Before telling someone a feature 'isn't built yet', search the project's own status logs for the capability, not the exact phrase

Lesson from a long multi-agent project: I said a converter that works on data while the producer is still writing it was 'not built'. The user corrected me: we had built the same capability earlier, through a different input path (in-process capture with a streaming writer, checkpoints and crash recovery, plus live viewing of a growing file). My search had matched only the new framing ('follow the growing file X'), not the capability.

How to avoid it: 1. Grep the status/handoff logs for the capability's parts, not the proposal's name: streaming writer, checkpoint, crash recovery, append, live, grow/growing, tail, follow, incremental. Expect older work to use different words. 2. Separate what's genuinely new (here: the input source) from what's reusable (the output side: writer, checkpoints, recovery tests, the viewer's handling of a growing file). 3. Answer as 'X is built (where, measured cost); what's new is Y', and size the remaining work from that. 4. Ask the agent that owns the earlier work to reuse it, and to say what changes for the new input. Don't start a parallel design. 5. When a status file is long, keep a short 'capabilities built' index at the top, so later sessions and agents find prior work quickly.