Back to plugin

Security audit

ClawPhylax — did that actually work?

Security checks for vulnerabilities and agentic risk

Overview

ClawPhylax is a disclosed local security/diagnostics plugin that records tool and network metadata on the machine and is observe-only by default.

Install only if you are comfortable with a local plugin observing agent network metadata and tool outcomes, storing them in a local SQLite ledger, and adding diagnostic guidance to agent behavior. Leave it in default observe mode unless you intentionally want denylist or secret-send blocking; review the separate breed-two-agents skill before using it because it can create and register new agent workspaces.

SkillSpector was not run because this plugin release contains no bundled skills.

Static analysis

Detected: suspicious.dangerous_exec, suspicious.prompt_injection_instructions

Shell command execution detected (child_process).

Critical
Code
suspicious.dangerous_exec
Location
skill/breed-two-agents/scripts/breed.mjs:338
Evidence
const root = spawnSync(process.platform === "win32" ? "npm.cmd" : "npm", ["root", "-g"], { encoding: "utf8", shell: process.platform === "win32" }).stdout.trim(...

Prompt-injection style instruction pattern detected.

Warn
Code
suspicious.prompt_injection_instructions
Location
README.md:101
Evidence
**Prompt guidance.** The `/phylax` command contributes two sentences to the agent's system prompt: verify before installing, ask the outlook before retrying, ru...