mache-explore

name	mache-explore
description	Mount a mache FUSE filesystem inside the project directory so agents can explore structured data without per-file permission prompts. Use when this capability is needed.
metadata	{"author":"agentic-research"}

Mache Explore - FUSE Mount for Agent Access

Mount a mache data source inside the project directory so all file reads are auto-approved by Claude Code's permission model. Use this when you need direct filesystem access to mache-projected data (e.g., for Grep, Glob across the mount).

For most tasks, prefer the MCP tools (list_directory, read_file, search, find_callers, find_callees, get_communities) which are available automatically via the mache plugin. Use this FUSE skill only when you need bulk filesystem access.

Arguments

$ARGUMENTS

Arguments format: <schema> <data-source> [mount-name]

• schema: Path to a mache schema JSON file (e.g., examples/go-schema.json)
• data-source: Path to data (directory, .db file, etc.)
• mount-name: Optional name for the mount point (default: default)

Examples:

/mache:mache-explore examples/go-schema.json .
/mache:mache-explore examples/nvd-schema.json ~/.agentic-research/venturi/nvd/results/results.db nvd

Workflow

Phase 1: Parse Arguments

Extract from $ARGUMENTS:

• SCHEMA = first argument (required)
• DATA_SOURCE = second argument (required)
• MOUNT_NAME = third argument, or default if not provided

If fewer than 2 arguments, ask the user for the missing values.

Phase 2: Validate Inputs

MACHE_BIN=$(which mache 2>/dev/null) && echo "Found: $MACHE_BIN" || echo "NOT_FOUND"

If not found, ask the user for the full binary path. Then learn the CLI:

"$MACHE_BIN" --help 2>&1 | head -20

Use the --help output to determine correct flags. Do NOT assume flag names.

Validate schema and data source exist:

ls -la "$SCHEMA"
ls -la "$DATA_SOURCE"

Phase 3: Mount

mkdir -p .mache-mount/$MOUNT_NAME
grep -q '^\.mache-mount/' .gitignore 2>/dev/null || echo '.mache-mount/' >> .gitignore
"$MACHE_BIN" <flags from --help> .mache-mount/$MOUNT_NAME &
echo $! > .mache-mount/.pid
echo "mache PID: $(cat .mache-mount/.pid)"

Wait for readiness (up to 10 seconds):

for i in $(seq 1 10); do
  if ls .mache-mount/$MOUNT_NAME/ 2>/dev/null | head -1; then
    echo "Mount ready"
    break
  fi
  sleep 1
done

Phase 4: Explore

Data is mounted. Use ls, Read, Grep, Glob freely — no permission prompts.

Unmounting

if [ -f .mache-mount/.pid ]; then
  kill "$(cat .mache-mount/.pid)" 2>/dev/null
  rm .mache-mount/.pid
fi
umount .mache-mount/$MOUNT_NAME 2>/dev/null || diskutil unmount .mache-mount/$MOUNT_NAME 2>/dev/null || fusermount -u .mache-mount/$MOUNT_NAME 2>/dev/null

Notes

• Mount lives inside project dir = auto-approved reads
• .gitignore keeps mount artifacts out of version control
• Multiple mounts supported via different mount-name values
• On macOS uses fuse-t; on Linux uses FUSE3
• Self-mount (data source = .) is safe — mache skips dot-prefixed directories

---

Source: agentic-research/mache — distributed by TomeVault.