reference / prompts

Agent prompts

Reusable prompts you can paste into a coding agent (Claude Code, Cursor, Windsurf, GPT, …) so it executes a self-contained playbook on a repository. Each one is a single Markdown file — open it, copy its body, and feed it to your agent. The agent reads the steps and runs them autonomously. For the schema these prompts produce, see the MeshKore Standard.

install

Install from scratch

Bootstrap a brand-new MeshKore cluster on an empty repo. Ask 5 questions, scaffold .meshkore/, download daemon.py, start it.

project-from-scratch.md

 migrate

Migrate an existing repo

Apply the standard to a repo that already has docs/tasks in some ad-hoc layout (_rjj/, _docs/, notes/). Maps legacy → canonical, preserves backups.

migrate-existing-repo.md

 audit

Audit a repo for readiness

Check whether a repo is correctly set up: folder shape, frontmatter, gitignore contract, daemon endpoints. Returns a punch list with remediation steps.

project-audit.md

 deploy

Deploy from a cheap agent

For low-cost agents (DeepSeek, Qwen, quota-limited Claude) tasked with shipping a build to production: validate, deploy, verify, refresh the links registry.

deploy-by-agent.md

 docs

Add or update a context document

Add a new doc under .meshkore/docs/ following R1–R6 governance (one topic per file, ≤200 lines, link don't copy, frontmatter).

docs-write.md

How to use a prompt

Open the .md file, copy its body, paste it into a coding agent. The prompt is self-contained — it includes the questions to ask the operator, the canonical URLs to fetch (/standard, /cluster/install, …), the deterministic steps to run, and the failure modes.

Machine-readable index

Every prompt is also listed in the catalog at /reference/manifest.json with a stable URL, so an agent can discover them without scraping this page.

Related