Corporate Claude Code Training at Red Beacon Asset Management
Summary
Case study: a one-day corporate Claude Code training for 10 staff at Red Beacon Asset Management on 28 May 2026 — custom commands, MCP tools, skills, sub-agents and hooks, anchored on a real lead-generation site.
On 28 May 2026, Tertiary Infotech Academy delivered a one-day corporate Claude Code training to 10 staff at Red Beacon Asset Management, an MAS-regulated wealth manager in Singapore. The cohort built a working asset-management landing page with a live lead-capture form, then layered on custom slash commands, MCP-driven testing, skills, agents and hooks — all on Claude Code. Book a corporate Claude Code training for your team →
Why a wealth manager needed agentic AI training
Red Beacon Asset Management operates under a Capital Markets Services licence from the Monetary Authority of Singapore. Like most regulated firms in Singapore, their digital surface — marketing site, lead capture, internal reporting — has to be auditable, fast to update and resilient. The brief from their team was straightforward: stop hand-rolling marketing pages and form integrations, and start producing them with an AI coding agent that anyone on the team can drive.
That is exactly the problem Claude Code is built for — a terminal-native agent that reads your repo, runs your tools, edits files and follows instructions persisted as commands, skills and hooks. The training therefore ran as a build-something-real day rather than a slide tour, anchored on the public WSQ Agentic AI Applications with Claude Code curriculum.
What "good" looks like with Claude Code
Most teams treat AI coding assistants as autocomplete on steroids. That underuses the agent. A team gets real leverage when four primitives are wired into their daily workflow:
Custom slash commands
Reusable prompts stored as Markdown files under .claude/commands/. Typing /deploy or /new-section runs the recipe the team agreed on — the same scaffolded HTML block, the same lead-form snippet, the same accessibility checks every time. The codified knowledge stops living in the senior developer's head.
MCP tools
The Model Context Protocol lets the agent reach external systems through a typed interface — a Playwright browser to test the form end-to-end, a Postgres connection to inspect leads, a GitHub server to file an issue. Instead of "ask the model to write a test", the agent runs the test.
Skills
Long-form domain knowledge — design conventions, copy voice, compliance constraints — packaged as a folder of Markdown the agent loads when relevant. Skills are how a Red Beacon-specific tone of voice survives a developer rotation without being re-explained every fortnight.
Agents & hooks
Sub-agents specialise the work (a reviewer agent, a screenshotter, an SEO auditor). Hooks fire on lifecycle events — for example, run a build and a Lighthouse pass on every PostToolUse save, so regressions surface inside the session rather than the next morning.
The four primitives — when to reach for which
| Primitive | Best for | Lives in | Red Beacon example |
|---|---|---|---|
| Custom command | Repeatable recipes the team runs by name | .claude/commands/*.md | /new-service-card scaffolds a USP card with brand tokens |
| MCP tool | Calling external systems (browser, DB, API) | .mcp.json | Playwright MCP fills and submits the lead form, asserts the thank-you redirect |
| Skill | Long-lived domain knowledge the agent should reload | .claude/skills/*/SKILL.md | brand-voice skill encodes Red Beacon's tone, banned words and disclosure copy |
| Sub-agent | Bounded, specialist work inside a larger task | .claude/agents/*.md | compliance-reviewer agent re-reads every published page against MAS disclosure rules |
| Hook | Automatic behaviour on lifecycle events | settings.json hooks block | PostToolUse hook runs the GitHub Pages build after each save |
How the day ran at Red Beacon
The cohort of 10 was a mix of investment and operations staff — not a developer-heavy room. By lunch every participant had a working Asset Invest landing page on their own GitHub Pages deployment, complete with hero, six USP cards, six service cards, testimonials and a validated lead-capture form posting to a confirmation page. In the afternoon we layered on the agentic mechanics.
- Custom commands. The team wrote
/new-sectionand/refresh-copyas Markdown files. Within an hour, adding a new service card became a single-line invocation rather than a 40-line edit. - MCP tools for testing. We attached the Playwright MCP server and asked Claude Code to drive the lead-generation form — fill name, email, message, submit, assert the thank-you page, screenshot the result. The agent caught a missing
requiredattribute on first run. - Skills for the website. Each participant authored a
brand-voiceskill capturing Red Beacon's tone — independent, evidence-led, no superlatives. Subsequent copy edits respected the constraints without being re-prompted. - Sub-agents. A
compliance-reviewersub-agent was set up to re-read pages against MAS disclosure expectations and flag missing risk language. - Hooks. A
PostToolUsehook ran the GitHub Pages build after each save, surfacing broken HTML in the same session it was introduced.
By end-of-day every staff member had shipped a working site to a public URL and had a personal .claude/ folder of commands, skills and hooks they could carry into the next project.
Why this format works for non-developer cohorts
The instinct with mixed cohorts is to keep the day shallow — a tool tour and a couple of toy prompts. That underestimates the room. Claude Code's command, skill and hook surfaces are text files; once a participant has authored one, they have authored them all. The skill transfer is conceptual (knowing when to reach for a command vs a skill vs a hook), not syntactic. We have run the same shape of workshop for developer-heavy rooms — see our earlier n8n workshop for 40 NTU staff — and the agentic mechanics translate cleanly between audiences.
If your team is starting from scratch, our AI solutions and corporate training service scopes the day around what your people actually ship — marketing pages, internal tools, compliance reviews — rather than a generic curriculum. Book a 30-minute walkthrough of the workshop format →
FAQ
Is the training only for software developers?
No. The Red Beacon cohort was investment and operations staff. The shape of the day is the same — every participant ships a working site — but the agentic exercises are pitched at the team's actual workflow, not a generic SaaS demo.
Is there a public, SSG-funded version?
Yes — the WSQ Agentic AI Applications with Claude Code course at Tertiary Courses Singapore is the public, SkillsFuture-fundable on-ramp covering the same material. The corporate version is private, customised to your repo and stack, and typically runs on-site for 8–15 staff.
What does a corporate Claude Code workshop cost?
Pricing depends on cohort size, on-site vs virtual, and whether you want our team to ship a starter repo before the day. Request a quote with your cohort size and target date →
Do we need to give the trainer access to our codebase?
Not for the standard day. Participants build on a scaffold we provide (the Asset Invest pattern, or one tailored to your industry). For an advanced engagement where the goal is to onboard Claude Code into your real repo, we sign an NDA and work inside your environment.
Where does Claude Code fit alongside other AI courses?
It is the agentic-coding pillar. The broader artificial intelligence courses catalogue at Tertiary Courses Singapore covers complementary tracks — machine learning, RAG, automation with n8n — that pair well with a Claude Code rollout.
What to do next
- Read the public curriculum. The WSQ Agentic AI Applications with Claude Code outline shows what your team would cover.
- Look at what staff actually ship. The Asset Invest repository is the artefact the Red Beacon team built — clone it, read it, picture your own team writing it.
- Talk to us. Book a scoping call for a corporate Claude Code workshop — we will come back with an agenda mapped to your team's real work.