Skills for your coding agent

Give your coding agent a reusable Graphrun workflow, then connect it to live diagrams and tools through MCP.

How skills and MCP work together

A skill is a reusable set of instructions and reference guides your coding agent can load for a task. The Graphrun skills repository provides author-graphrun-models, which teaches the agent how to inspect, build, validate, simulate, and explain a system model.

The skill supplies the workflow: discover what exists, make focused changes, and check the result. Graphrun MCP supplies the live connection: read authorized diagrams, edit their drafts, and run journeys when your access allows it. Use both together so the agent follows the workflow against the actual model.

Browse the source and installation instructions in the Graphrun skills repository.

Install for your coding agent

The MCP connection supplies live Graphrun state and tools. The author-graphrun-models skill gives a coding agent the matching workflow. It covers discovering components and their supported ports, authoring graph behavior and contracts, preparing and running journeys, deployment overlays, and metadata-only implementation mapping.

The skill also teaches safe edits: use current references and version tokens, make bounded changes, reread after edits, recover from conflicts, and verify the actual route and outcome of every critical journey. Structural validation alone is not proof that a journey works.

Supported behaviors include Gateway call_next_backend for round-robin calls and Database writeOperation: atomic-batch for 1–32 ordered writes to one collection. Batch writes use the intrinsic database handler; keep validation and recovery in the caller. Their data contract uses operation: write with transactional: atomic_batch. The skill covers both commit and rollback verification. General transactions across calls or collections remain unsupported. See the database component reference and data contract guidance.

Install for this project

npx skills add graphrun/skills \
  --skill author-graphrun-models

Install across your projects

npx skills add graphrun/skills \
  --skill author-graphrun-models \
  --global

Review the public Graphrun skills repository before installing it. Run npx skills add graphrun/skills --list to inspect available skills, or npx skills update author-graphrun-models to update an existing installation. Review local customizations before updating. These commands use the Skills CLI. Run the project command from your repository and follow the installer prompts to select your coding agent. Use the global command to make the skill available across projects. Reload your agent's session if the installed skill does not appear.

Note

Installing the skill adds agent instructions only. It does not configure the MCP endpoint, create an Agent Access Key, or grant access to a workspace. Skills do not add server capabilities or bypass permissions, workspace write controls, or execution availability. Repository access belongs to the coding client, not to Graphrun MCP.

Use the skill with MCP

Install the skill for your coding agent and confirm that the agent can find author-graphrun-models.
Follow Agent access & MCP to create an Agent Access Key, connect your client, and verify access. Installation and MCP setup are separate steps.
Choose a prompt below and replace the example diagram or system name with your own. Ask the agent to use the installed skill and Graphrun MCP, and confirm it has loaded the skill before making changes. Skill selection shortcuts vary by coding agent.

What you can do through MCP

Graphrun gives an MCP client a grant-scoped architecture workspace. The client can combine Graphrun with the files, tickets, or repositories it can already access. The Agent Access Key, workspace write controls, and server availability determine which Graphrun operations are usable. These prompts describe tasks for the agent, not additional built-in tools or a guarantee that every requested feature is modeled.

Note

These prompts name both the installed skill and Graphrun MCP. Use your coding agent's skill selector or supported invocation syntax if it requires one. Confirm the skill is loaded and the MCP connection works before authoring.

Understand an existing system

Suggested access: View diagrams only; Custom access for journey and contract evidence

Inspect checkpointed architecture and component structure without changing the workspace, then add read access only for the extra evidence the task needs.

Try this prompt

“Use the author-graphrun-models skill and Graphrun MCP to open the Checkout Platform diagram and explain the end-to-end purchase flow. Identify every service, queue, data store, external dependency, trust boundary, and modeled failure route. Separate facts from assumptions, do not make changes, and tell me if journey or contract access would provide useful missing evidence.”

Turn a brief into an executable architecture

Suggested access: Full MCP access; Custom needs View diagrams, Create and edit diagrams, View journeys, Create and edit journeys, the Diagrams and architecture and Journeys information types, the Real stored values payload policy, and journey execution enabled

Create a diagram, choose components from Graphrun's catalog, connect handler-owned interactions, lay it out, and add journeys that prove important routes.

Try this prompt

“Use the author-graphrun-models skill and Graphrun MCP to create a Graphrun diagram for a multi-tenant notification platform. Include the public API, tenant and preference services, scheduler, queue, email and SMS workers, provider APIs, database, cache, and dead-letter queue. Model one successful delivery journey and one provider-timeout journey, validate both, and summarize any assumptions.”

Map a codebase into Graphrun

Suggested access: Repository access in the client plus Full MCP access; Custom needs View diagrams, Create and edit diagrams, View journeys, Create and edit journeys, the Diagrams and architecture and Journeys information types, the Real stored values payload policy, and journey execution enabled

Let an agent that can also read your repository translate real applications, routes, dependencies, queues, and stores into a model while marking uncertain inferences.

