generated: '2026-08-26'
method: generated
source: >-
  Grounded in openapi/*.json in this repo and the Pypestream reference pages cited in each skill.
  Every operationId, path, parameter, status code, event name and published limit quoted in these
  skills was read from a Pypestream-published document; none is invented.
note: >-
  Pypestream publishes no AGENTS.md, no skills index and no agent-facing guide of any kind. Its
  developer portal declares an agent_skills_index feature but serves nothing at any skills path
  (all 404). These four skills are written by API Evangelist over Pypestream's published contracts.
provider_published_skills: false
skills:
- name: pypestream-embed-and-run-a-chat
  file: skills/pypestream-embed-and-run-a-chat.md
  api: pypestream:engagement-api
  operations: 7
  summary: >-
    The marquee flow. Half REST, half Phoenix Channels WebSocket, gated on chat:ready or you get a
    428.
- name: pypestream-escalate-to-a-live-agent
  file: skills/pypestream-escalate-to-a-live-agent.md
  api: pypestream:contact-center-api
  operations: 14
  summary: >-
    Webhook-driven escalation from an automated Pype to a human agent, with context handoff.
- name: pypestream-pull-conversation-and-agent-reporting
  file: skills/pypestream-pull-conversation-and-agent-reporting.md
  api: pypestream:reporting-api
  operations: 6
  summary: >-
    Warehouse extraction under a 400/hour limit, per-endpoint window maxima and a silent 30-day
    retention cliff.
- name: pypestream-export-analytics-to-a-warehouse
  file: skills/pypestream-export-analytics-to-a-warehouse.md
  api: pypestream:analytics-api
  operations: 22
  summary: >-
    Batch exports to S3 or Snowflake with the pause/unpause/backfill reversal path spelled out.
not_written:
- api: pypestream:javascript-sdk
  reason: >-
    The contract is documentation-only and describes client-side SDK methods, not callable HTTP
    operations. A skill grounded in it would teach an agent to POST to endpoints that do not exist.
- api: pypestream:insights-api
  reason: >-
    One operation, and every published server is a Vercel preview deployment or localhost. No
    production host to ground a skill in.
- api: pypestream:middleware-api
  reason: >-
    Twelve of its operations duplicate the Contact Center API on identical paths, and it publishes
    no host. Covered by the escalation skill instead.