Built for Agentic Workflows

Prompt Cookbook

Concrete starting points you can paste into your assistant. The goal is always the same: gather structure first, then write code.

Create an implementation plan
Turn a feature request into a realistic plan aligned with the existing architecture.
get-repository-overviewget-symbol-contextread-file
Example prompt
You are helping me implement a feature in a TypeScript monorepo.

Feature: Add a “Bulk rename tags” action to the Settings UI. Tags are stored in a backend `TagStore` and referenced by `TaskCard` and `TaskFilterBar`.

Use context master tools to propose a realistic implementation plan:
- Start with get-repository-overview and identify the most central symbols related to tags, settings, and filters.
- Use get-symbol-context on those symbols to map blast radius and key call sites.
- Use read-file only for the small set of files you expect to modify.

Output:
- A step-by-step plan (files to touch, high-level changes)
- Edge cases (duplicates, undo, optimistic updates)
- A minimal test plan.
Understand a legacy flow
Answer “how does this work?” with a traceable execution path, not guesses.
get-file-treeget-symbol-contextread-file
Example prompt
Help me understand a bug in our “Invite user” flow.

Question: When a user clicks “Invite” in `InviteMemberDialog`, how does the request propagate to the backend and where do we enforce role checks?

Use context master tools:
- Use get-file-tree to locate entrypoints (dialog component, API client, server handler).
- Use get-symbol-context to follow referenced-by edges for `inviteMember` and any auth/permission checks.
- Use read-file for only the few files that define the behavior.

Output a concrete call chain (UI -> client -> server -> persistence) and where errors are generated.
Plan a safe refactor
Map blast radius before touching code so you don’t ship breaking changes.
get-repository-overviewget-symbol-context
Example prompt
I want to refactor `parseMoney` to return a structured result instead of throwing:
`parseMoney(input: string): { amount: number; currency: string } | { error: string }`

Use context master tools:
- get-repository-overview to find where money parsing is central (billing, invoices, checkout).
- get-symbol-context for `parseMoney` and the top dependents (referenced-by).

Output:
- A step-by-step migration plan (add new API, update call sites, remove old behavior)
- A list of the highest-risk call sites that need careful updates.
Generate targeted tests
Write tests that match real usage, not imagined inputs.
get-symbol-contextread-file
Example prompt
I need tests for `computeRetryDelay` (used by our job queue).

Use context master tools:
- get-symbol-context on `computeRetryDelay` and enumerate referenced-by call sites.
- read-file around each call site to extract real inputs (attempt counts, jitter flags, max delay caps).

Then propose a test matrix that covers actual usage patterns and edge cases (first retry, max cap, jitter on/off).
Draft docs from code reality
Produce documentation that reflects what the code actually does today.
get-repository-overviewget-symbol-contextread-file
Example prompt
Draft internal docs for the `WebhookRouter` module.

Use context master tools:
- get-repository-overview to identify entrypoints and central webhook-related symbols.
- get-symbol-context to list handlers, dependents, and shared utilities.
- read-file for the authoritative module files and 1-2 representative handler examples.

Write docs with:
- Entry points and how requests are routed
- How signatures are verified
- Where retries/idempotency are handled
- How to add a new webhook handler (step-by-step).

Keep exploring