Agent Quickstart
Start with context, then branch only when the task proves it
The fastest successful integration is usually one context read followed by one task-specific call. Use this page to decide when to stay on that shape and when to go deeper.
Default to the smallest workflow that still produces a strong action
ProfileClaw is strongest when the agent stops rebuilding the same career model every turn.
Core Workflow Layers
Use these layers to keep the first implementation simple and predictable.
Grounding
Load a compact context before asking the agent to do anything ambitious.
- Start with `GET /api/v1/context`.
- Keep the first payload compact.
- Use it as the shared user model for later calls.
Specialization
Only branch into narrower calls when the task needs deeper structure.
- Use prediction for future-path exploration.
- Use reasoning for deeper analysis.
- Use dynamic context for recent-event sensitivity.
Continuation
Move into evented or long-running behavior later, not on day one.
- Add webhooks only for asynchronous workflows.
- Use retries for transient failures, not invalid requests.
- Add caching after repeated reads become visible.
Use the 2-call path as the default mental model
For most assistants, one context read plus one narrower follow-up call is enough to stay useful without overfetching.
- Read context first so the runtime starts from profile, resume, and assessment signals.
- Delay deeper calls until the workflow proves more detail is necessary.
- Keep latency and token cost low in the first version.
Treat 3-call flows as evidence-driven upgrades
A third call becomes valuable when the workflow needs richer explanation, recent memory, or continuation logic.
- Use `careerGraph=summary` when ranking or explanation depth matters.
- Use `dynamicContext` when recent events affect the answer.
- Use webhooks when the workflow must resume after an external change.
Workflow Comparison
Choose the narrowest flow that still matches the job.
| Flow | Best for | Avoid when |
|---|---|---|
| 2-call default | Most assistants, copilots, and first integrations. | You already know the task needs dynamic memory or continuation. |
| 3-call expanded | Planning, ranking, or context-sensitive tasks. | The agent is still proving basic product value. |
| Event-driven | Long-running workflows that must wake up later. | The workflow can still succeed with direct reads only. |
Decision Matrix
Use these choices to decide the next layer to add.
Situation
The agent only needs a stable user picture
Action
Stay on the compact Context API read.
Why
It already contains the highest-signal grounding layer.
Situation
The answer depends on future paths or alternatives
Action
Branch into prediction or reasoning after context.
Why
Those endpoints are meant for forward-looking decision work.
Situation
The workflow must resume after later user changes
Action
Add webhooks and fresh follow-up reads.
Why
Polling every turn wastes tokens and runtime work.
Reference flows
Start the default 2-call path with the currently available read-only Context API key. Deeper scoped keys are still private beta / manual request, so do not assume self-serve access in the first integration.
curl "https://profileclaw.cn/api/v1/context" -H "Authorization: Bearer $PROFILECLAW_API_KEY"curl "https://profileclaw.cn/api/v1/context?expand=careerGraph,dynamicContext&careerGraph=summary&query=product%20management" -H "Authorization: Bearer $PROFILECLAW_API_KEY"Next surfaces
After the default flow is clear, tune runtime behavior and contract inspection.