# Claude Experts — full index for language models Canonical site: https://claudeexperts.agency Publisher: Claude Experts, an independent studio. Not affiliated with or endorsed by Anthropic. Contact: book a 15-minute discussion at the Calendly URL on the homepage. ## Business Claude Experts embeds senior Claude specialists in client repositories. Engagements are fixed fee and cover production agents, Model Context Protocol servers, evaluation harnesses, guardrails and team enablement. IP stays with the client. Primary services: Claude Expert Audit, Production Agent Build, MCP and Tool Integration, Eval and Guardrail Harness, Team Enablement, Embedded Claude Expert. Home: https://claudeexperts.agency/ Services: https://claudeexperts.agency/#services Pricing: https://claudeexperts.agency/#pricing Insights: https://claudeexperts.agency/blog ## Insights abstracts ### Claude Agent SDK Session Recovery: Resume, Fork, Compact, and Checkpoint Safely URL: https://claudeexperts.agency/blog/claude-agent-sdk-session-recovery Published: 2026-09-01 Section: production Learn when to continue, resume, fork, compact, and checkpoint Claude Agent SDK sessions so long-running agents can recover safely. Long-running agents fail in ordinary ways: a worker restarts, a context window tightens, an experiment needs a clean branch, or a task moves to another host. A production recovery design starts by separating two things that are often treated as one: the conversation transcript and the durable work state. Use the right recovery operation Choose the operation from the failure mode, not from convenience. - Continue when... ### Claude Agent Observability: Trace Model Calls, Tools, Cost, and Failures URL: https://claudeexperts.agency/blog/claude-agent-observability-opentelemetry Published: 2026-08-31 Section: quality A production guide to tracing Claude model calls, tools, cost, failures, and task outcomes with OpenTelemetry-shaped telemetry. The short answer An agent that returns an answer is not necessarily healthy. Production teams need to know which request started a run, which model calls and tools followed, where time and cost accumulated, what failed, and whether the user task succeeded. This guide shows how to use an OpenTelemetry-shaped operating model for a Claude agent without turning traces into a second data leak. Start with a trace that foll... ### Why most Claude pilots never reach production URL: https://claudeexperts.agency/blog/why-claude-pilots-stall Published: 2026-08-25 Section: production Most Claude pilots stall on integration, evaluation and cost, not prompting. Here is what actually blocks a production launch, and how to unblock it. A surprising number of Claude pilots look healthy in week two and are still a demo in month nine. The prompt is rarely the reason. The system around the model is. This is the pattern we see when we are asked to audit a stalled build. Use it as a checklist before you spend another quarter on a prototype that cannot take traffic. The prompt was never the bottleneck Teams over-invest in prompt wording because it is visi... ### Put eval gates on your Claude agent before you scale traffic URL: https://claudeexperts.agency/blog/claude-eval-harness Published: 2026-08-18 Section: quality An eval harness turns Claude quality into a CI number. Here is the minimum setup that catches regressions before customers do. Most teams shipping Claude still decide quality in a meeting. Someone pastes two answers into Slack. Someone else has a feeling. The change ships anyway. That works until the agent is allowed to touch a real system. Then a quiet regression in retrieval, a tool schema tweak, or a model default you did not notice will land on a customer before anyone has language for what broke. An eval harness is how you stop arguing... ### MCP versus ad-hoc Claude tools: pick the contract, not the acronym URL: https://claudeexperts.agency/blog/mcp-vs-ad-hoc-claude-tools Published: 2026-08-11 Section: production MCP is a protocol. The production question is whether Claude has a narrow, permissioned contract with your systems of record. Model Context Protocol (MCP) is a standard way to expose tools and data to Claude. It is not, by itself, a production integration. The work that matters is the contract: which actions exist, who can call them, and what happens when they fail. Teams stall when they treat MCP as a wrapper around an internal API. Claude then sees fifty overlapping endpoints, weak error messages and write paths with no review. That is ho... ### Prompt caching is how most Claude bills actually come down URL: https://claudeexperts.agency/blog/prompt-caching-claude-cost Published: 2026-08-04 Section: quality Uncached long context is the usual 3x overspend. Cache the stable prefix, budget the rest, and measure cost per successful task. Prompt caching reuses the stable prefix of a Claude prompt so you do not pay full price to re-read the same system instructions, tool schemas and retrieval boilerplate on every turn. It is the highest-leverage cost control we install on inherited agents. The typical overspend we measure is about 3.4x, usually from context that did not need to be there. Teams notice the invoice before they notice the architecture. Cac... ### Claude agent architecture that survives real traffic URL: https://claudeexperts.agency/blog/claude-agent-architecture Published: 2026-07-28 Section: production A production Claude agent is a loop with budgets, retries and checkpoints, not a chat window with tools attached. A Claude agent is a model plus a loop: observe, choose a tool or a reply, act, and stop. Production