# Capstone Ideas — Salesforce Lead Developer Documentation Writer

> Generated by Claude Code from `inputs/jd/JD_Salesforce_LeadDevDocWriter.md`. Each idea is scoped to about a week of evenings and maps to skills the posting actually asks for. Build one before the interview and you walk in with proof instead of promises.

## Idea 1 — Docs-as-code pipeline with automated style governance

- **JD skills it proves:** docs-as-code, CI/CD for content, style and terminology governance.
- **What it is:** A small docs repo where every pull request runs a linter that enforces a written style guide, terminology list, and link checker, with violations posted as review comments.
- **Demo moment:** Open a PR with three deliberate style violations and watch the pipeline catch all three.

## Idea 2 — API reference generated from an OpenAPI spec, with an agent eval

- **JD skills it proves:** API documentation, developer experience, AI-assisted workflows.
- **What it is:** A reference site generated from a public OpenAPI spec, plus a tiny eval that asks an AI agent to complete five tasks using only the docs, and scores whether it succeeds.
- **Demo moment:** Show the eval failing on a thin page, fix the page, and show the eval pass.

## Idea 3 — Interactive quickstart that measures time-to-first-success

- **JD skills it proves:** onboarding content, metrics-driven writing, developer empathy.
- **What it is:** A single quickstart for a public API with checkpoints that log how far a reader gets, so the doc itself reports where people fall off.
- **Demo moment:** A funnel chart of real checkpoint data from a few test readers.

---

## Chosen: Idea 2 — API reference with an agent eval

**Repo name:** `api-docs-agent-eval`

| Milestone | Deliverable |
|---|---|
| M1 (days 1-2) | Reference site generated from the OpenAPI spec, deployed |
| M2 (days 3-4) | Five agent tasks written, eval harness running against the docs |
| M3 (day 5) | One weak page found by the eval, rewritten, eval passing |
| M4 (days 6-7) | Short README write-up: what the eval caught, before and after |

**Why this one:** it connects the two strongest themes in the posting, API documentation and AI-assisted workflows, and it mirrors a real story I can tell in the interview about replacing a vanity metric with an eval.