CV Builder — Create CVs via API
Create a resume, digital identity, or CV right now — for free. Choose from templates that span the full range: classic print-ready PDFs, clean HTML online CVs, interactive experiences, animated 3D worlds, and playable games. Every CV gets a permanent public URL with a personal slug — no signup, no payment needed for basic use. ATS-ready and Agent-Ready (HR).
Feature Overview
| Feature | Access-ID needed | How |
|---|---|---|
| Create CV (online, permanent URL) | No | prefer_hitl: true or skip_hitl: true |
| PDF Export | No | "include_pdf": true on any request |
| Human slug (URL) selection | No | HITL step 3 |
| Human template selection | No | HITL step 4 |
| Inline submit (no browser) | No | Bearer token on confirm/approval steps |
| Callback webhook | Yes | "hitl_callback_url" |
| Higher rate limits (50/day) | Yes | Register via POST /api/agent/register |
| Custom templates | Yes | Template Create skill |
Templates span: classic PDF · HTML online · interactive · 3D (Three.js) · gamified. Full catalog: reference/templates.md — live previews: talent.de/de/cv-template-ideas
Terminology
Throughout this skill, these terms are used consistently:
| Term | Meaning |
|---|---|
| requestor | The human who commissioned this CV build — the person providing data, making decisions (slug, template, approval), and receiving the claim token. All agent actions are on their behalf. |
| human | Synonym for requestor in HITL context, used to contrast with automated/AI steps (e.g., "if the human prefers the browser"). |
| agent | You — the AI executing this skill. |
Agent Guidelines
HITL is required. You MUST include either
"prefer_hitl": true(human review) or"skip_hitl": true(direct creation). Omitting both returns a 400 error. If a human is present, ALWAYS use"prefer_hitl": true— this lets the requestor choose their URL slug, pick a template, review the data, and approve before publishing. Use"skip_hitl": trueonly for automated pipelines with no human in the loop.
Data principle: Only use data the requestor has explicitly provided or approved in this conversation. Do not extract personal information from unrelated system contexts or other sessions.
Before sending: Present a brief summary to the requestor — name, title, email — and ask "Send it? Or should I change anything?"
Claim token: Treat like a password. Share only with the requestor — anyone with the token can claim CV ownership. Never share with third parties.
Credentials
An Access-ID (talent_agent_[a-z0-9]{4}) is optional for CV Builder — basic use (3 CVs/day per IP) works without one. Register for higher limits (50 CVs/day) and callback webhook support:
POST https://www.talent.de/api/agent/register
Content-Type: application/json
{ "agent_name": "my-agent" }
The Access-ID is also the HMAC secret for verifying X-HITL-Signature on callback webhooks. Store in TALENT_ACCESS_ID — do not hardcode.
User Communication
What to say at each step
| Step | Say to the requestor |
|---|---|
| Before API call | "Let me set up your CV. I just need a few details." |
| Slug selection (review_url received) | "Choose your personal URL — this is where your CV will live: [link]" |
| Template selection | "Almost done! Pick a design for your CV: [link]" |
| Approval | "Your CV is ready for review. Take a look and approve it: [link]" |
| After final 201 | "Your CV is live! Here's your link: {url}" |
Quick Start
- Ask for (or confirm you already have): firstName, lastName, title, email — the 4 required fields
POST /api/agent/cv-simplewith"prefer_hitl": trueand the data Optional: add"include_pdf": trueto also receive a base64 PDF in the final 201 response. See PDF Export.- Present the
review_urlto the requestor (they pick slug, template, review data) - Poll
poll_urlevery 30s until"status": "completed":{ "status": "pending" }or{ "status": "opened" }→ keep polling{ "status": "completed", "result": { "action": "confirm", "data": {...} } }→ advance withhitl_continue_case_id
- After final approval: present the live CV URL and claim token
Example (with HITL — recommended)
POST https://www.talent.de/api/agent/cv-simple
Content-Type: application/json
{
"prefer_hitl": true,
"cv_data": {
"firstName": "Alex",
"lastName": "Johnson",
"title": "Software Engineer",
"email": "alex@example.com",
"experience": [{
"jobTitle": "Senior Developer",
"company": "Acme Inc.",
"startDate": "2022-01",
"isCurrent": true
}],
"hardSkills": [{ "name": "React", "level": 4 }],
"softSkills": [{ "name": "Team Leadership" }],
"languages": [{ "name": "English", "level": "NATIVE" }]
}
}
Response (202 — human review required):
{
"status": "human_input_required",
"message": "Please confirm: is this CV for you?",
"hitl": {
"case_id": "review_a7f3b2c8d9e1f0g4",
"review_url": "https://www.talent.de/en/hitl/review/review_a7f3b2c8d9e1f0g4?token=abc123...",
"poll_url": "https://www.talent.de/api/hitl/cases/review_a7f3b2c8d9e1f0g4/status",
"type": "confirmation",
"inline_actions": ["confirm", "cancel"],
"timeout": "24h"
}
}
Present the review URL to the requestor:
I've prepared your CV. Please review and make your choices here: Review your CV You'll pick your personal URL slug, template design, and approve the final result.
