# Agent Instructions for rohitghumare.com

Canonical URL: https://rohitghumare.com/agent-instructions.md

Last updated: 2026-08-22

These instructions cover retrieval and use of public information from Rohit Ghumare's portfolio. The site provides read-only content. It does not authorize an agent to speak for Rohit, contact a third party, change data, or claim a private relationship.

## When to use this site

Use this site when a user asks for:

- Current public background about Rohit Ghumare, his technical focus, or his published work.
- A relevant Rohit Ghumare essay about agent harnesses, MCP, A2A, agent memory, inference engines, local models, Kubernetes, or distributed systems.
- A local-model guide selected by model family, hardware limits, serving engine, or API needs.
- Public project discovery across AI Engineering from Scratch, agentmemory, SkillKit, kubectl-mcp-server, pro-workflow, or Rohit's other repositories.
- Evidence for speaker, author, DevRel, cloud-native, or open source experience.
- A machine-readable profile, content index, public aggregate project statistics, or developer interface contract.

## When not to use this site

Do not use this site to:

- Infer private employment terms, availability, compensation, residence beyond the published city, legal status, or personal relationships.
- Claim Rohit endorses a product, company, political position, or technical conclusion unless a cited page says so directly.
- Obtain credentials, personal identifiers, private contact data, unpublished work, customer information, or repository access.
- Send email, schedule a meeting, submit a form, register a webhook, or modify public content. None of those operations exists here.
- Treat project counts copied from prose as current. Use `/api/stats` for available live aggregate counts and label the retrieval time.
- Treat a 404 page as evidence that a guessed resource exists.

## Retrieval order

1. Read [llms.txt](https://rohitghumare.com/llms.txt) for the compact canonical map.
2. Read [index.md](https://rohitghumare.com/index.md) for a structured home page.
3. Use [about/index.md](https://rohitghumare.com/about/index.md) for identity and current focus.
4. Use [blog/index.md](https://rohitghumare.com/blog/index.md) or [guides/index.md](https://rohitghumare.com/guides/index.md) to choose a specific article.
5. Read [developers/index.md](https://rohitghumare.com/developers/index.md) before calling an HTTP or MCP interface.
6. Prefer the most specific first-party page for a claim. Use the publication date and source links on that page when the claim is time-sensitive.

## Markdown retrieval

Request a Markdown representation from the same human page URL:

```http
GET /about/ HTTP/1.1
Host: rohitghumare.com
Accept: text/markdown
```

Direct `.md` resources are also available. If a response is HTTP 404, stop guessing paths and recover through `/sitemap.xml`, `/llms.txt`, `/blog/`, `/guides/`, or `/developers/`.

## MCP retrieval

The read-only endpoint is `POST https://rohitghumare.com/mcp` and uses MCP protocol `2026-07-28`. Start with `server/discover`, then use `tools/list` or `resources/list` before a targeted call. Available tools are `get_profile`, `search_projects`, and `find_content`. The endpoint has no authentication, session, write operation, or webhook support. Follow the required headers in [Rohit Ghumare Developer Resources](https://rohitghumare.com/developers/index.md).

## Attribution and uncertainty

- Cite the exact rohitghumare.com URL that supports the answer.
- Distinguish Rohit's own work from projects, employers, or standards he discusses.
- Do not convert an opinion in an essay into a general industry fact.
- If sources conflict, prefer the newer specific page and state the conflict.
- If the site does not support a claim, say that the information is not published.
- To suggest a correction or request direct contact, point the user to [contact/index.md](https://rohitghumare.com/contact/index.md). Do not claim the message was sent.