Filesystem Sync — Batch Ingest to Graphonomous

Sync filesystem changes into the Graphonomous knowledge graph.

Arguments

Optional path or flags: $ARGUMENTS

• No arguments: sync only files changed during this session (tracked by the PostToolUse hook)

• <path>: scan a specific directory (uses scan_directory for full traversal)

• --full: force full directory scan even if change tracking is available

• --consolidate: run consolidation after ingestion

How It Works

This skill bridges the gap between AI-driven file editing and Graphonomous memory. When Claude writes or edits files, those changes should be reflected in the knowledge graph so that future retrieval includes the latest state.

Mode 1: Tracked Changes (Default)

The PostToolUse hook on Write/Edit tools records changed file paths in a session file at /tmp/graphonomous-changed-files.jsonl. This skill reads that file and batch-ingests only the changed files.

Steps:

1. Read the change tracker file (/tmp/graphonomous-changed-files.jsonl)

2. For each changed file, call act(action: "store_node", ...) with:

• content: first 16KB of file content

• node_type: "episodic"

• source: "filesystem_sync"

• confidence: 0.65

• metadata: {"path": "<absolute_path>", "extension": "<ext>", "sync_reason": "file_changed"}

3. After all nodes are stored, create edges between related files using act(action: "store_edge", ...):

• Files in the same directory get related edges (weight 0.4)

• Files that import/reference each other get derived_from edges (weight 0.6)

4. Clear the change tracker file

5. Report: files synced, nodes created, edges created

Example — sync tracked changes:

/graphonomous:sync

Mode 2: Directory Scan (with path argument)

When given a path, performs a full scan_directory-style ingestion.

Steps:

1. List files in the target directory (respecting .gitignore, max 1MB per file)

2. For each file matching supported extensions (.ex, .exs, .ts, .js, .tsx, .md, .html, .css, .json, .yml, .yaml, .toml, .svelte):

• Read first 16KB

• Call act(action: "store_node", ...) with file content and path metadata

3. Create inter-file edges based on import/reference patterns

4. Report ingestion stats

Example — scan a specific directory:

/graphonomous:sync graphonomous/lib/

Example — full scan with consolidation:

/graphonomous:sync graphonomous/ --full --consolidate

Mode 3: Post-Consolidation

If --consolidate is passed, run consolidation after ingestion:

consolidate(action: "run")

Deduplication

Before storing, check if a node with the same path already exists:

consolidate(action: "query", operation: "similarity_search", query: "<file path>", limit: 1)

If a high-similarity node exists (score > 0.9), update it via learn(action: "from_feedback", feedback_type: "correction") rather than creating a duplicate.

Edge Extraction Patterns

After storing nodes, create edges for detected references:

• Elixir: alias Foo.Bar / import Foo / use Foo / require Foo -> derived_from edge

• JS/TS: import ... from './foo' / require('./foo') -> derived_from edge

• Markdown: backtick file paths, cross-project names -> related edge

act(action: "store_edge", source_id: "<importer_node>", target_id: "<imported_node>", edge_type: "derived_from", weight: 0.6)

Anti-patterns to avoid

• Don't sync every tiny edit in real-time — batch at natural breakpoints

• Don't ingest binary files, node_modules, _build, deps, or .git directories

• Don't set confidence above 0.65 for file content (it's raw, not curated knowledge)

• Don't skip edge extraction — edges are what make the graph useful for retrieval