Skip to content

obsidian-export.md

Latest commit

 

History

History
151 lines (118 loc) · 5.59 KB

obsidian-export.md

File metadata and controls

151 lines (118 loc) · 5.59 KB
Raw

Obsidian Vault Export

session-export can mirror your AI coding sessions into an Obsidian vault as structured Markdown, alongside the usual SQLite export. The feature is opt-in and shared across the supported harnesses — Kiro CLI, Codex, Claude Code, OpenCode, pi, and Grok Build all flow into one folder, so your hand-written notes stay clean.

Looking for the implementation spec or the GraphRAG follow-up? See private-docs/obsidian-export.md and private-docs/obsidian-graphrag-roadmap.md.

What it produces

<vault>/
└── AgentMemory/
    ├── 2026-06-01-claude-code-myproject-1a2b3c4d.md   ← one note per session
    └── MOC/
        ├── _index.md                                  ← project index
        └── myproject.md                               ← per-project session list

Each note carries Dataview-ready YAML frontmatter:

---
type: agent-memory
id: 2026-06-01-claude-code-myproject-1a2b3c4d
created: 2026-06-01
updated: 2026-06-01
status: active
source_tool: claude_code
source_project: myproject
session_id: <full id>
git_branch: main
tags: [agent-memory, claude_code]
date: 2026-06-01
about: []
content_hash: 39fa1138
---

The body holds a summary, key points, and a ## Related section of [[wikilink]] backlinks to vault topic notes whose titles or aliases match the session. Per-project MOC (map-of-content) notes list every session for a project in reverse-chronological order.

How AgentMemory/ differs from your study notes

obsidian_base (the flat config key) points StudyLoop at your study sources — the notes you write and study from. The obsidian: section configures this export sink — machine-generated session memory. Keeping them in a separate AgentMemory/ folder means hundreds of auto-generated notes never dilute your curated Sessions//Study/ material, while Dataview can still query across both.

Enabling it

Three ways, in order of convenience:

1. Per run (no config needed):

session-export --obsidian                      # sessions touched this run
session-export --obsidian --obsidian-backfill  # one-time: ALL history (idempotent)
session-export --obsidian --obsidian-dry-run   # preview counts, write nothing
session-export --obsidian --obsidian-vault ~/Obsidian/Personal  # override path
session-export --no-obsidian                   # force-off even if config enables it

2. Config (always on): add to ~/.config/studyloop/config.yaml:

obsidian:
  export_enabled: true            # turn the gate on
  vault_path: ~/Obsidian/Personal # defaults to obsidian_base if omitted
  memory_dir: AgentMemory
  moc_dir: AgentMemory/MOC
  backlinks: true
  granularity: both               # both | session

3. Setup wizard: studyloop setup asks whether to enable export at the Obsidian step and writes the section for you.

Incremental vs backfill

  • A plain --obsidian run only writes sessions added or updated in that run (computed by diffing session timestamps before/after export). It does not re-scan your whole history every time.
  • --obsidian-backfill writes a note for every session in the database — the one-time "import everything" path. It is idempotent: a content_hash in each note's frontmatter means unchanged notes are skipped on re-runs.

Querying exported sessions with Dataview

Because every note shares the type: agent-memory frontmatter, a Dataview block surfaces your session history anywhere in the vault:

```dataview
TABLE source_project AS Project, source_tool AS Tool, date
FROM "AgentMemory"
WHERE type = "agent-memory"
SORT date DESC
```

Health checks

studyloop doctor (category config) validates the setup:

  • the vault path exists and is a directory;
  • it contains an .obsidian/ marker (warns if not — likely not a vault root);
  • when export is enabled, the AgentMemory/ directory is writable.
studyloop doctor --category config

Safety notes

  • Notes are written after the SQLite export commits, so an export failure never leaves the vault half-written.
  • The writer is path-traversal hardened: untrusted session fields (e.g. timestamps) are sanitised and a containment guard refuses any write outside AgentMemory/.
  • Re-exports never duplicate: the content hash drives idempotent skips.

AgentMemory exports versus Study projections

If you also use the second-brain layer, your vault has two StudyLoop folders. They hold different things and neither reads the other.

AgentMemory/ (this page) Study/ (Second Brain)
What it contains one note per AI coding session your study plans, today's action, due reviews, learning records
Where it comes from the sessions database, via session-export --obsidian your plan Markdown documents
Written by agent-session-tools studyloop brain publish
Turned on with obsidian.export_enabled second_brain.provider
Your own notes not applicable Study/Plans/<plan-id>.notes.md, which StudyLoop only reads

The split is deliberate: session memory is machine-generated history, while a study plan is material you read and revise. Mixing them would put a wall of transcript notes in the folder you open to decide what to study.

Related