Claude Code Patchkit — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Claude Code Patchkit (Plugin) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Community-contributed binary patches for Claude Code CLI — unlock hard-coded limitations without waiting for upstream changes.
Disclaimer: This is unofficial and not supported by Anthropic. It modifies the Claude Code binary. Use at your own risk. Keep backups and know how to reinstall Claude Code.
<img width="1264" height="516" alt="image" src="https://github.com/user-attachments/assets/dbcb6e46-e1d4-4f9e-9708-1afbedae3efe" />
npx skills add huybuidac/claude-code-patchkit -gThe -g flag installs globally (user-level) so the skill is available in all projects. Without it, the skill is only available in the current project./plugin marketplace add huybuidac/claude-code-patchkit
/plugin install claude-patchmacOS / Linux:
codesign (macOS only — ships with macOS)Windows 10/11:
Both:
Inside a Claude Code session:
/claude-patch # Interactive mode
/claude-patch apply subagent-model # Apply specific patch
/claude-patch revert subagent-model # Revert to backup
/claude-patch status # Show all patch states| Patch | Description | Risk |
|---|---|---|
| subagent-model | Unlock model param on Agent tool — use any model id per-call | Low |
Claude Code is distributed as a bun-compiled binary with embedded JS bundles. Some behaviors are locked behind hard-coded Zod schemas. This is an Agent Skill — Claude reads the patch definitions and executes the procedure interactively with your confirmation at each step. It applies safe, length-preserving binary patches with:
<binary>.bak.<timestamp>HashMismatch (binary still runs)Note: the bun-compiled binary embeds the JS bundle 1 or 2 times depending on platform/version (macOS ≤ 2.1.132 = 2, macOS ≥ 2.1.133 = 1, Windows = 1). Patch definitions detect the count dynamically rather than asserting a fixed number.
claude.exe is locked while running, so apply/revert use a rename-swap pattern:
.patching sidecar at the known offsetRename-Item the live claude.exe to .replacing.<ts> (same-volume rename works on a running .exe)Move-Item the patched copy into place*.replacing.* after exitskills/claude-patch/patches/TEMPLATE.md.describe() text, not minified variable names)<binary>.bak.<timestamp> next to the originalcp <binary>.bak.<timestamp> <binary> && codesign --force --sign - <binary> (skip codesign on Linux)Rename-Item claude.exe claude.exe.broken; Move-Item claude.exe.bak.<ts> claude.exe. The backup retains the original Authenticode signature.*.replacing.* files after Claude Code has exited (they may stay on disk while the old process is still mapped)/claude-patch revert flow will detect this and offer in-place reverse-patch instead.~/.local/share/claude/ on Unix, %USERPROFILE%\.local\bin\claude.exe on Windows) and re-run the installerMIT
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.