I want flag three issues that with Claude's help I identified that keep coming up, in case they're
useful.
- Line numbers in contentManager.update
Any edit shifts subsequent line numbers, which means every
targeted update needs a fresh read immediately before it. A
find-and-replace style option — something like {find:
"string", replace: "string", nth?: 1} — as an alternative to
line targeting would eliminate the problem entirely without
breaking the existing API.
- property:set overwrites list fields
Calling property:set on tags, aliases, or similar list
properties replaces the whole list rather than appending. Any
automated update on an already-tagged note wipes existing
tags. A property:append tool or a merge mode flag on
property:set would solve this — the current workaround is
editing the raw frontmatter, which just reintroduces problem
#1.
- No confirmation from write/update responses
I've been informed that best practice is to read back immediately after
every write to confirm it landed, which doubles the tool
calls. If the response included the affected lines or a content hash that extra
read could be skipped.
Hope this helps.
I want flag three issues that with Claude's help I identified that keep coming up, in case they're
useful.
Any edit shifts subsequent line numbers, which means every
targeted update needs a fresh read immediately before it. A
find-and-replace style option — something like {find:
"string", replace: "string", nth?: 1} — as an alternative to
line targeting would eliminate the problem entirely without
breaking the existing API.
Calling property:set on tags, aliases, or similar list
properties replaces the whole list rather than appending. Any
automated update on an already-tagged note wipes existing
tags. A property:append tool or a merge mode flag on
property:set would solve this — the current workaround is
editing the raw frontmatter, which just reintroduces problem
#1.
I've been informed that best practice is to read back immediately after
every write to confirm it landed, which doubles the tool
calls. If the response included the affected lines or a content hash that extra
read could be skipped.
Hope this helps.