David decided: Mandated that workspace AI chat agent read every file end-to-end, not just triage

After investigating a workspace where the AI gave incomplete answers by doing only surface-level file triage, David directed the agent to update the shared workspace-chat file-discovery instructions:

does the system prompt used for personal workspaces mandate that the ai go through every single pdf and file on the users lms, that is even vaguely associated with their query, by using the view file tool over and over and over? it must!

When the agent proposed instructions that still allowed stopping at partial inspection, David pushed back:

also this is too vague. you need to mandate that it call view_file on any and all [relevant files]

He also explicitly drew the boundary between where the rule should and should not apply:

ok but will this affect the assignment agent runs or lecture notes agent runs? i dont want it to. only the interactive chat in those workspaces. i dont even want it to affect the openclaw assignment agent generations

The update was scoped to interactive workspace chat only, leaving async generation pipelines untouched.