chore(nix): modularize tooling and migrate hooks - #1139

@coderabbitai please review this PR.

No actionable comments were generated in the recent review. 🎉

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

#!/bin/bash
# Get PR details
gh pr view 1139 --repo ryoppippi/ccusage --json title,body,files,additions,deletions,changedFiles

# Get the list of changed files
gh pr diff 1139 --repo ryoppippi/ccusage --name-only

#!/bin/bash
# Get full diff
gh pr diff 1139 --repo ryoppippi/ccusage

@ryoppippi I'll review this PR now.

Review: chore(nix): migrate hooks to git-hooks.nix

This is a well-structured migration from lefthook to git-hooks.nix via prek, with a clean modularization of the flake into flake-parts. Here's my analysis:

✅ Strengths

• Clean flake.nix simplification: The monolithic 224-line flake is reduced to 47 lines by delegating to flake-parts modules. Very readable.
• passthru pattern in package.nix: Exposing { cargoArtifacts, commonArgs, version } via passthru allows nix/checks.nix and nix/static-package.nix to reuse build artifacts without duplicating logic. Good design.
• default.nix for non-flake users: Adding default.nix (pointing to package.nix via callPackage) enables nix-build / niv users, which is a nice addition.
• Nix shebang in scripts/sync-skills.ts: Allows running the script directly with nix + bun without a pre-installed global bun.
• shopt -s nullglob in nix/sync-skills.nix: Correctly handles empty directories; broken-symlink detection via [ -L "$dst" ] is also robust.

🔍 Observations & Questions

1. check.enable = false in nix/git-hooks.nix

pre-commit = {
  check.enable = false;  # ← disables the check derivation
  ...
};

This intentionally prevents nix flake check from generating a derivation that validates the hooks config. The PR validation uses nix flake check --no-build which is consistent. Just confirming this is intentional and not accidentally suppressing a useful build check.

2. renovate in hook runtime inputs → devShell via enabledPackages

renovate-config-validator = {
  enable = true;
  entry = "${lib.getExe pkgs.renovate} --strict config-validator";
  ...
};

pkgs.renovate has a very large closure (~500 MB+). Since config.pre-commit.settings.enabledPackages is added to devShells.default.buildInputs in nix/dev-shell.nix, this pulls renovate into every developer's shell. Is this intentional? You might want to consider whether a lighter validator or a conditional tool would be preferable here.

3. Repeated pkgs instantiation across modules

Every module under nix/ independently instantiates:

pkgs = import inputs.nixpkgs {
  inherit system;
  overlays = [ inputs.rust-overlay.overlays.default ];
};

This is a known-acceptable pattern in flake-parts (Nix laziness prevents repeated evaluation), but it means the overlay is applied 6 times in isolation. The idiomatic alternative is a shared nixpkgs flake-parts module (e.g., via perSystem.pkgs). Not a bug, just worth noting for future maintenance.

4. nix/checks.nix — typos and oxfmt checks removed from flake check outputs

The old flake.nix exposed explicit checks.typos and checks.oxfmt in the flake outputs. These are now handled by treefmt-nix (which exposes checks.treefmt). Confirmed by the PR validation command:

nix build .#checks.aarch64-darwin.treefmt

This is the right approach. Just noting the change in surface area for nix flake check consumers.

5. nix/git-hooks.nix — ccusage-treefmt hook pass_filenames defaults to true

ccusage-treefmt = {
  entry = lib.getExe config.treefmt.build.wrapper;
  files = ".*";
  stages = [ "pre-commit" ];
  # pass_filenames not set → defaults to true
};

ccusage-treefmt-check explicitly sets pass_filenames = false. For consistency and clarity, it might be worth being explicit here too (even if treefmt does accept file arguments):

ccusage-treefmt = {
  entry = lib.getExe config.treefmt.build.wrapper;
  files = ".*";
  pass_filenames = false; # run treefmt on everything, not just staged files
  stages = [ "pre-commit" ];
  priority = 10;
};

6. Minor: import ./../default.nix in nix/packages.nix

ccusage = import ./../default.nix {

./../ is equivalent to ../ — the leading ./ is redundant. Not a bug, but ../default.nix is more conventional.

📋 Summary

@coderabbitai I pushed a follow-up that makes the treefmt hook filename behavior explicit and cleans up the package import path. Please review the latest commit.

@cubic-dev-ai please review the latest commit as well.

@cubic-dev-ai please review the latest commit as well.

@ryoppippi I have started the AI code review. It will take a few minutes to complete.

