Add tree-sitter parsers for bash (.sh) and JSON (.json)

[mattmillartech]

Summary

Graphify currently doesn't index .sh (bash) or .json files at the AST level — these extensions aren't in the extension→extractor map in graphify/extract.py, and the corresponding tree-sitter-bash / tree-sitter-json packages aren't dependencies.

Result: graphs of repositories that include hook scripts, CI configs, MCP configs, package manifests, and other operationally-relevant files have no nodes for any of that content. graphify get_node "scripts/hooks/my-hook.sh" returns "not found" even when the file exists.

Reproduction

In a fresh graphify venv:

# Verify parsers
pip list | grep -E "tree.?sitter"
# No tree_sitter_bash or tree_sitter_json

# Verify extension map
grep -E "\.(sh|json)" graphify/extract.py
# No .sh entry; only stray .json references for tsconfig.json path resolution

# Verify empirical
graphify update /path/to/repo-with-shell-scripts
graphify query "any-bash-symbol-in-the-repo"   # zero results

Why this matters

For users on AI agent platforms (Claude Code, Codex, Cursor, etc.), a significant portion of the "operating environment" lives in shell + JSON:

• .claude/settings.json — hook registration, MCP servers, permissions
• scripts/hooks/*.sh — PreToolUse / PostToolUse / Stop hooks
• .mcp.json — MCP server configuration
• package.json — script entry points

These files contain real cross-references (a settings.json matcher points at a hook script; a hook script invokes another shell script; a package.json script references a .sh entry-point). Without bash/json indexing, queries like "how does PreToolUse on Bash get gated?" return zero hits — agents fall back to manual file walks.

Proposed fix

1. Add tree-sitter-bash and tree-sitter-json to pyproject.toml dependencies.
2. Register the parsers in graphify's language map (whichever file holds the tree-sitter language registration).
3. Add extension→extractor entries:

   ".sh": extract_bash,
   ".bash": extract_bash,
   ".json": extract_json,

4. Write extract_bash (functions + top-level commands) and extract_json (top-level keys + nested object refs) — both can be lightweight; the goal is file + symbol nodes, not deep semantic analysis.

Workaround

For users who can't wait, a workspace-local monkey-patch:

# scripts/persistence/graphify-bash-json-shim.py
import graphify.extract as e
# ... register bash + json extractors ...
import sys; sys.exit(__import__("graphify.__main__").main())

But this is fragile and breaks on graphify upgrades. Upstream support is the right answer.

Context

Surfaced while building a 6-layer agent intelligence stack (graphify + lightrag + ruflo + mempalace + nexus + obsidian-mind). Graphify's blind spot for the agent operating environment was the root cause of a protocol violation: when the protocol demanded "query graphify before grep," graphify had nothing to return for hook-mechanism questions, so the fallback path (manual grep) became the only option.

Happy to contribute the PR if there's interest in this scope.

[safishamsi]

Fixed in branch v8, commit 40b9b84.

Added extract_bash() and extract_json() extractors and registered .sh, .bash, .json in _DISPATCH. Added tree-sitter-bash and tree-sitter-json as dependencies.

Bash extracts: functions (with defines edges), cross-function calls edges, source/.\ imports resolved to file paths, export/declare` variable declarations.

JSON extracts: top-level and nested keys (contains tree), dependencies/devDependencies → imports edges, extends edges (tsconfig/eslint), $ref references. Size guard skips files > 1 MiB.

19 new tests, 798 passing total. The monkey-patch workaround mentioned in the issue is no longer needed.