A privacy-first Claude skill that turns your profile into a structured study abroad + immigration pathway — school shortlist, EOI/visa pathway, cost breakdown, and a concrete next-steps list.
You give it 6 things about yourself. It gives you a multi-year plan.
Input (collected via conversation, never stored):
| # | Field | Example |
|---|---|---|
| 1 | Age at graduation | 27 |
| 2 | English test + score | PTE 65 (60, 65, 60, 65) |
| 3 | Highest education + GPA | Bachelor in Engineering, GPA 3.5/4.0 |
| 4 | Total budget | 500000 RMB |
| 5 | Target country + field | Australia, Master of OT |
| 6 | Constraints (optional) | Prefer regional area for PR points |
Output: a 6-section markdown report
- Profile snapshot
- Shortlist (3-5 schools, sorted by fit, not prestige)
- Visa pathway (EOI / points / state nomination / timeline)
- Cost & cashflow (with budget gap called out explicitly)
- Risk register (3-5 risks, each with mitigation)
- Recommended next actions (this week, with verifiable outcomes)
v0.3.0 had a single 180-day freshness threshold for all data. v0.3.1 fixes that with a per-data-type policy table + override mechanism + gate wrapper:
| Tool | Use it for |
|---|---|
bin/release-precheck.mjs |
Pre-push orchestrator: PII scrub + source-validator in one call (100ms) |
bin/synthesise.mjs |
The gate every LLM must call before writing a report. Reads release-precheck exit code, decides proceed/caveat/block, writes audit log |
data/freshness-policy.json |
9 data_type thresholds (e.g. eoi_threshold 7d warn / 30d error; english_standard 30d / 60d) |
data/known-stale.json |
Official override mechanism — mark data as "best-available, intentionally not refreshed" with audit trail |
data/source-tiers.json |
Now includes domain_tiers block (45 domains mapped; was hardcoded in .mjs) |
Why it matters: when AHPRA ELS / DHA SkillSelect / state nomination rules change (which happens yearly), the v0.3.0 validator wouldn't catch it until 180 days passed. v0.3.1 catches the same change in 7-30 days automatically — no code change. The override mechanism also means known-stale historical data (e.g. SkillSelect round archive) doesn't block pushes.
v0.2.0 shipped CLI tools. v0.3.0 adds a 5-step honesty pipeline so the report says what it doesn't know, not what looks plausible:
| Stage | What it does |
|---|---|
| validate | bin/source-validator.mjs scans all data/*.json for source tier (1-5, social media banned) + freshness periods + warning markers |
| gate | Hard eligibility gate (e.g. AHPRA ELS 3 English gates, PTE/Age/Prereq) — binding constraints called out explicitly |
| analyse | Tier-based PR pathway comparison (高/中/低) — no fabricated precision percentages |
| synthesise | 6-section report grounded in the same data the validator just approved |
| self-check | 5-item honesty checklist before output: source tier, freshness, confidence stated, fabricated numbers, mismatch with audit |
New in v0.3.0:
| File | Purpose |
|---|---|
bin/source-validator.mjs |
5-tier source classifier + freshness periods + tier_5 (Reddit/Facebook) ban |
data/source-tiers.json |
Domain → tier mapping (tier_1: gov/university, tier_4: aggregator, tier_5: social media) |
data/ahpra-english-standard.json |
3 English gates (CDU entry 65/58 · AHPRA ELS 65/65 · EOI superior 79/79) + 4 ELS exemption pathways |
data/seven-school-pr-pathway-comparison.json |
7 OT schools ranked by PR pathway, with explicit "real grant-by-school data not public" note |
data/cdu-ot-fieldwork-compliance.json |
1000h placement breakdown + 6-item compliance checklist (police / WWCC / immunisation / NDIS screening) |
data/australian-courses.json |
Now 11 courses (Monash added) + La Trobe Bendigo → Bundoora correction + IELTS separate fields + PTE Academic test type verified |
Why it matters: v0.2.0 reported "85% PR conversion" for the top school. v0.3.0 reports "高" with confidence + basis + the explicit caveat that AHPRA does not publish school-of-graduation data. Same conclusion, no false precision.
v0.1 was static markdown docs. v0.2.0 adds a CLI toolset and a queryable JSON dataset:
| Tool | Use it for |
|---|---|
bin/eoi-calculator.mjs |
EOI points breakdown by age / English / qualification / state nomination — 10 seconds, no mental math |
bin/aus-course-filter.mjs |
Filter 10 courses by PTE / budget / regional / profession / path / prerequisites — 1 second, no grep |
data/australian-courses.json |
10 courses × 12 fields (CDU / La Trobe / UTAS / Flinders / UniSA / CQU / UniMelb / USyd) — queryable, version-controlled |
Why it matters: a privacy-first skill that only outputs markdown would still need the user (or agent) to manually re-derive EOI math, or grep through dozens of web pages. v0.2.0 ships the data + tools so the 6-section report is grounded in the same source the user can audit.