architecture is the set of limits around that loop so it cannot wander, overspend or write the wrong record. Demos hide the loop. They run three happy-path steps on a staging tenant. Traffic introduces timeouts, partial payloads, ambiguous user intent and tools that return empty. The architecture has to assume those ar... ### Prompt injection defense for Claude agents that can write URL: https://claudeexperts.agency/blog/prompt-injection-defense-claude Published: 2026-07-21 Section: shipping Injection is a product and permissions problem. Treat untrusted text as data, constrain tools, and put a human on irreversible actions. Prompt injection is when untrusted text (a webpage, an email, a retrieved document) tries to override the instructions you gave Claude. For a chatbot that only talks, the blast radius is embarrassment. For an agent that can send mail or update Salesforce, the blast radius is a real incident. Defense is not a magic system prompt. It is layering: untrusted content marked as data, tools that cannot do everything, and a... ### Claude Code workflows that your engineers will actually keep URL: https://claudeexperts.agency/blog/claude-code-engineering-workflows Published: 2026-07-14 Section: shipping Claude Code is useful when the repository has conventions, evals and a review path. Without those it is a faster way to merge guesswork. Claude Code is Anthropic's agentic coding workflow in the terminal and the IDE. It is not a replacement for engineering judgment. Used well, it shortens the path from task to pull request. Used as "just let it rip on main," it creates review debt you will pay for months. The teams that keep it are the ones who treat the repository as the product: conventions, tests, and a human review gate that still means something.... ### Shipping Claude agents under HIPAA, SOC 2 and SOX URL: https://claudeexperts.agency/blog/hipaa-soc2-claude-agents Published: 2026-07-07 Section: shipping Regulated Claude work is a deployment and audit-log problem first. Model quality is necessary and still not sufficient for a security review. A regulated Claude agent is one that can see protected health information, financial records or privileged legal text, and that has to survive a security review. The model is the easy part. The hard part is data flow, logging, access and a refusal path a lawyer can read. We work in HIPAA, SOC 2, SOX and privilege-aware legal settings. Those engagements ship a compliance architecture document with the code. They do no... ### Retrieval for Claude agents: hybrid search, not a bigger window URL: https://claudeexperts.agency/blog/retrieval-for-claude-agents Published: 2026-06-30 Section: production Long context does not replace retrieval. Production agents need a small, cited, testable set of chunks and a budget for what enters the prompt. Retrieval is how a Claude agent gets the facts it was not trained on: tickets, policies, SKUs, the current account. Dumping a 200k-token dump into the window feels like retrieval. It is usually just an expensive way to confuse the model and the bill. See prompt caching and cost (/blog/prompt-caching-claude-cost). Production retrieval is hybrid when it needs to be: keyword for identifiers and exact policy clauses, emb... ### Where to put a human in a Claude agent loop URL: https://claudeexperts.agency/blog/human-in-the-loop-claude-agents Published: 2026-06-23 Section: shipping Human review belongs on irreversible writes and ambiguous intent, not on every token. Design the queue like a product, not a guilt valve. Human-in-the-loop means a person approves, edits or rejects an agent action before it lands. It is not a synonym for "we read the Slack dump." If the queue is unstructured, people rubber-stamp and you have theatre. Put humans where the expected cost of a mistake is high: money movement, customer messages, record updates in a system of record, anything regulated. Let the agent run freely on drafts, classifications and... ### Structured outputs make Claude agents callable by software URL: https://claudeexperts.agency/blog/structured-outputs-claude Published: 2026-06-16 Section: quality If downstream code cannot parse the answer, you do not have an integration. Use schemas, validate, and eval the JSON, not the vibe. Structured output means Claude returns data that software can parse: JSON that matches a schema, not a paragraph that happens to contain a number. Chat UIs can live on prose. Production agents cannot. The next hop is your API, your queue, your warehouse. We treat schema failures as test failures. "It usually looks right" is how silent writes go missing. What to structure - Tool arguments - Decision records (action, c... ### Claude tool schema design: small actions, loud errors URL: https://claudeexperts.agency/blog/claude-tool-schema-design Published: 2026-06-09 Section: production Tool schemas are the product Claude actually uses. Narrow the action, require the fields, and make failure obvious in traces. A tool schema is the JSON contract Claude sees for a side effect: name, description, required fields, types. It is not documentation for humans. It is the API of your agent. Weak schemas produce retries, invented ids and writes you cannot explain. Good schemas look boring. That is the point. Design rules we use on audits 1. One job per tool. Split "search" from "update." 