Skip to content

example-prompt.md

Latest commit

 

History

History
59 lines (46 loc) · 3.05 KB

example-prompt.md

File metadata and controls

59 lines (46 loc) · 3.05 KB
Raw

Worked example — implement a feature with the full toolchain

A copy-paste prompt that drives one feature through every tool ai-dev-kit sets up: graphify (orient), Grep MCP (prior art), Superpowers / the AGENTS.md workflow (plan + TDD), and private-journal (record decisions).

Works in Claude Code, Codex, or Copilot. Swap the feature in bold for your own.

The prompt

Implement rate limiting for our HTTP API. Use our full toolchain and follow our
workflow — don't jump straight to code.

1. Orient (graphify). Run:
     graphify query "how do incoming HTTP requests flow and where is middleware registered"
     graphify path "<router>" "<handler>"
   to find where a limiter should sit and what it touches. Summarize the integration points.

2. Prior art (Grep MCP). Search public repos via the Grep MCP for real-world rate-limiter
   middleware in our language (token-bucket / sliding-window). Show 2-3 concrete examples
   and their trade-offs — ground the design in how others actually do it.

3. Decide + record (private-journal). Choose an algorithm and config (limit, window,
   storage, per-key strategy). Record the decision AND the rationale in your journal so
   future sessions know why — e.g. "chose token-bucket backed by Redis because ...;
   rejected sliding-window-log because ...".

4. Plan + build (workflow). Write a failing test for the limiter first, implement it, wire
   it into the integration points from step 1, then run the test / lint / typecheck commands from
   AGENTS.md. Self-review the diff against the Standards (including Code review).

5. Wrap up. Summarize what changed, run `graphify update .` to refresh the graph, and tell
   me what you recorded in the journal. Don't push or open a PR.

What each tool does here

Step Tool Role
1 graphify Map the codebase — where the feature plugs in and what it touches (recall, not grep)
2 Grep MCP Ground the design in real implementations across ~1M public repos
3 private-journal Persist the design decision + rationale across sessions
4 Superpowers / AGENTS.md workflow Enforce plan -> failing test -> implement -> verify
5 graphify update . Keep the graph current after the change

Notes

  • Tool availability. graphify is on by default; Grep MCP needs --with-grep, private-journal needs --with-journal. If a tool isn't installed, the agent will just skip that step.
  • Superpowers enforces the brainstorm -> plan -> TDD -> review flow in Claude Code. Codex and the Copilot cloud agent can't run the plugin, so they follow the same flow from the Development workflow section of AGENTS.md.
  • Want an HTML write-up of the result? Add "...and give me an HTML report of the change" or run /report-html afterward (see AGENTS.md -> Output formats).
  • Adapt freely. This is one ordering that uses every tool; on a real task the agent may loop (e.g. re-query graphify after finding prior art). The point is the habit: orient -> prior art -> decide+record -> TDD -> verify.