Live data sources (instead of hand-quoted values that go stale):
- MLTSSL occupation list → https://www.anzscosearch.com/mltssl/
- QILT course + employment → https://qilt.edu.au/comparED
- Skilled occupation list 2026 → https://www.australianvisaonline.com/occupations/skilled-occupation-list-sol-2026
- Home Affairs points calculator → https://immi.homeaffairs.gov.au/help-support/tools/points-calculator
The skill's references/australia.md and references/data-sources.md now point at these live sources for any value that changes annually.
This skill contains zero personal data and stores nothing. It is safe to publish to a public repository.
- ❌ Never bakes a real name, age, score, university, or salary into the skill
- ❌ Never reads
user-profile.jsonfrom disk unless you explicitly pass a path - ❌ Never echoes your data into skill logs or telemetry
- ✅ Every example uses
<placeholder>syntax — copy-paste safe - ✅ The skill always collects your profile fresh at invocation time
If you generate a report for a real person, save it outside the skill directory and never commit that report alongside the skill.
Drop the skill folder into your Claude skills directory:
# Clone the repo
git clone https://github.com/<owner>/study-abroad-planner.git
# Or copy manually
cp -r study-abroad-planner ~/.claude/skills/study-abroad-planner
# Verify it shows up in your skills list
ls ~/.claude/skills/study-abroad-planner/
# SKILL.md LICENSE README.md .gitignore references/ examples/Restart Claude Code. The skill is now active and will auto-trigger on keywords like 留学, study abroad, EOI, 189, PR pathway, etc.
Upload SKILL.md and the references/ folder as a project skill. The
exact upload UI varies; refer to Anthropic's current skill documentation.
The skill is pure markdown + reference files. Any agent that supports the
SKILL.md-with-frontmatter convention can use it. Point the agent at the
SKILL.md file.
Just ask Claude anything that sounds like study-abroad planning. The skill fires automatically.
Triggers (English):
"Help me plan a master's in Australia" "What's my EOI score for 189?" "I want to study OT in the UK" "Compare universities for nursing in New Zealand" "PR pathway for software engineers in Canada"
Triggers (Chinese):
"帮我规划一下澳洲留学" "189 签证 EOI 怎么打分" "我想去英国读 OT 硕士" "新西兰护理专业选哪所大学" "加拿大软件工程师 PR 路径"
Trigger-then-flow:
The skill first asks you the 6 questions. You answer. It confirms the profile in one short table. You confirm or correct. It then generates the full report.
| Country | Status | Reference |
|---|---|---|
| Australia 🇦🇺 | ✅ Full (189 / 190 / 491 + EOI + QILT) | references/australia.md |
| United Kingdom 🇬🇧 | 🟡 Stub (Graduate Route + Skilled Worker) | references/countries.md |
| New Zealand 🇳🇿 | 🟡 Stub (SMC + Green List) | references/countries.md |
| Canada 🇨🇦 | 🟡 Stub (Express Entry + PNP) | references/countries.md |
| Others | ❌ Not covered | — |
To add a new country: drop a references/<country>.md that follows the
same structure as the existing references (visa subclasses, points table,
occupation lists, skills assessment bodies, recommended reading order).
Open a PR.
study-abroad-planner/
├── SKILL.md # Main entry — frontmatter + pipeline + output template
├── README.md # This file
├── LICENSE # MIT
├── .gitignore # Excludes eval workspaces + any user-data files
├── bin/
│ ├── eoi-calculator.mjs # EOI points breakdown CLI
│ ├── aus-course-filter.mjs # Course filter by PTE/budget/regional/prereq
│ ├── source-validator.mjs # 5-tier source classifier + freshness-by-policy check
│ ├── release-precheck.mjs # v0.3.1 — orchestrator: PII + validator in one call
│ └── synthesise.mjs # v0.3.1 — gate wrapper (LLM must call before writing report)
├── data/
│ ├── australian-courses.json # 11 courses × ~15 fields
│ ├── source-tiers.json # 5-tier definitions + 45 domain mappings
│ ├── freshness-policy.json # v0.3.1 — 9 data_type thresholds
│ ├── known-stale.json # v0.3.1 — official override mechanism
│ ├── ahpra-english-standard.json # 3 gates + 4 ELS exemptions
│ ├── ahpra-ot-registration-pathway.json # OTBA Pathway A + OTC migration
│ ├── seven-school-pr-pathway-comparison.json # tiered honest format
│ ├── cdu-ot-fieldwork-compliance.json # 1000h + 6-item checklist
│ └── skillselect-ot-invitations.json # Historical SkillSelect data
├── references/
│ ├── australia.md # Full AU 189/190/491 + EOI + QILT
│ ├── countries.md # UK / NZ / CA short pathways
│ └── data-sources.md # Authoritative public links per country
└── examples/
└── sample-output.md # Sample report with placeholder values
The skill ships with one worked example in examples/sample-output.md. To
add more coverage, add new test profiles (all <placeholder> values) to
the same folder and run them through Claude with the skill loaded.
