Extended checkbox markers ([/], [>], [?], [X]) mint junk observation categories

[phernandez]

Sibling of #1219, found while reviewing checkbox semantics: the observation exclusion list
only knows GFM's three markers ([ ], [x], [-]), so Obsidian's extended checkbox
vocabulary leaks into the graph as junk observation categories:

- [/] in progress task  ->  {'category': '/', 'content': 'in progress task'}
- [>] deferred task     ->  {'category': '>', 'content': 'deferred task'}
- [?] maybe task        ->  {'category': '?', 'content': 'maybe task'}

[X] (uppercase, GFM-legal) also slips through as category X.

Fix direction

Deliberately narrow, like #1239: treat any single-character non-alphanumeric bracket
prefix (plus [X]) as task-marker-shaped, never an observation category. Single-char
alphanumeric categories other than x/X ([a], [1]) keep parsing as today.

Longer term these lines belong to first-class task extraction (#1242), which will
own the full marker family; this exclusion is the interim guard.

[claude]

Claude finished @phernandez's task in 45s —— View job

---

Triage

Type: Bug — observation_plugin's task-marker exclusion (src/basic_memory/markdown/plugins.py:22) only checks for [ ], [x], [-], so Obsidian's extended checkbox vocabulary ([/], [>], [?]) and uppercase [X] fall through into the [category] content regex and mint junk observation categories like /, >, ?, X.

Component: Markdown parsing / sync (core). Not a cloud-specific issue.

Priority: Medium — doesn't break ingestion, but pollutes the knowledge graph with garbage categories that show up in search/browsing.

Complexity: Simple — narrow, targeted fix in the same spirit as #1239: widen the task-marker guard at plugins.py:22 to treat any single-character non-alphanumeric bracket prefix (plus [X]/[x]) as task-marker-shaped, while leaving other single-char alphanumeric categories ([a], [1]) parsing as observations. Confirmed the exclusion list and reproduced the described gap in code.

Related but not a duplicate: sibling of #1219 / #1239, which cover the timestamp-shaped-category exclusion. This issue is scoped separately to the checkbox-marker family, per the report. No further action taken — implementation is out of scope for triage.