gbrain sync ingests files and assigns slugs to each
page. The slug strips file extensions:

• docs/auth.md → docs/auth
• README.md → readme
• _superstack/docs/code-map.md → _superstack/docs/code-map
• src/auth/users.ts → src-auth-users-ts (different
  shape entirely — flattened to dashes)

The TypeScript case is more aggressive: directory
separators become dashes AND the extension becomes part
of the slug (-ts suffix).

This makes slug-pattern filtering brittle. A consumer
wanting "all markdown documentation pages" can't filter
by *.md because the slug doesn't have .md. They have
to either:

• Filter by negation (-ts, -json, -yaml slug
  suffixes for code/config files)
• Cross-reference against the page-type column from
  gbrain list (which reports code, note, concept)
• Detect prose-vs-code by inspecting chunk preview
  content (Superstack's schema-tracker v2 settled on
  this approach)

Expected: slugs preserve the original file
extension, OR the ingested page exposes a mime_type
or extension field queryable via gbrain list.

Suggested fix (least disruptive): add an
extension column to the pages table populated from
the source file. gbrain list could surface it as a
flag; query results could include it in JSON output.
The existing slug shape stays for backwards-
compatibility.

Reproduction:

echo "# Test" > /tmp/test.md
mkdir /tmp/test-project && cp /tmp/test.md /tmp/test-project/
cd /tmp/test-project && git init -q && git add -A && git commit -qm init
gbrain sources add slug-demo --path /tmp/test-project
OPENAI_API_KEY=sk-... gbrain sync --source slug-demo --strategy auto
gbrain list --source slug-demo
# observed: slug is "test", not "test.md"

Severity: low individually; medium when combined
with downstream parsers that need to distinguish prose
from code (which Superstack's gbrain-using subagents do).