todo

**where to start when you cannot tell:** everything ranked urgent against important - three streams: reference checks, doc-vs-reality, and recent agent logs - categorizes every opportunity on an urgent/important matrix - broken references are one signal among many, never the point # Instructions ## Telemetry ```! "${CLAUDE_PLUGIN_ROOT}"/skills/todo/todo.sh $ARGUMENTS echo "sidecar exit: $?" ``` - `help: requested` → the run was refused before it started; `## Help` below is the whole turn - it already ran, so there is no command to issue - report-only; it never fails the run — capture its telemetry (broken references + code markers) 1. reconcile docs against reality — run this block once per objective, one objective at a time: - **read:** the source of truth (the code, dirs, or config the doc describes) - **search:** where the doc makes claims about it - **reconcile:** does every claim still match reality? - **flag:** each drift as an opportunity (note the file and the mismatch) objectives (one at a time): - **README.md** — does it still describe the project, stack, and setup accurately? - **AGENTS.md** — do its prose rules still hold (naming, css, imports, mirroring, etc.)? - **plugins/\*/** — does each skill's doc still match its own sidecar's flags and behavior? 2. read the 5 most recent agent logs in `.construct/retardify/log/`, the shape `/retardify:log` defines - extract observations, pain points, unfinished tasks, recurring bugs, or architectural ...