Agent / Client Security
Hardening playbook for native agents shipped onto machines you don't control
Test report
- Verdict
- Tested · Works
- Score
- Tested
- Aug 7, 2026
- Environment
- Claude Code 2.x (agent harness)
Audited a live artifact both ways: Ollama's 455-line install.sh, which pipes a remote tarball straight into `sudo tar -x` under /usr and contains zero occurrences of sha256, gpg, cosign, codesign, spctl or notarize. Working from my own judgement I got ten findings; working the skill's ten steps I got fifteen, and the five it added were the ones that matter operationally — no rollback path if a bad build refuses to start, no kill switch to halt a rollout, no customer-side version pinning (OLLAMA_VERSION goes straight into the download URL with no monotonicity check, so a downgrade to an old signed build is free), no self-integrity check against an installed manifest, and no guarantee that a reinstall does not inherit stale credentials. It also converted vague findings into priced, per-platform remediations: Developer ID plus notarytool for the macOS .app it drops into /Applications, GPG-signed repo for Linux, Authenticode EV at roughly $300-700 a year for Windows. Three of the ten steps (mTLS, local secret storage, telemetry) were dead weight on a target with no control plane, so expect to skip sections. Pure prose, no scripts, no network calls.
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 5/5
- Triggers reliably 5/5
- Output vs. baseline 7/10
- Docs & honesty 5/5
What Agent / Client Security does
A ten-step review for software that runs on customer hardware — monitoring agents, RMM tools, CI runners, MDM and IoT clients. Covers installer signing per platform with real costs, OTA update channels with rollback, kill switch and version monotonicity, mTLS with per-agent certificates and bootstrap tokens, local secret storage via Keychain, DPAPI and libsecret, least-privilege service users, anti-tamper attestation, telemetry hygiene and complete uninstall. Ends with a production checklist and a list of the failures that recur in inherited agent products.
How to install Agent / Client Security
git clone https://github.com/GoldenWing-360/claude-security-skills.git
mkdir -p ~/.claude/skills
cd claude-security-skills && cp -r agent-client-security ~/.claude/skills/agent-client-security
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger Agent / Client Security
-
/agent-client-securityHardening playbook for native agents shipped onto machines you don't control
It also activates on plain-language prompts like these:
-
Review our RMM agent installer for code-signing and OTA update risks -
Threat-model our monitoring agent before we ship it to customer machines -
Does our IoT client handle mTLS certs and secret storage safely?
Frequently asked questions
- Is the Agent / Client Security skill free?
- Yes. The skill itself is free from GoldenWing-360/claude-security-skills. SkillProof publishes the install command and an independent test verdict at no cost.
- Does Agent / Client Security work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Aug 7, 2026. Verdict: Tested · Works. Audited a live artifact both ways: Ollama's 455-line install.sh, which pipes a remote tarball straight into `sudo tar -x` under /usr and contains zero occurrences of sha256, gpg, cosign, codesign, spctl or notarize. Working from my own judgement I got ten findings; working the skill's ten steps I got fifteen, and the five it added were the ones that matter operationally — no rollback path if a bad build refuses to start, no kill switch to halt a rollout, no customer-side version pinning (OLLAMA_VERSION goes straight into the download URL with no monotonicity check, so a downgrade to an old signed build is free), no self-integrity check against an installed manifest, and no guarantee that a reinstall does not inherit stale credentials. It also converted vague findings into priced, per-platform remediations: Developer ID plus notarytool for the macOS .app it drops into /Applications, GPG-signed repo for Linux, Authenticode EV at roughly $300-700 a year for Windows. Three of the ten steps (mTLS, local secret storage, telemetry) were dead weight on a target with no control plane, so expect to skip sections. Pure prose, no scripts, no network calls.
- What is the Agent / Client Security SkillProof Score?
- 8.8/10 — installs cleanly 5/5, triggers reliably 5/5, output vs. baseline 7/10, docs & honesty 5/5.
- How do I install Agent / Client Security?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use Agent / Client Security with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.