brek-ai-skill

Integrate with Brek Partner Core Chat API for hotel-search and booking assistant flows. Use when an agent needs to create or continue Brek chat sessions, send user events safely, enforce anti-abuse call controls, and run payment setup or payment confirmation without collecting raw card data.

Safety Notice

This listing is from the official public ClawHub registry. Review SKILL.md and referenced scripts before running.

Copy this and send it to your AI assistant to learn

Install skill "brek-ai-skill" with this command: npx skills add leo9/brek-ai

Brek AI Partner Core Chat

Execute Brek through /api/partner/v1/core-chat.

Use this execution order:

  1. Create one session per end user (POST /sessions) with a stable actor.actorId.
  2. Reuse that session for all follow-up messages (POST /events).
  3. Read the latest state when needed (GET /sessions/{sessionId}).

Do not share one session across different users.

Required runtime inputs

Require these inputs before calling Brek:

  • BREK_BASE_URL
  • BREK_PARTNER_API_KEY
  • actorId (stable partner-side end-user ID)
  • workspaceId or tenant context if your product uses workspaces
  • partnerId (stable partner tenant ID for idempotency and billing grouping)
  • clientActionId for each write-like event kind

If one required input is missing, stop and request it.

If BREK_PARTNER_API_KEY is missing:

  • stop outbound calls
  • ask the user to get the key from their internal owner or approved support channel
  • never request secrets through unapproved channels

Call safety guardrails

Apply these guardrails before every upstream call:

  1. Enforce local budget limits from references/call-control.md.
  2. Attach a deterministic idempotencyKey for all write-like event kinds.
  3. Respect 429 with retry-after and exponential backoff.
  4. Open a circuit breaker after repeated 5xx or timeout failures.
  5. Log x-request-id, x-partner-id, x-ratelimit-limit, and x-ratelimit-remaining.

Never retry booking or payment-confirm actions without the same idempotencyKey.

Event-kind rules

When kind is one of:

  • command_book_by_option_id
  • action_book_option
  • action_confirm_price_change
  • action_confirm_payment_card
  • action_cancel_booking

Always include idempotencyKey.

Generate idempotencyKey as:

  • <partnerId>:<sessionId>:<kind>:<clientActionId>

If partnerId is unavailable in your runtime, use stable tenant context (for example workspaceId) and keep the key format deterministic.

Payment handling

Handle payment in two layers:

  1. End-user card setup and confirmation in secure portal flow.
  2. Agent-to-agent usage billing and settlement.

Follow references/payment-and-billing.md for both layers.

Hard rules:

  • Never request or store raw card number, CVV, or full PAN in chat.
  • Accept only tokenized paymentMethodId from provider-hosted fields.
  • Require explicit user confirmation before action_confirm_payment_card and booking actions.

Response mapping

Map Brek response as:

  • data.result.status -> state machine key for UI and orchestration
  • data.result.message.text -> user-visible assistant text
  • data.result.artifacts -> structured payload (shortlist, payment setup URL, booking metadata)

Error handling

  • 400: request validation failed. Fix payload.
  • 401/403: API key issue. Stop calls. Tell user to rotate or provision key through their internal owner or approved support channel.
  • 404: session not found or wrong tenant.
  • 409: actor/session mismatch. Recreate correct session.
  • 429: throttle locally and retry by retry-after.
  • 5xx: retry with backoff, then open breaker.

References

Read only what you need:

  • API payload templates: references/api-templates.md
  • Call-control and anti-abuse policy: references/call-control.md
  • Payment and billing orchestration: references/payment-and-billing.md

Source Transparency

This detail page is rendered from real SKILL.md content. Trust labels are metadata-based hints, not a safety guarantee.

Related Skills

Related by shared tags or category signals.

Automation

Cypress Agent Skill

Production-grade Cypress E2E and component testing — selectors, network stubbing, auth, CI parallelization, flake elimination, Page Object Model, and TypeScr...

Registry SourceRecently Updated
Automation

Ichiro-Mind

Ichiro-Mind: The ultimate unified memory system for AI agents. 4-layer architecture (HOT→WARM→COLD→ARCHIVE) with neural graph, vector search, experience lear...

Registry SourceRecently Updated
1128
hudul
Automation

Reddit Engagement

Create and execute robust Reddit engagement workflows (create post, add comment, upvote) using browser accessibility-tree semantics instead of brittle DOM id...

Registry SourceRecently Updated