of a plain SDK loop.

If the three answers do not converge on one option — or the task names a framework that must be justified against alternatives — open references/framework-selection.md.

Runtime Deferral

Runtime detail is out of scope by design. Availability is per-developer, not per-project: never assume a skill is installed, so every row names its substitute route.

QuestionRoute
Bedrock / AgentCore operation, Harness, guardrails configamazon-bedrock (when present; official AWS Bedrock docs otherwise)
Anthropic SDK, tool use, MCPclaude-api (when present; official Anthropic docs otherwise)
Lambda / ECS / API Gateway implementation of a chosen hostingaws-serverless — packaged stack skill (when present; official AWS docs otherwise)
Framework API signatures, version-specific detailPinned official docs — see the ## Sources block of each reference

references/aws-deployment.md owns only the choice between hostings. Implementing the chosen one is the third row's job.

References — load per task

ReadWhen the task involves
references/framework-selection.mdChoosing or justifying a framework or runtime; the three routing questions did not converge; a user proposes an option you must place (including CrewAI or AutoGen)
references/langgraph-patterns.mdLangGraph already chosen or in use — state design, nodes and edges versus prebuilt agents, durable execution, HITL interrupts, streaming, memory, subgraphs
references/akili-spec-mapping.mdWriting spec documents for an agent feature — behaviors and evals as scenarios, architecture as design decisions, eval work as tasks with verification commands
references/aws-deployment.mdDeciding where an agent workload runs on AWS — AgentCore Harness versus Lambda versus ECS trade-offs for agent workloads

More than one can apply: choosing a framework and then specifying the feature is framework-selection.md + akili-spec-mapping.md.

Maturity

The agent ecosystem churns faster than this skill is revalidated. Every factual claim here and in the references is pinned to an official-docs link, and the matrix carries a Last verified: date — check it before relying on a comparison.

Guidance not yet validated in a real build is marked inline with > Unvalidated: rather than omitted. The marker is greppable (grep -rn "> Unvalidated:" .claude/skills/ai-agent-development/), so the maturity boundary is auditable instead of implied. Treat marked content as a starting hypothesis to verify against the pinned source, not as settled practice.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.