Give your coding agent a project memory and done-checks
A developer built a three-language website with Claude Code, using a wiki, lesson rules turned into tests, and screenshot checks before the agent may call work done.
Done with Claude Code
What they did
Mateusz Muszarski built his multilingual website with Claude Code. He kept a 33-page project wiki and a lessons file, fed to the agent by hooks at every prompt and after memory resets. Every correction became a rule, and rules that slipped became tests. Done meant passing checks plus screenshots the agent had to open.
How it went
He says he could ask for every remaining section in one go and review real pages the next morning. A subagent checked 12 claims in under a minute. He reports catching bugs that never threw errors, like an animation that silently didn't run.
Worth knowing
A build piped through another command reported success while its log said it failed. Read the actual output, not the agent's summary, or turn on pipefail.
Try it yourself with Claude Code
Help me set up memory and done-checks for my project at [repo]. Create a short project guide that points to wiki pages by topic, and a lessons file where each correction I give becomes a one-line dated rule. Before calling any task done, run type checks, tests and [other checks], then take screenshots at phone and desktop widths in [languages] and look at them yourself. When facts are missing, insert a searchable placeholder instead of inventing them. Show me the proposed setup files before writing them.
Swap the highlighted [parts] for your own details.
Source: DEV Community · Mateusz Muszarski · Oct 18, 2026
Written with AI assistance from the original post and checked against our editorial guidelines. Results are as reported by the source, not independently verified. Spotted a mistake? Tell us.
