Before any implementation, Claude needed to read the memory files at:
~/.claude/projects/-Users-kcossabo-Documents-Coding-Container-Managment/memory/
Specifically:
– project_container_pipeline.md — the Gitea/Komodo workflow, compose.yml conventions, unRAID labels, known landmines
– project_komodo_containers.md — server inventory, IP ranges, bridge names, Komodo field gotchas
Without this context, Claude would generate generic compose files that don’t match the existing pipeline (wrong network names, wrong IP ranges, wrong Komodo field behavior).
“I am thinking through the documentation, outside of the AI context/memory.
Think wiki-like, file-based. I mount a NAS folder that text/MD files are created with AI at the end of a project. The ‘app’ uses the file store as backend for its presentation layer. I can go to knowledge.cossaboon.net and see what and when things were done, projects that provide value, search across them. AI prompts would be a key part of this as in 6 months this desktop app might be completely different and AI tools grown, and need the ‘institutional’ knowledge to live on.
Based on this present a solution that would meet these needs and anything else that would be of value.”
At the end of any session, use this to generate the wiki entry:
“Generate a project knowledge entry for
/Volumes/knowledge/docs/projects/[YYYY-MM-slug]/.
Create:
–index.md— overview, architecture, key decisions, outcome
–ai-prompts.md— the context Claude needed, prompts that worked, what to watch for
–lessons.md— what broke, what surprised, what to do differentlyUse frontmatter with title, date, tags (pick from: spacedock, dockyard, dockofthebay, containers, infrastructure, media, documentation, ai-prompts), server, status, ip.
Then run: kcommit ‘[description of what was added]'”
kcommit alias only works if /Volumes/knowledge is mounted. If the NAS is unreachable, the commit will fail silently by landing in the wrong directory.