2. Required fields for anything the system of... ### Claude on Bedrock, Vertex or the Anthropic API: pick for control, not fashion URL: https://claudeexperts.agency/blog/claude-bedrock-vertex-or-api Published: 2026-06-02 Section: shipping The right Claude runtime is about data residency, IAM and procurement, not a leaderboard. Match the platform to the review you have to pass. Claude runs on the Anthropic API , Amazon Bedrock and Google Vertex . Quality differences on a given model generation are usually smaller than the differences in IAM, networking, logging and legal paper. Pick the runtime that your security review can sign. A startup with no residency constraint can start on the Anthropic API and move later. A bank that already lives in AWS will often want Bedrock. A public-sector buy... ### What a two-week Claude Expert Audit actually inspects URL: https://claudeexperts.agency/blog/claude-expert-audit-checklist Published: 2026-05-26 Section: quality An audit is traces, spend, tools, safety and a ranked remediation plan you can retest. It is not a slide deck of prompting tips. A Claude Expert Audit is two weeks inside your codebase and traces. The job is to find where quality, safety or money leaks, then hand back a ranked plan with baselines. It exists because most stalled pilots do not need a rewrite. They need a diagnosis. Background: why pilots never reach production (/blog/why-claude-pilots-stall). The inspection list 1. Prompt and context. What is cached, what is retrieved, what is d... ### Context engineering for Claude: budget the window like memory URL: https://claudeexperts.agency/blog/context-engineering-claude Published: 2026-05-19 Section: production Context engineering is choosing what Claude sees, in what order, and what is cached. It is the discipline behind quality and spend. Context engineering is the work of assembling what Claude sees on each turn: instructions, tools, retrieval, history, and the user. Prompt wording is a subset. Most production failures we inherit are context failures: too much, in the wrong order, uncached, uncited. Treat the window like a memory budget, not a junk drawer. A working budget - Policy and tools first, stable, cached. See prompt caching (/blog/prompt-cac... ### Multi-agent Claude systems: when a second agent earns its keep URL: https://claudeexperts.agency/blog/multi-agent-vs-single-agent-claude Published: 2026-05-12 Section: production Split agents when you have a real seam (research versus mutate, or two systems of record). Do not split them to look advanced. A multi-agent Claude system is more than one model loop coordinating: a researcher, a writer, a reviewer, a specialist on SAP. It is justified when a single agent with a small tool set cannot keep permissions or context straight. It is not justified because a diagram looked impressive. Start single. Add a second agent when you can name the seam. Seams that are real - Read-only research versus mutating a system of rec... ### Put Claude eval gates in CI or accept shipping on instinct URL: https://claudeexperts.agency/blog/ci-eval-gates-claude Published: 2026-05-05 Section: quality If the golden set is not a pull request check, it will not run the week you are late. Block on safety and writes; warn on softer slices. An eval gate is a check that fails a pull request when Claude quality, safety or tool behaviour drops. Notebooks are research. Gates are how you ship. The longer version of the design lives in how to put eval gates on a Claude agent (/blog/claude-eval-harness). This note is the CI shape. What to block on Block merge when: - A safety or injection case fails - A write-path case fails (wrong tool, missing fields, skippe... ### Claude latency in production: timeouts, streaming and budgets URL: https://claudeexperts.agency/blog/claude-latency-production Published: 2026-04-28 Section: quality Latency is a budget. Cap steps, cache the prefix, stream when a human is waiting, and fail before the user does. Latency for a Claude agent is not one model round-trip. It is the sum of retrieval, each tool, each model call and your own queue. Users feel the sum. Your SLOs should too. Unbounded loops are how a six-second demo becomes a 90-second production stall. Put ceilings on steps and wall-clock time in agent architecture (/blog/claude-agent-architecture), then tune. Levers that actually move p95 1. Cache the stable prefix.... ### Train your engineers on Claude using your codebase, not a sample app URL: https://claudeexperts.agency/blog/claude-team-enablement Published: 2026-04-21 Section: shipping Enablement that sticks is prompting, tools, evals and Claude Code on the repository you ship. Exam prep is optional. Habits are not. Team enablement here means your engineers can change a Claude system without waiting for us, and without guessing. A slide deck about prompting does not do that. Four weeks on your repository does. We teach prompting, tool design, evaluation and agentic coding where the code already lives. Anthropic exam prep is available for people who want the credential. The credential is not the point. The handover (/ process) is... ### Claude computer use in production: isolate the desktop, log the clicks URL: https://claudeexperts.agency/blog/claude-computer-use-production Published: 2026-04-14 Section: production Computer use is an interface of last resort. Sandbox it, constrain the goal, and prefer APIs or MCP when the system of record has one. Computer use lets Claude operate a GUI: browser, desktop, internal thick clients that have no API. It is powerful and easy to romanticise. In production it is an interface of last resort. If Salesforce has an API, use MCP or tools (/blog/mcp-vs-ad-hoc-claude-tools). Do not click the UI because a demo looked like a human. When the system of record truly has no API, computer use can still be the right seam. Treat it li...