Content
# Anthropic Architect Path
Personal upskilling journey aligned to the **Anthropic Skilljar canonical dev-track** progression. Tracks course completion, deep-dive notes, runnable sample builds, and a multi-cloud capstone — all in public so the journey itself becomes portfolio evidence.
## Owner
[@oianmol](https://github.com/oianmol) — started 2026-05-28
---
## Why this path
The fastest way to be hireable as an "AI/Claude solutions architect" is to:
1. Hold every Anthropic Skilljar dev-track credential (signals foundations).
2. Pass the **Claude Certified Architect — Foundations** exam (signals exam-grade rigor across the five official domains).
3. Ship at least one MCP server publicly (signals you grok the protocol, not just the docs).
4. Deploy a non-trivial multi-agent system on **both** Bedrock and Vertex (signals multi-cloud production discipline).
5. Have an eval harness backing your work (signals you take quality seriously).
This repo is the structured path to all five signals — the Skilljar courses, the supplementary depth Skilljar doesn't cover (prompt-engineering interactive tutorial, eval harness, computer use, engineering-blog reading, Agent SDK hooks, structured-output extraction, escalation/HITL, CI/CD headless, batch/files APIs, extended thinking), and the capstone. Everything builds toward one shipped artifact that exercises every Claude Certified Architect exam domain.
## Who this is for
Primarily: me. Secondarily: a public study guide for anyone walking the same Skilljar-to-architect ladder. Recruiters and hiring managers welcome.
## Prerequisites assumed
- Working knowledge of one programming language (Python or TypeScript ideal — most examples use Python)
- Comfort with `git`, the shell, and Docker basics
- An Anthropic API key (paid tier — caching benchmarks need actual cache hits)
- Familiarity with at least one of: AWS, GCP (needed in Phase 4)
If any of those are gaps, fill them before starting Phase 1.
---
## Skilljar Canonical Ladder (dev track)
Per [anthropic.skilljar.com](https://anthropic.skilljar.com), in order. **Progress: 5 / 10.**
| # | Course | Status | Credential | Phase |
|---|---|---|---|---|
| 2 | Claude Code 101 | ✓ | `46c4ruc82785` | (baseline) |
| 3 | Introduction to Claude Cowork | ☐ | — | [P1·m5](./phase-1-deepen-core/m5-claude-cowork/) |
| 4 | Claude Code in Action | ✓ | `ueoz6nmdn6d7` | (baseline) |
| 5 | Building with the Claude API | ✓ | `fdb463aapoa9` | (baseline) |
| 6 | Introduction to Model Context Protocol | ✓ | `ofh5hyqgjgqp` | (baseline) |
| 7 | Model Context Protocol: Advanced Topics | ☐ | — | [P2·m7](./phase-2-agents/m7-mcp-advanced/) |
| 8 | Introduction to agent skills | ✓ | `hjwkk9ige9y5` | (baseline) |
| 9 | Introduction to subagents | ☐ | — | [P2·m6](./phase-2-agents/m6-subagents-course/) |
| 10 | Claude with Amazon Bedrock | ☐ | — | [P4·m2](./phase-4-ship/m2-deploy-bedrock/) |
| 11 | Claude with Google Cloud's Vertex AI | ☐ | — | [P4·m3](./phase-4-ship/m3-deploy-vertex/) |
> Claude 101 (#1) skipped — foundational content covered by other completed courses.
>
> Optional non-dev tracks (AI Fluency variants, AI Capabilities and Limitations) not in scope.
---
## How to use this repo
1. **Pick** a Planned item (table below) and move it to In Progress.
2. **Read** the linked module README — it lists prerequisites, key concepts, exercises, and exit criteria.
3. **Study** the upstream content (course video, doc page, notebook, blog post).
4. **Build / annotate** — write personal notes in `<module>/notes/`, code in `<module>/lab/`.
5. **Verify** against the module's exit criteria checklist.
6. **Tick** the canonical-ladder table when a Skilljar course is done (record credential ID).
7. **Move** the row back to Planned (no — strike it out / archive) once complete.
### Study cadence
Target 8–12 hrs/week. A full Skilljar course takes 3–6 hrs. A lab module (build) typically takes 4–10 hrs. Eng-blog reads take 30–60 min each. The 12-week plan assumes ~10 hrs/week; scale accordingly.
