/graph_add — Smart Content Addition to Obsidian Vault (v3) Skill

/graph_add — Smart Content Addition to Obsidian Vault (v3)

Thesis: 2-pass Gemini pipeline — Gemini navigates vault tree, reads candidate files, decides UPDATE/CREATE/REFACTOR, and writes the actual content. Claude reviews the plan and executes it.

Execution Protocol

Step 1: Prepare Session Summary

Claude (the agent) writes a concise session summary covering:

All topics/tasks worked on
Key entities (customers, tools, people, technologies)
Artifacts created or modified (files, skills, configs)
Patterns, decisions, gotchas discovered

Keep it 3-8 sentences. This is the input for Gemini.

Step 2: Run graph_add() Python Function

claude_venv/bin/python -c "
import sys, json
sys.path.insert(0, '.')
from data_sources.obsidian.ops import graph_add

plan = graph_add('''SESSION_SUMMARY_HERE''')

print(json.dumps({
    'action': plan.action,
    'filepath': plan.filepath,
    'section': plan.section,
    'content': plan.content,
    'remove_lines': plan.remove_lines,
    'reasoning': plan.reasoning,
    'secondary_updates': plan.secondary_updates,
}, ensure_ascii=False, indent=2))
"

This runs the full pipeline:

Gemini #1 (Flash): Navigates compact vault tree → finds 3-5 candidate files/folders
Python: Expands top-3 candidates (reads file content, up to 100 lines each)
Gemini #2 (Pro): Receives session summary + expanded content → returns ActionPlan

Step 3: Review Gemini's Plan

Claude MUST review before executing:

Is the target file correct?
Is the section appropriate?
Does the content make sense?
Are remove_lines actually contradictory?
Do secondary_updates look reasonable?

If anything seems wrong — adjust or ask user.

Step 4: Execute

Based on action field:

UPDATE

mcp__obsidian__patch_vault_file(
    filename="filepath from plan",
    target="section from plan",
    targetType="heading",
    operation="append",
    content="content from plan"
)

If remove_lines is not null — Read file, Edit to remove those lines first, then append.

CREATE

mcp__obsidian__create_vault_file(
    filename="filepath from plan",
    content="content from plan"
)

REFACTOR

Follow Gemini's plan: create consolidated file, patch old files with > [!info] Consolidated into [[New File]].

For secondary_updates — apply each one the same way.

Architecture

Claude (session summary)
    → Python: graph_add(summary)
        → Gemini Flash: suggest_placement → candidates
        → Python: expand_nodes → file contents
        → Gemini Pro: decide_action → ActionPlan
    ← ActionPlan {action, filepath, section, content}
Claude reviews → executes via MCP/Edit

Gemini = author (navigates, decides, writes content) Claude = reviewer + executor (validates, uses tools)

Content Rules

Wikilinks: [[File Name]] (name only, no paths)
Tags: in frontmatter tags: array
Dates: ISO YYYY-MM-DD
No secrets: per CLAUDE.md §2.1 ¶4
Session tracking: add session_id: to frontmatter per §2.3

Related Skills

| Skill | Relationship | |-------|-------------| | /graph_find | Independent — graph_add has its own search via suggest_placement | | /obsidian | MCP tools for execution (create, patch, show) | | /knowledge-framework | Apply MECE structure for docs >50 lines |

Created: 2026-02-05 | Version: 3.0.0 Changes: v3 — 2-pass Gemini pipeline replacing graph_find dependency. Gemini writes content, Claude reviews+executes.

Agent Skills: /graph_add — Smart Content Addition to Obsidian Vault (v3)

Install this agent skill to your local

Skill Files