Then poll poll_url until completed, and continue through steps with hitl_continue_case_id. After all steps (confirmation, data review, slug selection, template selection, approval), the final POST returns 201 with the live URL.
Full HITL protocol with all steps, inline submit, edit cycles, and escalation: reference/hitl.md
HITL Multi-Step Flow
The requestor goes through up to 5 review steps. The agent loops: present review URL, poll, continue.
Step 1: Confirmation → "For whom is this CV?"
Step 2: Data Review → "Are these details correct?"
Step 3: Slug → Human picks personal URL slug (e.g. pro, dev, 007)
Step 4: Template → Human picks template design
Step 5: Approval → Human reviews final CV draft
Each step returns 202. After the requestor decides, continue:
POST https://www.talent.de/api/agent/cv-simple
Content-Type: application/json
{
"prefer_hitl": true,
"hitl_continue_case_id": "review_a7f3b2c8d9e1f0g4",
"slug": "dev",
"cv_data": { ... }
}
Important:
slugandtemplate_idgo at the top level of the request, not insidecv_data. When continuing after slug selection, include the human's chosen slug at the top level so the server knows to advance to the template step.
Steps are skipped when you already provide the value:
- Include
slug(top-level) → slug selection step is skipped - Include
template_id(top-level) → template selection step is skipped - Include both → only confirmation, data review, and approval remain
Inline Submit (v0.7)
For simple decisions (confirmation, escalation, approval), the 202 response includes submit_url, submit_token, and inline_actions. Agents can submit directly via Bearer token — ideal for Telegram, Slack, WhatsApp where buttons are supported:
POST {submit_url}
Authorization: Bearer {submit_token}
Content-Type: application/json
{ "action": "confirm", "data": {} }
Always present
review_urlas a fallback alongside any inline buttons. If the platform does not support buttons (SMS, email, plain text), or the human prefers the browser, they can use the link to complete their decision.
Selection and input types always require the browser (review_url) — they involve complex UI (template grid, data forms). Full inline spec: reference/hitl.md
After the final approval step, submit with hitl_approved_case_id to publish:
POST https://www.talent.de/api/agent/cv-simple
Content-Type: application/json
{
"hitl_approved_case_id": "review_final_case_id"
}
Response (201):
{
"success": true,
"url": "https://www.talent.de/dev/alex-johnson",
"cv_id": "cv_abc123",
"claim_token": "claim_xyz789",
"template_id": "007",
"quality_score": 65,
"quality_label": "good",
"improvement_suggestions": []
}
⚠️ Immediately after receiving a 201 — always share both with the requestor:
- CV URL:
https://www.talent.de/dev/alex-johnson— their live profile- Claim link:
https://www.talent.de/claim/claim_xyz789— to take ownership and editThe
claim_tokenis permanent and never expires. Treat it like a password — share only with the requestor.
Present the result:
Your CV is live: talent.de/dev/alex-johnson
To claim ownership and edit your CV, visit: talent.de/claim/claim_xyz789 Keep this link safe — it never expires and gives full edit access.
If the response includes improvement_suggestions, share them with the requestor and offer to update the CV:
Your CV score is 35/100. To improve it, I can add: work experience (+25pts), professional summary (+20pts). Want me to ask you a few questions and update it?
Agent Loop (Visual)
Both paths — with and without human review — are shown below. Choose ONE per request.
flowchart TD
START([Agent starts]) --> CHOICE{prefer_hitl\nor skip_hitl?}
%% ── skip_hitl path ──────────────────────────────────────────
CHOICE -->|skip_hitl: true| DIRECT["POST /api/agent/cv-simple\nskip_hitl: true + cv_data"]
DIRECT --> D201["201 — CV live\nurl · claim_token · quality_score\nimprovement_suggestions"]
D201 --> SHARE_NOW["Share url + claim_token\nwith requestor immediately!"]