---
## In Progress
_Nothing active yet. Pick next item from Planned below._
## Planned
Queue ordered by Skilljar canonical sequence, then supplemental priority.
| Item | Type | Source | Phase | Priority |
|---|---|---|---|---|
| Introduction to Claude Cowork | Course | Skilljar | 1 | H |
| Prompt Eng Interactive Tutorial (13 notebooks) | Notebook | `phase-1-deepen-core/m1-prompt-engineering/lab/` (submodule) | 1 | H |
| Tool Use Lab (weather / sql / parallel file search) | Lab | `phase-1-deepen-core/m3-tool-use/lab/` | 1 | H |
| Caching Bench (long-system scenario) | Lab | `phase-1-deepen-core/m4-prompt-caching/lab/` | 1 | H |
| Tool use docs (parallel, `tool_choice`, streaming) | Reading | [docs.claude.com/en/docs/build-with-claude/tool-use](https://docs.claude.com/en/docs/build-with-claude/tool-use) | 1 | H |
| Prompt caching docs (breakpoints, TTL, pricing) | Reading | [docs.claude.com/en/docs/build-with-claude/prompt-caching](https://docs.claude.com/en/docs/build-with-claude/prompt-caching) | 1 | M |
| Real World Prompting | Course | Anthropic Skilljar | 1 | M |
| Introduction to subagents | Course | Skilljar | 2 | H |
| Model Context Protocol: Advanced Topics | Course | Skilljar | 2 | H |
| Claude Agent SDK — full docs read | Reading | [docs.claude.com/en/api/agent-sdk](https://docs.claude.com/en/api/agent-sdk) | 2 | H |
| "Building effective agents" | Reading | [anthropic.com/engineering/building-effective-agents](https://www.anthropic.com/engineering/building-effective-agents) | 2 | H |
| "How we built Claude Code" | Reading | Anthropic engineering blog | 2 | H |
| "Multi-agent research system" | Reading | Anthropic engineering blog | 2 | H |
| MCP server build × 2 (stdio + HTTP) | Lab | `phase-2-agents/m2-mcp-stdio/lab/`, `m3-mcp-http/lab/` | 2 | H |
| MCP spec — full read | Reading | [modelcontextprotocol.io](https://modelcontextprotocol.io) | 2 | M |
| Agent SDK hooks (PreToolUse / PostToolUse policy enforcement) | Lab | `phase-2-agents/m8-hooks/lab/` | 2 | H |
| Claude Code session management (`fork_session`, `--resume`, `/compact`, `/memory`) | Lab | `phase-2-agents/m9-sessions/lab/` | 2 | M |
| Computer Use + Bash tool sandbox | Lab | `phase-2-agents/m4-computer-use/lab/` | 2 | M |
| Anthropic Cookbook — every notebook | Notebook | [anthropics/anthropic-cookbook](https://github.com/anthropics/anthropic-cookbook) | 1–3 | M |
| Structured-output extraction (JSON Schema + Pydantic validate-retry) | Lab | `phase-3-production/m7-structured-output/lab/` | 3 | H |
| Claude Code CI/CD headless (`-p`, `--output-format json`, `--json-schema`) | Lab | `phase-3-production/m8-ci-cd/lab/` | 3 | H |
| Escalation + Human-in-the-Loop (structured handoff protocols) | Lab | `phase-3-production/m9-escalation/lab/` | 3 | H |
| Provenance & multi-source synthesis (conflict handling, attribution) | Lab | `phase-3-production/m10-provenance/lab/` | 3 | M |
| Error propagation in multi-agent systems (transient/validation/business/permission) | Lab | `phase-3-production/m11-error-propagation/lab/` | 3 | H |
| Eval harness (LLM-as-judge + golden set) | Lab | `evals/` + `phase-3-production/m1-eval-harness/` | 3 | H |
| Batch API — bulk pipeline w/ `custom_id` correlation + partial-failure recovery | Lab | `phase-3-production/m2-batch-api/lab/` | 3 | M |
| Files API workflows | Reading + Lab | docs.claude.com + `phase-3-production/m3-files-api/lab/` | 3 | M |
| Extended thinking (interleaved w/ tools) | Reading + Lab | docs.claude.com + `phase-3-production/m4-extended-thinking/` | 3 | H |
| Context engineering (compaction, memory tool, sub-agent isolation) | Reading | Anthropic eng blog | 3 | H |
| Constitutional AI + Responsible Scaling Policy | Reading | anthropic.com | 3 | L |
| Claude Certified Architect — Foundations exam prep + sit | Exam | [exam guide](https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD) | 4 | H |
| Claude with Amazon Bedrock | Course | Skilljar | 4 | H |
| Claude with Google Cloud's Vertex AI | Course | Skilljar | 4 | H |
| Bedrock deployment guide | Reading | [docs.claude.com/en/api/claude-on-amazon-bedrock](https://docs.claude.com/en/api/claude-on-amazon-bedrock) | 4 | H |
| Vertex AI deployment guide | Reading | [docs.claude.com/en/api/claude-on-vertex-ai](https://docs.claude.com/en/api/claude-on-vertex-ai) | 4 | M |
| Capstone build (multi-agent + MCP + skills + subagents + evals) | Build | `phase-4-ship/capstone/` | 4 | H |
