feat: Hookdeck Outpost quickstarts and agent onboarding prompt

[leggetter]

TL;DR

• Hookdeck-managed quickstarts (curl, TypeScript, Python, Go) plus a single agent prompt template for published docs and dashboard onboarding ({{PLACEHOLDERS}} for API base, topics, test destination, docs URL).
• Agent evaluation harness under docs/agent-evaluation/: 10 scenarios (basics → app stacks → integrate-into-existing), Claude Agent SDK runner, heuristic transcript scoring + LLM judge, lifecycle sidecars, read/bash sandbox, authoring rules (AGENTS.md, Cursor rule).
• CI slice: workflow Docs agent eval (CI slice) runs scenarios 01 + 02 against local repo docs (EVAL_LOCAL_DOCS=1), then executes generated curl + TypeScript against live Outpost; supports workflow_dispatch for manual runs.
• OpenAPI: add DestinationSchemaField.key so the published spec matches the API.

Stacking: This PR targets feat/refactor-docs. The broader docs platform / content restructure (Markdoc layout, nav, redirects migration, etc.) is not introduced here—it lives on the base branch. This branch adds quickstarts, the onboarding prompt, the eval harness, CI, and targeted doc updates on top of that foundation.

---

Goals

1. Operators can follow Hookdeck-hosted quickstarts with a consistent story across languages.
2. The same agent prompt can ship in docs and be injected in the Outpost dashboard (placeholders documented on the prompt page).
3. We can regression-test that prompt via repeatable agent scenarios and a small, billable CI slice that catches obvious drift.

---

Hookdeck quickstarts and agent prompt

• New/updated quickstart pages under docs/content/quickstarts/ (curl, TypeScript, Python, Go, overview).
• docs/content/quickstarts/hookdeck-outpost-agent-prompt.mdoc: copy-paste template with explicit rules (no API key in chat, topics reconciliation, test destination, links into the rest of the docs).
• {{DOCS_URL}} and related copy aligned with production Hookdeck docs URLs where intended.
• Supporting content updates on this branch include concepts (e.g. OSS vs managed framing, destinations hub tweak), Building your own UI, mental model / UI-guide links, Option 3 full-stack integration guidance, and the list-topics example fix for string[] API shape.

---

Agent evaluation (docs/agent-evaluation/)

Area	What shipped
Scenarios	01–10 under scenarios/: basics (curl/TS/Python/Go), app templates (Next.js, FastAPI, Go HTTP), existing-app integration (08–10).
Runner	TypeScript (run-agent-eval.ts), harness (eval-harness.ts), declarative Eval sections / pre-steps, sandboxed read/bash.
Scoring	score-transcript.ts (per-scenario heuristics, including 08–10 publish_beyond_test_only), optional llm-judge.ts against each scenario’s Success criteria.
Operations	SCENARIO-RUN-TRACKER.md, results/ layout + templates, .env.example, AGENTS.md, .cursor/rules/agent-evaluation-authoring.mdc.
CI execution	execute-ci-artifacts.sh runs generated curl + TypeScript against secrets; smoke-test-execute-ci-artifacts.sh + npm run smoke:execute-ci for local verification.

Local runs are opt-in per scenario (--scenario / --scenarios / --all); full suite and 08–10 are slow by design (clones, installs, multi-turn agents).

Here are some examples of what was built:

Basic with some UI

Integration into SaaS templates

---

CI

• Workflow: .github/workflows/docs-agent-eval-ci.yml — Docs agent eval (CI slice).
• Triggers: push to main, pull_request (same-repo only for the eval job), and workflow_dispatch.
• Steps: ci-eval.sh → npm run eval:ci (scenarios 01, 02 with heuristic + LLM judge), then execute-ci-artifacts.sh with OUTPOST_API_KEY, webhook URL from EVAL_TEST_DESTINATION_URL, and explicit OUTPOST_API_BASE_URL / OUTPOST_CI_PUBLISH_TOPIC.
• Note: Anthropic-backed steps consume API usage; fork PRs skip the job when secrets are unavailable.

---

API / SDK docs

• docs/apis/openapi.yaml: document DestinationSchemaField.key (and related spec consistency).

---

How to review

1. Quickstarts + prompt: Read hookdeck-outpost-agent-prompt.mdoc and one language quickstart end-to-end; check placeholders and links.
2. Eval harness: docs/agent-evaluation/README.md + AGENTS.md; spot-check a scenario file’s Success criteria vs score-transcript.ts / judge behavior.
3. CI: Confirm org secrets exist for ANTHROPIC_API_KEY, EVAL_TEST_DESTINATION_URL, OUTPOST_API_KEY; optional manual workflow_dispatch after merge if needed.
4. Diff scope: Compare against feat/refactor-docs in the PR “Files changed” tab for the exact delta (this description highlights the onboarding/eval work; other touched files should match reviewer expectations).

---

Follow-ups (optional)

• Expand CI to more scenarios only if cost/latency and flake budget are acceptable.
• Keep SCENARIO-RUN-TRACKER.md updated as scenarios or the prompt change.