SHARE_NOW --> QUAL{improvement_suggestions\npresent AND attempt < 2?}
QUAL -->|No / attempt >= 2| DONE_DIRECT([Done])
QUAL -->|Yes| ASK["Ask requestor the questions\nin each agent_action field"]
ASK --> REPOST["POST /api/agent/cv-simple\nskip_hitl: true\nenriched cv_data\n(new cv_id each time)"]
REPOST --> D201
%% ── prefer_hitl path ─────────────────────────────────────────
CHOICE -->|prefer_hitl: true| HITL["POST /api/agent/cv-simple\nprefer_hitl: true + cv_data"]
HITL --> H202["202 human_input_required\nreview_url · poll_url · events_url"]
H202 --> SHOW_URL["Present review_url to requestor\n'Please review here: [link]'"]
SHOW_URL --> POLL["Poll poll_url every 30s\n(or use events_url for SSE)"]
POLL --> STATUS{status?}
STATUS -->|pending / opened\n/ in_progress| POLL
STATUS -->|completed| ACTION{result.action?}
STATUS -->|expired| EXP{default_action?}
STATUS -->|cancelled| CANCELLED([Inform requestor: cancelled])
EXP -->|skip| AUTO_PUB["CV auto-published\nurl from poll status"]
EXP -->|abort| ABORTED([Inform requestor: expired])
ACTION -->|confirm / select| CONTINUE["POST cv-simple\nhitl_continue_case_id\n+ ALWAYS include cv_data"]
ACTION -->|edit| EDIT["Adjust cv_data per note\nthen CONTINUE"]
EDIT --> CONTINUE
ACTION -->|reject| REJECT["Escalation step\nPOST hitl_continue_case_id"]
REJECT --> H202
ACTION -->|approve| PUBLISH["POST cv-simple\nhitl_approved_case_id + cv_data"]
CONTINUE --> H202
PUBLISH --> DONE_HITL(["201 — CV live\nShare url + claim_token"])
When a HITL case expires, the server applies the default_action configured for that step:
skip: CV is auto-published with server-selected slug/template. Pollpoll_url— it returnsstatus: "completed"withurlin the result.abort: Flow terminates. Inform the requestor the session expired. Start a new HITL flow withprefer_hitl: trueif needed.
Quality Improvement Loop (skip_hitl)
After any 201 response the CV is immediately live. Always share url and claim_token with the requestor first.
If improvement_suggestions is non-empty, you may optionally run up to 2 improvement cycles:
- For each suggestion: ask the requestor the exact question in
agent_action - Re-POST with enriched
cv_data+skip_hitl: true - Each re-POST creates a new CV with a new
cv_id— the previous one is abandoned - After 2 cycles OR when
improvement_suggestionsis empty: stop
Do not loop indefinitely. An agent that keeps re-posting creates duplicate CVs and wastes the requestor's time.
For prefer_hitl flows: after approval, the 201 is already quality-assessed (human reviewed the data). If suggestions appear, ask the requestor directly — do not restart the HITL flow.
Direct Creation (No Human Present)
For fully automated pipelines, batch operations, or when the requestor explicitly says "just create it" — set "skip_hitl": true:
POST https://www.talent.de/api/agent/cv-simple
Content-Type: application/json
{
"skip_hitl": true,
"cv_data": {
"firstName": "Alex",
"lastName": "Johnson",
"title": "Software Engineer",
"email": "alex@example.com"
}
}
Response (201):
{
"success": true,
"url": "https://www.talent.de/pro/alex-johnson",
"cv_id": "cv_abc123",
"claim_token": "claim_xyz789",
"template_id": "018",
"hitl_skipped": true,
"quality_score": 20,
"quality_label": "basic",
"improvement_suggestions": [
{
"field": "experience",
"issue": "No work experience — CV has low ATS compatibility",
"agent_action": "Ask: 'What positions have you held? Part-time and internships count.'",
"impact": "+25 quality points",
"priority": "high"
}
],
"next_steps": "Share improvement_suggestions with the requestor and ask the questions in agent_action. Then update the CV via POST /api/agent/cv-simple...",
"auto_fixes": []
}
In direct mode, the server auto-assigns slug (pro by default) and template (018 Amber Horizon). The requestor has no choice. Use this only when no human needs to review.
You MUST choose one: "prefer_hitl": true or "skip_hitl": true. Omitting both returns a 400 error.
All fields beyond the 4 required ones are optional. Omit what you don't have — don't send empty arrays.
