Custom Claude Skill or Agent Specification
A complete proprietary specification for a bespoke Claude Skill or autonomous agent. PRD, architecture, system prompts, guardrails, evaluation metrics, deployment plan, observability framework, inference cost model. The spec you hand to your engineering partner before they spend €200K on the wrong build.
PRICE
From €4,900
DELIVERY
12 days
FORMAT
Partner-led
DISCUSS YOUR PROJECT →
SEE SAMPLE SPECIFICATION →
§ 01 · The problem
"We greenlit a six-figure Claude project on a two-page brief."
Most enterprise AI projects start with a deck slide and a confident "we should build this." Engineering scopes from rough requirements. Six months in, the system works on the demo dataset but fails in production. Guardrails are missing. Costs explode. Nobody can answer the simple question: "is this actually working?"
The Custom Specification fixes this at the source. Before a single line of code is written, we deliver the document that defines what to build, how to test it, what could go wrong, and what success actually looks like. The engineering team builds the right thing. The CFO knows what they're paying for. The board can be told a defensible story.
§ 02 · What you get
A complete build specification — engineering-ready.
Six deliverables that turn an ambition into a buildable plan. Detailed enough for your engineering partner to start work the day after delivery — without "we'll figure that out later."
DELIVERABLE 01
Product Requirements Document
20-30 page PRD: user journeys, functional requirements, non-functional constraints, success criteria. The document your engineering partner asks for and you usually don't have.
DELIVERABLE 02
Target architecture
System diagram, data flow, integration points, model selection rationale, fallback patterns. The diagram your CTO needs to validate before signing the build budget.
DELIVERABLE 03
System prompts & guardrails
Drafted system prompts for each agent role. Refusal patterns. Tool-use boundaries. Escalation triggers. The behavioral spec that prevents your assistant from doing something it shouldn't.
DELIVERABLE 04
Evaluation framework
Test sets, metrics definitions, acceptance thresholds, golden examples. How you'll know it works — before it ships, after it ships, every quarter after that.
DELIVERABLE 05
Deployment & observability plan
Rollout strategy (shadow, pilot, full deployment), telemetry schema, dashboards, incident response. The operational backbone that turns a demo into a system you can run.
DELIVERABLE 06
Inference cost model
3-year inference cost projection: model mix, token volumes, caching strategy, scaling assumptions. The number your CFO wants to see before the project starts, not after.
§ 03 · The method
A specification framework refined across enterprise builds.
Five-stage spec methodology, each stage producing artefacts your engineering team will actually use. We don't ship vague PowerPoint — we ship the document the build is contracted against.
STAGE 01
Discovery & intent
What problem are we actually solving? Who uses this? What does success look like in measurable terms? We pressure-test the ambition against operational reality before specifying anything.
STAGE 02
Behavior specification
User journeys, agent capabilities, refusal patterns, tool-use boundaries. The behavioral envelope of what the Skill will do — and the explicit list of what it won't.
STAGE 03
Architecture & integration
Target architecture, model selection rationale, integration touchpoints, data flows, fallback patterns, latency budget. The technical envelope your engineering team needs.
STAGE 04
Evaluation & observability
Test sets, golden examples, acceptance thresholds, telemetry schema, monitoring dashboards. The instrumentation that turns "we built it" into "we know it works."
STAGE 05
Economics & rollout
Inference cost model over 3 years, deployment plan (shadow → pilot → production), risk register, incident playbook. The economic and operational reality check.
§ 04 · Process & timeline
Twelve days. Four sessions. Engineering-ready spec.
Partner-led from intent to engineering handover. The same person who scopes your problem writes your spec. No team rotation, no junior interpretation.
DAYS 01–02
Discovery
90-min framing session + 2-3 stakeholder interviews. We map the problem, the users, the constraints, the success criteria. Hypotheses surface.
DAYS 03–06
Spec drafting
PRD, behavior spec, architecture diagram, draft prompts & guardrails. Co-authored with your tech lead via async exchanges. First draft delivered.
DAY 07
Pressure-test session
90-min joint session with your engineering and product leads. We stress-test every assumption, surface gaps, refine the spec. Risk register populated.
DAYS 08–11
Finalization
Evaluation framework, observability schema, inference cost model, deployment plan. All six deliverables finalized and cross-referenced.
DAY 12
Engineering handover
90-min handover session with your engineering partner. Walkthrough, Q&A, edge case discussion. They leave ready to scope and build. You leave with a defensible artefact.
§ 05 · Who it's for
For leaders about to greenlight a six-figure build.
The right time to invest in a specification is before the development budget commits — not after. Built for the four roles that own the decision.
CHIEF TECHNOLOGY OFFICER
You're being asked to scope a Claude build from a 4-slide deck. You need a real specification before committing engineering capacity for six months.
HEAD OF AI / CDO
You own the AI roadmap and a custom Skill keeps surfacing in conversations. You need a defensible spec before approaching engineering partners for quotes.
CHIEF PRODUCT OFFICER
You're embedding an AI capability into your core product. The PRD, evaluation framework, and guardrails need to be rigorous before engineering starts the work.
PE OPERATING PARTNER
A participation wants to build a custom AI capability. You want third-party rigor on the spec before they spend the next funding round on engineering.
§ 06 · Frequently asked
Six honest answers to common questions.
Why "from €4,900"? What drives the price up?
Three factors: scope complexity (single Skill vs multi-agent system), number of integration touchpoints (CRM, ERP, internal APIs), and regulated industry constraints (healthcare, finance, public sector). The framing call sets the final price — and we're transparent about what drives it.
Do you build the Skill afterwards?
No. Odyssey specifies — we don't build. We hand the spec to your existing engineering team or your preferred build partner. If you don't have one, we can recommend three vetted partners we've worked with. We stay independent on build economics.
Why not have my engineering team write the spec?
They can — and many do. But engineering teams often skip the parts they'll figure out later: evaluation framework, inference cost model, refusal patterns, observability. Those are exactly the parts that cost the most to retrofit. We bring the discipline that builds time-pressured teams under-invest in.
Is this only for Claude, or any LLM?
Our methodology is model-agnostic but our depth is on Anthropic's Claude — system prompts, tool use, Skills primitive, Claude Code. We specify multi-model architectures (Claude + smaller models for routing) and call out where model choice matters. We don't ship vendor-neutral mush.
What if we discover the project isn't viable?
That's a valid outcome — and a valuable one. We've delivered specs that concluded "this should be a workflow automation, not a Skill" or "the value isn't there for the build cost." Saving €200K on a wrong build is worth €4,900. We won't manufacture a "yes" you didn't ask for.
How does this fit with the Make or Buy Strategy (Skill 06)?
Sequential. Skill 06 answers "should we build, buy, or partner on this AI capability?" If the answer is "build", this Custom Specification answers "how, exactly?". You can pair them, or use this Spec directly if Make-vs-Buy is already settled.
§ Ready when you are
Before €200K on the wrong build, €5K on the right spec.
From €4,900, twelve days. The document your engineering partner needs and your board can defend.
DISCUSS YOUR PROJECT →
SEE SAMPLE SPECIFICATION →