**Legend** — Type: Video / Reading / Notebook / Lab / Course / Build. Priority: H = do soon, M = mid, L = nice-to-have.
---
## Cloning
This repo references upstream Anthropic learning material via submodules. Clone with:
```bash
git clone --recursive https://github.com/oianmol/anthropic-architect-path.git
# or, if already cloned:
git submodule update --init --recursive
```
Pattern: upstream lives in a `lab/` submodule (untouched), personal notes/exercises live in sibling `notes/` directory.
---
## Active phase
**Phase 1: Deepen Core** — see [LEARNING_PATH.md](./LEARNING_PATH.md)
## Structure
```
.
├── LEARNING_PATH.md # Full 12-week plan, Skilljar-aligned, week-by-week
├── phase-1-deepen-core/ # m1 prompt-eng · m2 real-world · m3 tool-use · m4 caching · m5 cowork★
├── phase-2-agents/ # m1 agent-sdk · m2-3 mcp · m4 computer-use · m5 eng-blog · m6 subagents★ · m7 mcp-advanced★ · m8 hooks · m9 sessions
├── phase-3-production/ # m1 evals · m2 batch · m3 files · m4 thinking · m5 context-eng · m6 safety · m7 structured-output · m8 ci-cd · m9 escalation · m10 provenance · m11 error-propagation
├── phase-4-ship/ # m1 architecture · m2 bedrock★ · m3 vertex★ · capstone
├── projects/ # Standalone sample builds outside the phase ladder
├── evals/ # Eval harnesses + golden sets (cross-phase)
├── notes/ # Reading notes (blog, papers, cookbook)
└── resources/ # Links, prompts, snippets
★ = Skilljar canonical course
```
## Sample builds (runnable)
Standalone projects under [`projects/`](./projects/) — built to internalize a specific concept, runnable today.
| # | Project | Topic | Run |
|---|---|---|---|
| 01 | [`01-java-mcp-poetry-sampling`](./projects/01-java-mcp-poetry-sampling/) | MCP **sampling** primitive end-to-end (Kotlin server + Kotlin client; client bridges sampling to Anthropic Java SDK) | `cd projects/01-java-mcp-poetry-sampling && ./gradlew installDist && ANTHROPIC_API_KEY=… ./gradlew runClient -Ptext="city lights"` |
| 02 | [`02-mcp-progress-logging`](./projects/02-mcp-progress-logging/) | MCP **logging + progress** notifications (Kotlin server + Kotlin client; multi-step sampling tool emits progress and log frames per stanza) | `cd projects/02-mcp-progress-logging && ./gradlew installDist && ANTHROPIC_API_KEY=… ./gradlew runClient -Ptheme="monsoon rain" -Pcount=4` |
See each project's README for prerequisites and the Inspector-based run path.
---
Each module follows the same shape:
```
mN-<slug>/
├── README.md # why-it-matters + key concepts + exercises + exit criteria
├── notes/ # personal writeups, pattern cards
└── lab/ # runnable code (when applicable)
```
---
## What "done" looks like (architect signal at end)
- [ ] 10/10 Skilljar dev-track credentials captured here
- [ ] Claude Certified Architect — Foundations exam passed (score ≥ 720); badge linked
- [ ] At least one MCP server published (npm or registry)
- [ ] One multi-agent capstone deployed on **both** Bedrock and Vertex
- [ ] Eval harness running against the capstone with a published cost + quality report
- [ ] Architecture diagram + ADRs (architecture decision records) in `phase-4-ship/capstone/architecture/`
- [ ] Demo video linked from capstone README
When all seven items are ticked, the repo itself is the credential — a recruiter can verify every claim by clicking through.
