Problem
search_conversations (and the underlying conversation index) does not index cldctrl notepad / scratchpad .md files. Content that was drafted with an agent and saved to a notepad is therefore invisible to conversation search, even though notepad artifacts are exactly the kind of thing a user searches for later.
Real example (repro)
A peer-review draft, tolmachev-breathing-swallowing-review.md, was authored via cldctrl and saved to the scratch project (AppData/Roaming/cldctrl/scratch/). Later, search_conversations for the review ("swallowing breathing half-center oscillator review", "deglutition", etc.) returned nothing relevant: the token "swallowing" appeared in only 4 indexed sessions, none of them the review. The file was recovered only by a filesystem grep, not by conversation search. From the user's point of view the work was effectively lost until a manual disk search found it.
Proposal
Index notepad/scratchpad markdown as a searchable source. Either:
- fold the notes stores into the same index that
search_conversations queries (so notepad hits come back inline), or
- add a dedicated notepad search, and/or have
search_conversations return notepad matches tagged with a source: "notepad" (vs source: "session") field.
Locations to crawl: at minimum the scratch project dir (AppData/Roaming/cldctrl/scratch/*.md) and the control project's drafts/recaps, plus wherever open_scratchpad / save_scratchpad persist. Index on save (or crawl on demand) and keep it incremental.
Why it matters
Notepads accumulate the highest-value drafted deliverables (peer reviews, email drafts, abstracts, recaps). They are the most valuable class of artifact to be able to find again, and right now they are the one class the search cannot see.
Problem
search_conversations(and the underlying conversation index) does not index cldctrl notepad / scratchpad.mdfiles. Content that was drafted with an agent and saved to a notepad is therefore invisible to conversation search, even though notepad artifacts are exactly the kind of thing a user searches for later.Real example (repro)
A peer-review draft,
tolmachev-breathing-swallowing-review.md, was authored via cldctrl and saved to thescratchproject (AppData/Roaming/cldctrl/scratch/). Later,search_conversationsfor the review ("swallowing breathing half-center oscillator review", "deglutition", etc.) returned nothing relevant: the token "swallowing" appeared in only 4 indexed sessions, none of them the review. The file was recovered only by a filesystemgrep, not by conversation search. From the user's point of view the work was effectively lost until a manual disk search found it.Proposal
Index notepad/scratchpad markdown as a searchable source. Either:
search_conversationsqueries (so notepad hits come back inline), orsearch_conversationsreturn notepad matches tagged with asource: "notepad"(vssource: "session") field.Locations to crawl: at minimum the
scratchproject dir (AppData/Roaming/cldctrl/scratch/*.md) and thecontrolproject's drafts/recaps, plus whereveropen_scratchpad/save_scratchpadpersist. Index on save (or crawl on demand) and keep it incremental.Why it matters
Notepads accumulate the highest-value drafted deliverables (peer reviews, email drafts, abstracts, recaps). They are the most valuable class of artifact to be able to find again, and right now they are the one class the search cannot see.