Frontmatter fence detection is not line-anchored: inline '---' in single-line content gets parsed as frontmatter and writes garbage YAML keys

[groksrc]

Summary

has_frontmatter(), parse_frontmatter(), and remove_frontmatter() in src/basic_memory/file_utils.py detect frontmatter by substring/split rather than line-anchored fences. The frontmatter convention requires --- fences on their own lines; the current logic (content.startswith("---") + content.split("---", 2)) accepts inline fences anywhere.

Repro

Send single-line content that starts with --- — e.g. a shell command where \n doesn't expand (a very common agent/CLI input shape):

bm tool write-note --title "Meeting Notes" --folder meetings \
  --content "---\nstatus: active\n---\nDiscussed Q3 roadmap with Anthony."

The CLI receives the literal one-line string ---\nstatus: active\n---\nDiscussed Q3 roadmap with Anthony.

Resulting file on disk:

---
title: Meeting Notes
type: note
permalink: main/meetings/meeting-notes
\nstatus: active\n
tags:
- meeting
---

\nDiscussed Q3 roadmap with Anthony.

Two failures:

1. A garbage YAML key literally named \nstatus is merged into the note's frontmatter and written to disk.
2. The body is silently transformed (the inline ---…--- segment is stripped out of the content).

Root cause

src/basic_memory/file_utils.py:305-323 (has_frontmatter) and :326-348 (parse_frontmatter, content.split("---", 2)). remove_frontmatter shares the pattern.

Proposed fix

Require line-anchored fences (^---\s*$ multiline). The repo already depends on python-frontmatter (used by dump_frontmatter in the same module), whose loader is line-anchored — either delegate to it or anchor the split. Single-line content starting with --- should be treated as plain content, never as frontmatter. Apply consistently across all three helpers.

Context

Found during manual QA of the June 11 PR batch; present on main @ v0.22.0, unrelated to the open PRs.

🤖 Generated with Claude Code

[claude]

Claude finished @groksrc's task in 4m 52s —— View job • Create PR ➔

---

Triage & Fix: Frontmatter fence detection not line-anchored

Triage:

• Type: Bug — data corruption
• Priority: High — garbage YAML keys written to disk; content silently stripped
• Component: Core (file_utils.py), affects CLI and MCP write-note
• Complexity: Simple — anchor the --- detection to line boundaries

Root cause: has_frontmatter, parse_frontmatter, and remove_frontmatter used content.startswith("---") + content.split("---", 2) which matches --- anywhere in the string. A literal ---\nkey: val\n---\nbody (backslash-n, no real newlines — common from CLI/agent input) was falsely treated as frontmatter, writing garbage YAML keys to disk and silently stripping the body.

Fix applied (src/basic_memory/file_utils.py):

Added a module-level constant:

_FRONTMATTER_FENCE_RE = re.compile(r"^---[ \t]*$", re.MULTILINE)

All three helpers now use _FRONTMATTER_FENCE_RE.split(content, 2) and require parts[0] to be empty (opening fence at position 0). A string with no real newlines after --- produces zero regex matches → treated as plain content.

Regression test added (tests/utils/test_file_utils.py → test_inline_fence_not_treated_as_frontmatter): covers the exact repro from the issue.

• Gather context and read issue
• Read file_utils.py to understand root cause
• Add bug label
• Fix has_frontmatter(), parse_frontmatter(), remove_frontmatter() with line-anchored detection
• Add regression tests
• Run tests (requires approval for test execution)
• Commit (db36c45) — push blocked by repo permissions