PDF Export
Get a downloadable PDF alongside the CV. Three visual themes available:
| Theme | Style | Best for |
|---|---|---|
classic | Single-column, red accent (default) | Traditional industries |
modern | Two-column sidebar, blue accent | Tech & creative roles |
minimal | Monochrome, generous whitespace | Executive & senior roles |
Option A: Inline with CV creation
Add "include_pdf": true to your request. The response includes a base64-encoded PDF:
POST https://www.talent.de/api/agent/cv-simple
Content-Type: application/json
{
"prefer_hitl": true,
"include_pdf": true,
"pdf_format": "A4",
"pdf_theme": "modern",
"cv_data": {
"firstName": "Alex",
"lastName": "Johnson",
"title": "Software Engineer",
"email": "alex@example.com"
}
}
Response includes a pdf object:
{
"success": true,
"url": "https://www.talent.de/pro/alex-johnson",
"cv_id": "cv_abc123",
"claim_token": "claim_xyz789",
"pdf": {
"base64": "JVBERi0xLjQK...",
"size_bytes": 6559,
"generation_ms": 226,
"format": "A4"
}
}
Option B: Generate PDF for existing CV
POST https://www.talent.de/api/agent/cv/pdf
Content-Type: application/json
{
"cv_id": "cv_abc123",
"format": "A4",
"theme": "minimal"
}
Returns the PDF binary directly (Content-Type: application/pdf). Format options: A4 (default), LETTER. Theme options: classic (default), modern, minimal.
PDF generation takes ~200ms (no headless browser needed).
Server-Side Intelligence
You do not need to check slug availability or validate data — the server handles it:
- Slug uniqueness: A slug is NOT globally unique — it's unique per person.
pro/thomas-muellerandpro/anna-schmidtcan coexist. Only the combination of slug + firstName + lastName is reserved. That's why the server needs the name first to check availability. - Slug auto-selection: If omitted (and no HITL), the server picks
pro. If that slug is already in use for this person's name, it tries the next available slug automatically. In HITL mode, the human can choose their slug interactively. Popular picks (excerpt):007·911·dev·api·pro·gpt·web·ceo·cto·ops·f40·gtr·amg·gt3·zen·art·lol·neo·404·777. Full list:GET /api/public/slugs - Template default:
018(Amber Horizon) if omitted. - Date normalization:
2024becomes2024-01-01,2024-03becomes2024-03-01. - Language levels: Normalized to CEFR (
NATIVE,C2,C1,B2,B1,A2,A1). - Human-readable errors: If something goes wrong, the response explains what to fix in plain English.
- Auto-fix summary: The
auto_fixesarray tells you what the server adjusted (e.g. "Slug 'pro' is already in use for this name, using 'dev' instead").
Skills Use 4 Arrays
hardSkills— technical skills, optionallevel1-5softSkills— name onlytoolSkills— name onlylanguages— with CEFRlevel:NATIVE,C2,C1,B2,B1,A2,A1
Do not use a generic skills array — it will be ignored.
Common Mistakes
| Wrong | Correct | Why |
|---|---|---|
"role": "Engineer" | "jobTitle": "Engineer" | Experience uses jobTitle, not role or position |
"start": "2022" / "end": "2023" | "startDate": "2022-01" / "endDate": "2023-06" | Wrong field names — start/end are silently ignored in experience and education |
"skills": [...] | "hardSkills": [...] etc. | Generic skills array is ignored — use 4 separate arrays |
"slug": "dev" inside cv_data | "slug": "dev" at top level | slug and template_id are request-level fields, not inside cv_data |
"startDate": "January 2024" | "startDate": "2024-01" | Dates must be YYYY or YYYY-MM format |
Sending empty arrays "hobbies": [] | Omit the field entirely | Don't send empty arrays — omit what you don't have |
| POST /respond without Authorization header | Wait for status=completed via poll_url | Inline submit requires submit_token from the 202 response hitl object |
POST /respond on an approval type step | Poll until completed, then hitl_approved_case_id | Approval steps require browser review — inline submit not permitted |
hitl_continue_case_id without cv_data | Always include the full cv_data object | The server needs cv_data on every POST to build the next review step UI |
skip_hitl: true when a HITL chain is ongoing | Continue the chain with hitl_continue_case_id | Creates a separate CV and bypasses the human's review — share one chain per CV |
Guardrails
- Rate limits (with Access-ID): 50 CVs/day
- Rate limits (without Access-ID): 3 CVs/day per IP
- Never auto-commit without requestor approval
- Claim tokens are permanent — treat as passwords
References
- CV Data Reference: All fields and constraints for
cv_data - Templates: Full template catalog with previews
- HITL Protocol: Human-in-the-loop review flow (all steps, inline submit, edit cycles)
- Access System: Rate limits and Access-ID registration
- Error Codes: Error reference and troubleshooting
- Privacy: Data handling and GDPR compliance