#!/bin/bash
# Get the latest commit on the PR branch
gh pr view 1139 --repo ryoppippi/ccusage --json headRefName,headRefOid

# Get the diff of the latest commit
gh api repos/ryoppippi/ccusage/commits/$(gh pr view 1139 --repo ryoppippi/ccusage --json headRefOid --jq '.headRefOid') --jq '.files[] | {filename: .filename, patch: .patch}'

@ryoppippi Both changes look good in commit 3bc54ac:

1. nix/git-hooks.nix — pass_filenames = true is now explicit on ccusage-treefmt. The value makes sense semantically: the format hook only processes staged files, while ccusage-treefmt-check keeps pass_filenames = false to check the full tree. That's a clean and intentional distinction between the two hooks. ✅

2. nix/packages.nix — ./../default.nix → ../default.nix. Clean. ✅

Both follow-up items are resolved. The PR looks good overall! 🐇

If you found this review helpful, would you consider giving us a shout-out on X?

Thank you for using CodeRabbit!

Open in StackBlitz

npx https://pkg.pr.new/ryoppippi/ccusage@1139

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-darwin-arm64@1139

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-darwin-x64@1139

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-linux-arm64@1139

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-linux-x64@1139

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-win32-arm64@1139

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-win32-x64@1139

commit: 00cc77f

@coderabbitai @cubic-dev-ai I pushed a follow-up fix for the nested build-output filtering issue in the Nix checks. Please review the latest commit.

@coderabbitai @cubic-dev-ai I pushed a follow-up fix for the nested build-output filtering issue in the Nix checks. Please review the latest commit.

@ryoppippi I have started the AI code review. It will take a few minutes to complete.

#!/bin/bash
# Get the latest commit SHA on the PR branch
LATEST_SHA=$(gh pr view 1139 --repo ryoppippi/ccusage --json headRefOid --jq '.headRefOid')
echo "Latest commit: $LATEST_SHA"

# Get the diff of the latest commit
gh api "repos/ryoppippi/ccusage/commits/$LATEST_SHA" --jq '.files[] | {filename: .filename, patch: .patch}'

@ryoppippi The fix in commit c289636 looks correct and addresses the issue cleanly.

nix/checks.nix — nested build-output filtering

The old hasPrefix-based approach only excluded top-level occurrences — a path like rust/crates/ccusage/target/release/foo would not have been filtered since it doesn't start with "target/". The new approach splits the path into segments and checks whether any segment exactly matches an ignored directory name, so nested build outputs at any depth are now correctly excluded. ✅

A couple of small notes (non-blocking):

1. Exact segment matching is correct here — a path like dist-utils/foo.ts splits into ["dist-utils", "foo.ts"] and builtins.elem "dist" [...] returns false, so there are no false positives from similarly-named directories. 👍
2. Files named dist, target, etc. (as bare files, not directories) would technically be excluded too, but that's an unlikely edge case and an acceptable trade-off.

The fix is solid. PR looks good overall! 🐇

ccusage performance comparison

PR SHA: d6f6ad0afc68
Base SHA: 3e912262be1d

This compares the Rust PR release binary against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 3e912262be1d; PR package: d6f6ad0afc68. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

ccusage performance comparison

PR SHA: d6f6ad0afc68
Base SHA: 3e912262be1d

This compares the PR package against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 3e912262be1d; PR package: d6f6ad0afc68. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

ccusage performance comparison

PR SHA: 3bc54ac3a328
Base SHA: 3e912262be1d

This compares the Rust PR release binary against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 3e912262be1d; PR package: 3bc54ac3a328. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

ccusage performance comparison

PR SHA: 3bc54ac3a328
Base SHA: 3e912262be1d

This compares the PR package against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 3e912262be1d; PR package: 3bc54ac3a328. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

ccusage performance comparison

PR SHA: c289636057c8
Base SHA: 3e912262be1d

This compares the PR package against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 3e912262be1d; PR package: c289636057c8. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

ccusage performance comparison

PR SHA: c289636057c8
Base SHA: 3e912262be1d

This compares the Rust PR release binary against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 3e912262be1d; PR package: c289636057c8. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

@coderabbitai @cubic-dev-ai I pushed a follow-up that moves repo-local skill generation to agent-skills-nix, keeps it local-only via the Nix dev shell, ignores generated .claude/skills, and updates the PR description. Please review the latest commit.

@coderabbitai @cubic-dev-ai I pushed a follow-up that moves repo-local skill generation to agent-skills-nix, keeps it local-only via the Nix dev shell, ignores generated .claude/skills, and updates the PR description. Please review the latest commit.

@ryoppippi I have started the AI code review. It will take a few minutes to complete.

#!/bin/bash
# Get the latest commit SHA on the PR branch
LATEST_SHA=$(gh pr view 1139 --repo ryoppippi/ccusage --json headRefOid --jq '.headRefOid')
echo "Latest commit: $LATEST_SHA"

# Get the diff of the latest commit
gh api "repos/ryoppippi/ccusage/commits/$LATEST_SHA" --jq '.files[] | {filename: .filename, patch: .patch}'

@ryoppippi Commit 00cc77f looks great — this is a clean, well-motivated simplification.

