$ man context-wiki/taxonomy

How you organize files is context engineering. If your repo is a mess, your AI outputs will be a mess. Not because the model is confused by messy folders. Because messy folders mean the model cannot find what it needs, loads the wrong files, or skips context entirely. Predictable structure means predictable results. When every partner folder has the same subfolders (research/, prompts/, workflows/, resources/), Claude knows exactly where to look for ICP definitions, campaign copy, and qualification prompts. It does not search. It does not guess. It goes to the predictable location and reads the file.

PATTERN

content/drafts/ holds work in progress. content/published/ holds shipped posts with date prefixes. partners/{name}/research/ holds ICP, personas, and competitive intel. partners/{name}/prompts/ holds Clay qualification and research prompts. partners/{name}/workflows/ holds campaign sequences and routing logic. skills/ holds every reusable workflow as a SKILL.md file. data/ holds generated assets like dashboards and progression stats. scripts/ holds Python automation. Every folder has a purpose. Every file has a predictable name. When I tell Claude to "check the partner research," it knows exactly where to look because the structure is the same for every partner. That consistency is the taxonomy.

PRO TIP

Date-prefixed drafts: 2026-02-16_topic-name.md. This sorts chronologically and tells Claude when something was written. Partner folders use lowercase slugs: partners/acme/, partners/contax/. Skill files are always SKILL.md inside a named folder: skills/deploy/SKILL.md, skills/tracker/SKILL.md. Scripts use snake_case: daily_scan.py, rpg_sprites.py. The naming convention is not about aesthetics. It is about predictability. When Claude needs to find the deploy skill, it looks for skills/deploy/SKILL.md. Not deploy-skill.md. Not DeploySkill.md. Not deploy.skill. One pattern. Every time. Predictability eliminates search time and hallucinated file paths.

ANTI-PATTERN

The worst thing you can do is dump everything in one folder. I have seen repos with 200 files in the root directory. No subfolders. No naming convention. File names like "final_v2_ACTUAL.md" and "notes_old_backup.txt". Claude cannot make sense of that. Neither can you. If you cannot find a file in under 5 seconds by navigating the folder tree, your taxonomy is broken. Fix the structure before you add more content. A clean repo with 20 files will outperform a messy repo with 200.