# Keep the build context to what the builder's COPY actually needs: the

# workspace manifests, crate/service sources, and the vendored proto/ trees the

# build.rs protox passes compile. Everything below is either a build artifact,

# local dev state, or documentation the build never reads.

# Rust build artifacts (rebuilt inside the builder stage).

target/

**/*.rs.bk

*.profraw

# Local dev / test state. .claude/ includes any local git worktrees a session

# left behind (e.g. under .claude/worktrees/); each is a full duplicate source

# tree that is never needed by the build and has been observed to inflate the

# build context by an order of magnitude on a dev machine with a few stale

# ones lying around. Fleet executors keep gate logs and cargo-install trees in

# .gate-logs/ and .dd-tools/ inside the checkout (see .gitignore), and graft

# keeps its regenerable graph cache in graft/; none of the three is read by

# the build.

rustfs-data/

.cache/

.superpowers/

.claude/

node_modules/

.DS_Store

.gate-logs/

.dd-tools/

graft/

# Version control and CI metadata.

.git/

.github/

.gitignore

# Docs and top-level markdown: not read by any build.rs or by cargo.

docs/

*.md

# Deployment manifests and shell scripts: same reason as docs/, and they matter

# more than docs do. `COPY . .` is one layer, so before these were excluded,

# editing a Kubernetes manifest or a kind script invalidated the cached builder

# layer and forced a full release rebuild of the workspace (tens of minutes) to

# produce a byte-identical binary. No build.rs or cargo target reads either

# tree; the release binaries do not change when they do.

deploy/

scripts/

# Editor/OS cruft.

**/__pycache__/