✅ What's improved

• Custom sync code eliminated: nix/sync-skills.nix (90 lines), scripts/sync-skills.ts (108 lines), two git hooks, and two package.json scripts are all gone. Replaced by the 60-line nix/agent-skills.nix delegating to the upstream agent-skills-nix library. Net reduction ~200 lines of bespoke maintenance burden.
• nix/checks.nix — nix-filter upgrade: The previous commit's custom cleanSourceWith + builtins.elem path-segment filter is replaced with the idiomatic nixFilter.matchName predicates. Cleaner and more readable. ✅
• .claude/skills/ now gitignored and generated locally: The 20 tracked symlink files are removed, .gitignore updated, and .claude/CLAUDE.md documents the new model clearly. ✅
• checks.agent-skills = bundle: CI now verifies the bundle builds cleanly. ✅

🔍 Two minor observations (non-blocking)

1. home-manager as a transitive dependency

agent-skills-nix follows nixpkgs but brings in home-manager as a transitive input (visible in flake.lock). This is a non-trivial closure addition. It won't affect runtime behavior but is worth keeping in mind if you ever see a heavier-than-expected nix flake lock or shell startup. Nothing to change, just good to know.

2. syncAgentSkills runs on every shell entry

${lib.getExe config.packages.syncAgentSkills}
${config.pre-commit.shellHook}

Since syncAgentSkills runs unconditionally in the shellHook, it fires on every nix develop / direnv reload. If agent-skills-nix's install script is fast (idempotent copy with early-exit), this is fine. If it becomes noticeable, a lockfile stamp approach (similar to the existing pnpm-lock.yaml check above it) could gate the sync. Not an issue now, just worth revisiting if shell startup feels sluggish later.

📋 Summary

PR looks good overall. 🐇

@coderabbitai @cubic-dev-ai I pushed a follow-up that switches repo-local skill sync to a local-only symlinked agent-skills-nix bundle and updates the validation guidance to rely on hooks/CI for covered checks. Please review the latest commit.

@coderabbitai @cubic-dev-ai I pushed a follow-up that switches repo-local skill sync to a local-only symlinked agent-skills-nix bundle and updates the validation guidance to rely on hooks/CI for covered checks. Please review the latest commit.

@ryoppippi I have started the AI code review. It will take a few minutes to complete.

#!/bin/bash
# Get the latest commit SHA on the PR branch
LATEST_SHA=$(gh pr view 1139 --repo ryoppippi/ccusage --json headRefOid --jq '.headRefOid')
echo "Latest commit: $LATEST_SHA"

# Get the diff of the latest commit
gh api "repos/ryoppippi/ccusage/commits/$LATEST_SHA" --jq '.files[] | {filename: .filename, patch: .patch}'

@ryoppippi Commit f1ccccd looks solid. Here's the breakdown:

✅ nix/agent-skills.nix — "copy-tree" → "link"

Switching structure to "link" means .claude/skills becomes a single directory symlink into the Nix store, rather than a tree of copied files. This is simpler, lighter, and automatically stays in sync — no risk of stale copies. ✅

The new guard:

if [ -d "$target" ] && [ ! -L "$target" ]; then
  echo "$target already exists as a directory." >&2
  echo "Remove it before syncing Nix-managed agent skills." >&2
  exit 1
fi

is exactly right for the migration path. [ -d ] && [ ! -L ] correctly identifies a plain directory (not a symlink), fails fast with an actionable message, and won't trip on an existing symlink from a previous run. ✅

The ${AGENT_SKILLS_ROOT:-$PWD} env override is a nice flexibility touch for non-standard root setups. ✅

✅ .gitignore — trailing slash removed

.claude/skills/ → .claude/skills

This is important for correctness: git's handling of trailing-slash patterns for symlinks-to-directories is inconsistent across versions. The no-slash form reliably matches both the symlink entry and any directory at that path. ✅

✅ Documentation updates

• .claude/CLAUDE.md correctly drops the nix run .#sync-agent-skills command (hook handles it now) and points to nix/agent-skills.nix for wiring details.
• CLAUDE.md validation guidance appropriately relaxed — "run format when formatting may apply, rely on hooks/CI, run typecheck/test manually for behavior changes" is a realistic and accurate model.
• Skill creator and development skill docs are consistent with the new model. ✅

📋 Summary

PR looks good! 🐇

ccusage performance comparison

PR SHA: 00cc77f1b12d
Base SHA: 3e912262be1d

This compares the Rust PR release binary against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 3e912262be1d; PR package: 00cc77f1b12d. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

ccusage performance comparison

PR SHA: 00cc77f1b12d
Base SHA: 3e912262be1d

This compares the PR package against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 3e912262be1d; PR package: 00cc77f1b12d. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

ccusage performance comparison

PR SHA: f1ccccddc9b3
Base SHA: 3e912262be1d

This compares the PR package against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 3e912262be1d; PR package: f1ccccddc9b3. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

ccusage performance comparison

PR SHA: f1ccccddc9b3
Base SHA: 3e912262be1d

This compares the Rust PR release binary against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 3e912262be1d; PR package: f1ccccddc9b3. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.
Peak RSS is measured separately with /usr/bin/time using 1 runs. Lower RSS ratios are better.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.