0009: Markdown is the only source of truth; checker keeps no index
Context: Agents and humans edit docs directly. A planned checker CLI (check, qa run,
touching, tags, links) needs to read them.
Decision: The checker parses files on every call. No stored index or cache.
Zero-install (uvx), with a guide subcommand like aesop tools. No search command: agents use ripgrep.
Consequences: Never stale; no sync step. Fine while repo docs are small (hundreds of files). → reflected in architecture: tooling.
Alternatives: | option | why not | |---|---| | stored index (SQLite/JSON) | goes stale on every direct edit | | docs in a database, rendered to md | agents and git diff work on files | | built-in search | ripgrep already does it |