# Edit SKILL.md
$EDITOR ~/.claude/skills/study-abroad-planner/SKILL.md
# Restart Claude Code to pick up the change
# (skills are loaded at session start)
# Run a test invocation
claude "Help me plan study in Australia"For full eval/benchmark workflows (assertion-based testing, blind
comparisons, description optimization), see Anthropic's
skill-creator skill.
PRs welcome. Two specific ways to help:
- Add a country reference — write
references/<country>.mdfollowing the same structure asreferences/australia.md. Include: visa subclasses, points table, occupation lists, skills assessment bodies, recommended reading order. - Improve the output template — the 6-section template is in
SKILL.md. Make the risk register sharper, or the next-actions list more verifiable, etc.
Run all four before git push:
# 1. PII grep — replace patterns with ones matching your own PII
grep -rE "<suspicious-pattern>" \
~/.claude/skills/study-abroad-planner/ || echo "OK no PII"
# 2. Placeholder audit — every example value must be <wrapped>
grep -hroE "[^<a-z_<>][a-z_]+[0-9]?[^a-z_<>]*" \
~/.claude/skills/study-abroad-planner/examples/ \
| grep -v "^<" || echo "OK all placeholders"
# 3. Frontmatter check
head -3 ~/.claude/skills/study-abroad-planner/SKILL.md
# 4. No fabricated URLs
grep -rhE "https?://[^ )]+" ~/.claude/skills/study-abroad-planner/ | wc -l- It does not file visa applications, draft statements, or contact universities on your behalf.
- It does not guarantee admission or visa grant.
- It does not replace a licensed migration agent. If your case is non-trivial (e.g. dependent family, prior refusals, complex work history), consult a MARA-registered agent (AU) / OISC-regulated adviser (UK) / ICCRC consultant (CA) for the legal step. Planning is free; the legal step is not.
MIT © 2026 study-abroad-planner contributors
- UK / NZ / CA references expanded to full (currently stubs)
- Multi-language UI for the conversation flow (currently EN + ZH keywords only)
- Optional
user-profile.jsonsupport with explicit consent - Eval suite with 3+ realistic test profiles
- GitHub Action that runs the preflight checks on every PR
v0.3.0 所有数据共用 180 天保鲜期，v0.3.1 改成按 data_type 分级 + override + 守门 wrapper：
| 工具 | 用途 |
|---|---|
bin/release-precheck.mjs |
发布前编排：PII 扫描 + 源校验一键（100ms） |
bin/synthesise.mjs |
LLM 写报告前必调的守门。读 release-precheck 退出码，决定 proceed/caveat/block，写审计日志 |
data/freshness-policy.json |
9 个 data_type 阈值（如 eoi_threshold 7d warn / 30d error；english_standard 30d / 60d） |
data/known-stale.json |
官方 override 机制 — 标"明知过期但用这个"并留审计痕迹 |
data/source-tiers.json |
新增 domain_tiers 块（45 个域名映射；之前硬编码在 .mjs） |
为什么重要：当 AHPRA ELS / DHA SkillSelect / 州担保规则变更（每年都会发生），v0.3.0 validator 要 180 天才察觉。v0.3.1 7-30 天自动捕获——不用改代码。Override 机制也让已知过期的历史数据（如 SkillSelect 归档轮次）不再 block push。
v0.2.0 出了 CLI 工具。v0.3.0 加了 5 步诚实 pipeline，让报告说"不知道"而不是"看着像承诺"：
| 步骤 | 做什么 |
|---|---|
| validate | bin/source-validator.mjs 扫所有 data/*.json 检查源 tier (1-5，社媒禁) + 保鲜期 + 警告标记 |
