Quartz 4

agentic-coding

1 min read

Prerequisite

Workflow

key idea is to

  • leverage “free memory” of agents
  • document your intentions specifically
  • code in natural language

For example, structure as follows:

example-repo
|____docs
| |____space
| | |____agent        # agentic (temporary; like gh issue)
| | |____project      # project-related (evergreen; always up-to-date)
| | |____index.md
| |____compose.yml    # silverbullet docker compose file
|____src
| |____example_repo
| | |____example.py

NOTE hierarchy of documents:

  • high-level intention ⇒ docs
  • code structure ⇒ README.md
  • fine-grained ⇒ inside code as comment

This way,

  • document reside in an ideal place
    • agents have more access to the repo
    • users have easy-access human-readable docs
  • you can code in natural language through agent

code in natural language

For example, make your issue in docs/space/agent:

example-issue

And refer to that issue. For example:

opencode .
 
# 1. if you wrote solution by yourself
[Plan] Check if @docs/space/agent/example-issue.md is valid and update that with specific TODO list.
 
[Build] Proceed as planned
 
# 2. if you want agent to finish solution of the issue
[Build] Fill out solution in @docs/space/agent/example-issue.md then validate. Afterwards, update that with specific TODO list.
 
[Build] Proceed as planned

Graph View

Backlinks