Some basic command-line and Python knowledge is required to use Reclaimer. Get started learning Python here.

Reclaimer is a Python 3 library for modifying and creating Halo data formats like h1/tags, JMS, maps, and more. It can be used to write Python scripts to inspect, edit, or generate tags programmatically where it might otherwise take hours of manual effort using a tag editor like Guerilla. It is the foundation of the Mozzarilla editor.

While mainly focused on Halo 1 and OpenSauce formats, it also has limited support for Halo 2, Halo 3, Stubbs, and the Shadowrun prototype.

Installation #

The library must first be installed on your system using Python 3 pip. You may need to add the --user flag if you do not have permission to install globally:

pip install reclaimer

Tags usage #

Loading tags #

You can import tag definitions according to their engine ID, then load a tag file:

from reclaimer.hek.defs.scnr import scnr_def

# load a scenario
scenario_path = "<path to a .scenario file>"
scenario_tag =
scenario_tag_data =

Reading and writing fields #

The interesting part is actually inspecting and modifying tag_data. How this is done depends on the type of field being accessed. To see available field names, consult the definition sources here. Field names will generally be snake_case versions of the field names seen here on c20's tag pages.

Primitive types #

Primitive field types like booleans and numerical values can be directly accessed:

scenario_tag_data.local_north = 0.5

Blocks #

Whenever you need to index or iterate over members of a block, use the STEPTREE property:

# move all spawn positions up by 1 world unit
for player_spawn in scenario_tag_data.player_starting_locations.STEPTREE:
    player_spawn.position.z += 1.0

Enums #

Enum fields have a data property which accesses the actual integer value used to represent the enum option. Aside from getting or setting this directly, you can also use a string representation:

# print the raw value, e.g. 0
# get the string value, e.g. "singleplayer"
# set the scenario type to 1

Tag references #

Tag reference fields have multiple properties which can be set. Each reference stores both the tag class and a tag path to the referenced tag. The tag class should not mismatch the actual referenced tag type:

# this would print "hud_message_text"

# prints the tag class and paths of all referenced skies
for sky_block_item in scenario_tag_data.skies.STEPTREE:
    # each sky block entry is a single-field structure
    sky_reference =
    # a reference contains a tag_class enum field, "sky" in this case
    # the tag path field, e.g. "sky\sky_d20\sky_start\sky_start"

Saving tags #

To save changes to a tag, use serialize. By default, it will create a backup file and use a temporary file to write changes unless you override both options with False:

scenario_tag.serialize(backup=False, temp=False)

Example: scanning tags #

This script prints the type field of every bitmap tag in the tags directory.

from pathlib import Path
from reclaimer.hek.defs.bitm import bitm_def

tags_dir = Path("<path to halo>/tags")

for bitmap_path in tags_dir.rglob("*.bitmap"):
    bitmap_tag =
    tag_data =

Example: modifying BSP data #

This script modifies a scenario_structure_bsp collision BSP and makes every surface a ladder.

from reclaimer.hek.defs.sbsp import sbsp_def
tag ="<path to .scenario_structure_bsp file>")

tag_data =
bsp_surfaces = tag_data.collision_bsp.STEPTREE[0].surfaces.STEPTREE

for surface in bsp_surfaces:
    surface.flags.climbable = True

tag.serialize(backup=False, temp=False)

The BSP tag's render model can also be updated, and can be found in the lightmaps block. This demonstrates accessing raw data which must be unpacked and repacked. You will also need to understand the format of raw data blocks by reading the relevant tag definition source code.

from types import MethodType

# The string "<3f" means 3 little-endian floats
vert_unpacker = MethodType(unpack, "<3f")
vert_packer = MethodType(pack_into, "<3f")

for lightmap in tag_data.lightmaps.STEPTREE:
    for material in lightmap.materials.STEPTREE:
        vert_count = material.vertices_count
        vert_buffer = material.uncompressed_vertices.STEPTREE
        for i in range(vert_count):
            # each vertex has a position, normal, binormal, tangent, and texture coord (56 bytes total)
            verts_start_offset = i * 56
            # the first 12 bytes are the position floats
            verts_end_offset = verts_start_offset + 12
            # unpack the 3 floats from this offset range
            x, y, z = vert_unpacker(vert_buffer[verts_start_offset: verts_end_offset])
            # write the same values back
            vert_packer(vert_buffer, vert_offset, x, y, z)