| gate | 硬性资格门槛（如 AHPRA ELS 3 道英语门）— 显式标出 binding constraint |
| analyse | 分级 PR 路径对比 (高/中/低) — 不编造精确百分比 |
| synthesise | 6 段报告，用刚才 validator 批准的数据 |
| self-check | 5 项诚实清单：源 tier / 保鲜期 / 信心度 / 假数据 / 与审计不一致 |
v0.3.0 新增：
bin/source-validator.mjs— 5 级源分类 + 保鲜期 + tier_5 社媒禁入data/source-tiers.json— domain → tier 映射（tier_1: gov/uni，tier_5: 社媒）data/ahpra-english-standard.json— 3 道英语门（CDU 65/58 · ELS 65/65 · EOI superior 79/79）+ 4 豁免路径data/seven-school-pr-pathway-comparison.json— 7 校 PR 路径分级 + 显式"真实 grant-by-school 数据不公开"标注data/cdu-ot-fieldwork-compliance.json— 1000h 实习拆解 + 6 项合规清单data/australian-courses.json— 11 校（新增 Monash） + La Trobe Bendigo→Bundoora 修正 + IELTS 字段 + PTE Academic 验证
给它 6 个关于你的信息，它会给你一个多年计划。
输入（通过对话收集，永不存储）：
| # | 字段 | 示例 |
|---|---|---|
| 1 | 毕业时年龄 | 27 |
| 2 | 英语考试 + 分数 | PTE 65 (60, 65, 60, 65) |
| 3 | 最高学历 + GPA | Bachelor in Engineering, GPA 3.5/4.0 |
| 4 | 总预算 | 500000 RMB |
| 5 | 目标国家 + 专业 | Australia, Master of OT |
| 6 | 约束（可选） | 倾向偏远地区加分 |
输出：6 段 markdown 报告
- Profile 速览
- 候选学校（3-5 所，按适配度而非排名排序）
- 签证路径（EOI / 打分 / 州担保 / 时间线）
- 成本 & 现金流（明确指出预算缺口）
- 风险登记册（3-5 项，每项有缓解措施）
- 这周可执行的下一步（每项有可验证的产出）
这个 skill 包含 0 真实个人信息，不存任何东西。 可以放心发到公开仓库。
- ❌ 绝不把真实姓名、年龄、分数、学校、薪资写进 skill
- ❌ 绝不读
user-profile.json，除非你显式传路径 - ❌ 绝不在 skill 日志或 telemetry 里 echo 你的数据
- ✅ 所有示例都用
<placeholder>语法 - ✅ 每次调用都现场收集你的 profile
如果你为真人生成报告，保存到 skill 目录外，绝不要 commit 这份报告。
把 skill 文件夹扔到你的 skills 目录：
git clone https://github.com/<owner>/study-abroad-planner.git
cp -r study-abroad-planner ~/.claude/skills/study-abroad-planner
ls ~/.claude/skills/study-abroad-planner/重启 Claude Code。Skill 就激活了，关键词自动触发。
直接问 Claude 任何听起来像"留学规划"的问题，skill 自动触发。
触发词（中文）：
"帮我规划澳洲留学" "189 签证 EOI 怎么打分" "我想去英国读 OT 硕士" "新西兰护理专业选哪所" "加拿大软件工程师 PR 路径"
触发流程：
skill 先问你 6 个问题。你回答。它用 1 个简短表格确认。你确认或更正。然后生成完整报告。
study-abroad-planner/
├── SKILL.md # 主入口 — frontmatter + pipeline + 输出模板
├── README.md # 这个文件
├── LICENSE # MIT
├── .gitignore
├── bin/
│ ├── eoi-calculator.mjs # EOI 打分 CLI
│ ├── aus-course-filter.mjs # 按 PTE/预算/偏远/前置 筛选课程
│ ├── source-validator.mjs # 5 级源分类 + 按政策检查保鲜期
│ ├── release-precheck.mjs # v0.3.1 — 编排器：PII + validator 一键跑
│ └── synthesise.mjs # v0.3.1 — 守门 wrapper（LLM 写报告前必调）
├── data/
│ ├── australian-courses.json # 11 校 × 15 字段
│ ├── source-tiers.json # 5 级定义 + 45 域名映射
│ ├── freshness-policy.json # v0.3.1 — 9 个 data_type 阈值
│ ├── known-stale.json # v0.3.1 — 官方 override 机制
│ ├── ahpra-english-standard.json # 3 道英语门 + 4 豁免路径
│ ├── ahpra-ot-registration-pathway.json # OTBA Pathway A + OTC 移民评估
│ ├── seven-school-pr-pathway-comparison.json # 分级诚实格式
│ ├── cdu-ot-fieldwork-compliance.json # 1000h + 6 项合规清单
│ └── skillselect-ot-invitations.json # SkillSelect 历史数据
├── references/
│ ├── australia.md # 完整 AU 189/190/491 + EOI + QILT
│ ├── countries.md # UK / NZ / CA 短路径
│ └── data-sources.md # 各国公开权威源
└── examples/
└── sample-output.md # 占位符示例报告
- 不代办签证申请、文书、联系大学
- 不保证录取或签证获批
- 不替代持牌移民律师。复杂案例（家庭随行、有拒签史、复杂工作史）请咨询 MARA（澳洲）/ OISC（英国）/ ICCRC（加拿大）持牌顾问。规划免费，法律那一步不免费。
MIT © 2026 study-abroad-planner contributors