Try this prompt

“Use the author-graphrun-models skill and Graphrun MCP to study this repository, then create a Graphrun diagram of the system it implements. Trace entry points, service calls, asynchronous messages, data stores, and third-party APIs back to file paths. Add the main user journey and one failure journey. Mark anything inferred rather than evidenced, then validate the result.”

Design APIs, events, and data contracts

Suggested access: Full MCP access; Custom needs View diagrams, View contracts, Create and edit contracts, and the Diagrams and architecture and Contracts information types, and the Real stored values payload policy

Author HTTP operations, message operations, schemas, examples, and logical resources, then use the gap report to find missing bindings or inconsistent shapes.

Try this prompt

“Use the author-graphrun-models skill and Graphrun MCP to define the create-order HTTP contract, the OrderAccepted event, and the orders data resource for the Order Service. Include request, success, validation-error, and conflict shapes. Bind each operation to the correct component handler, run the contract gap report, and fix blocking gaps without approving or publishing anything.”

Exercise success and failure journeys

Suggested access: Full MCP access; Custom needs View diagrams, View journeys, Create and edit journeys, the Diagrams and architecture and Journeys information types, the Real stored values payload policy, and journey execution enabled

Create exact saved journeys with fixtures, run them against the current draft, and inspect bounded event, payload, and failure evidence.

Try this prompt

“Use the author-graphrun-models skill and Graphrun MCP to create and run two checkout journeys: a successful card payment and a payment-provider timeout followed by retry exhaustion. Use deterministic fixtures, verify the expected response or failure at each step, and report the actual route, state changes, emitted messages, and first unexpected event.”

Review resilience before an incident

Inspect timeouts, retries, backoff, acknowledgements, dead-letter paths, handler failures, and state mutations, then prove the improved behavior with a journey.

Try this prompt

“Use the author-graphrun-models skill and Graphrun MCP to review the payment path for reliability gaps. Check timeout ownership, retry limits, backoff, duplicate delivery, idempotency, queue ACK/NACK behavior, dead-letter routing, and failure responses. Propose the smallest supported changes first; after I confirm, apply them and run a journey that proves the failure route terminates safely.”

Plan a deployment topology

Suggested access: Full MCP access; Custom needs View diagrams, Create and edit diagrams, and the Diagrams and architecture information type

Add a deployment overlay for regions, zones, runtimes, networks, and replicas without changing the logical behavior of the graph.

Try this prompt

“Use the author-graphrun-models skill and Graphrun MCP to create a deployment view for the existing checkout system across two regions. Place edge, API, worker, database, cache, and queue workloads into appropriate zones and network boundaries. Show the active-region path and failover dependencies, validate the overlay, and call out where the logical model does not contain enough evidence for a deployment decision.”

Assess a proposed architecture change

Suggested access: Custom diagram, journey, and contract viewing to assess; Full or suitable Custom write access to implement

Use the current model as evidence for change-impact analysis before touching the draft, then apply only an approved bounded slice.

Try this prompt

“Use the author-graphrun-models skill and Graphrun MCP to assess what would change if inventory reservation moved from a synchronous request to an event-driven workflow. Trace affected components, contracts, ordering guarantees, failure handling, journeys, and operational trade-offs. Do not edit the diagram yet. Give me a minimal change plan and the journeys we should run to compare the designs.”

Find and repair model gaps

Suggested access: Full MCP access; Custom needs diagram, journey, and contract read and write actions, the Diagrams and architecture, Journeys, and Contracts information types, the Real stored values payload policy, and journey execution enabled

Validate the graph and contracts, find disconnected or non-executable behavior, and make focused repairs while preserving correct work.

Try this prompt

“Use the author-graphrun-models skill and Graphrun MCP to audit the Order Processing diagram for validation failures, disconnected components, interactions without arrival-handler behavior, producers without message emission, unbound contracts, missing failure outcomes, and journeys that cannot execute. Rank the findings, fix only blocking issues, then revalidate and run the critical journey.”

Create a review-ready checkpoint

Suggested access: Full MCP access; Custom needs View diagrams, View journeys, Create and edit journeys, View contracts, Create checkpoints, the Diagrams and architecture, Journeys, and Contracts information types, the Real stored values payload policy, and journey execution enabled

Freeze an immutable graph revision with selected journey snapshots after validation, giving reviewers a stable point-in-time model.

Try this prompt

“Use the author-graphrun-models skill and Graphrun MCP to prepare the current checkout draft for architecture review. Validate the graph and contracts, run the successful checkout and payment-timeout journeys, summarize unresolved warnings, and create a checkpoint containing the exact validated graph and those saved journeys. Do not approve, publish, share, or export anything.”

Tip

Strong prompts name the target diagram, the outcome, the evidence to inspect, whether changes are allowed, and the verification you expect. Ask the agent to mark assumptions whenever the model or connected repository does not provide enough evidence.