memory write: --update can't find notes; --category ignored and type dir mis-pluralized

[JaySmith]

Summary

cortex memory write computes the note file path from --type alone (naive type + "s"), so:

1. It ignores --category entirely when building the target directory.
2. Naive pluralization produces wrong dir names (entity -> entitys, knowledge -> knowledges).
3. --update therefore cannot find existing notes created by the encoder/import, which live under <type-plural>/<category>/<id>.md (e.g. entities/systems/, knowledge/api/).

Net effect: --update fails with "Note does not exist" for notes that clearly exist (cortex memory get <id> returns them), and fresh writes land in a flat, wrong directory (I found a stranded knowledges/cortex-lint-fix-...md from an earlier write).

Repro

# Note exists and is readable:
cortex memory get jira            # works -> entities/systems/jira.md

# But update fails:
cortex memory write --update --title "jira" --type entity --category systems \
  --tier skill:jira --body-file ./x.md --no-encode
# -> "Note does not exist at <vault>/entitys/jira.md"
#    (wrong: naive plural 'entitys', and category 'systems' dropped)

Same for --type knowledge --category api -> looks in knowledges/ instead of knowledge/api/.

Root cause

cortex/cli/main.py:1133-1137:

type_dir_name = note_type + ("s" if not note_type.endswith("s") else "")
target_dir = vault_path / type_dir_name          # <-- --category never used
target_dir.mkdir(parents=True, exist_ok=True)
note_path = target_dir / f"{note_id}.md"

• entity + "s" -> entitys (should be entities).
• --category is accepted as an option but never contributes to the path.

Suggested fix

The encoder already solves this correctly for the update case — cortex/encoder/core.py:897-901 does a recursive lookup by id before falling back to a type dir:

for md in vault_root.rglob("*.md"):
    meta, _ = parse_frontmatter(md.read_text(...))
    if meta.get("id") == note_id:
        local_path = md
        break

memory write should do the same:

1. On --update: locate the existing note by id via rglob (matching the encoder), regardless of directory, instead of reconstructing the path from --type.
2. On create: build the path as <type-plural>/<category>/<id>.md using --category, and use a correct pluralization map (entity -> entities, knowledge -> knowledge/knowledges per convention) rather than blind + "s".

Workaround

Edit the note files directly in the vault (preserve frontmatter, bump updated), then run cortex encode.

Environment

• cortex-ai 2.0.0 (uv tool install from ~/Projects/Cortex-AI)
• macOS (darwin), Python 3.12