perfguard

Performance anti-pattern scanner — finds N+1 queries, sync I/O, missing pagination, and memory leaks before they hit production

PerfGuard appears to implement a local, grep-based performance scanner and an optional git pre-commit hook installer; this is coherent with its description. Before installing: - Decide whether you want a tool that can modify your repository's lefthook.yml and run code on staged files (hooks installation appends or copies config and runs lefthook install). Review the config/lefthook.yml and how the hook sources the skill scripts to ensure it points to the intended skill installation path. - Understand license behavior: PERFGUARD_LICENSE_KEY unlocks Pro/Team features and is the declared primary credential. The offline JWT verification does not cryptographically verify tokens unless CLAWHUB_JWT_SECRET is set (an undeclared env var). That means, without the secret, the scripts perform only structural checks on the token; this weakens license enforcement and could allow local bypass of paid checks. If you rely on the signature check, set CLAWHUB_JWT_SECRET in a secure location. - The skill may use optional local tools if present (node, openssl); those are not declared as required. Ensure your environment's node/openssl behavior is acceptable. - The tool claims 'zero telemetry' and 'no external uploads' — the code is local and contains no outgoing network calls, but always inspect any hooks you install and the skill path to be sure you're sourcing the expected scripts. If these caveats are acceptable and you trust the skill source/homepage, the tool is functionally coherent; if you need stronger license guarantees or want to avoid repo modifications, proceed cautiously or request changes (declare CLAWHUB_JWT_SECRET, make signature verification mandatory, or require explicit consent before modifying lefthook.yml).

Type: OpenClaw Skill Name: perfguard Version: 1.0.2 The skill contains a critical command injection vulnerability in `scripts/license.sh`. The `extract_field` and `decode_jwt_payload` functions interpolate decoded JWT payload data (user-controlled via the license key) directly into `python3 -c` and `node -e` execution strings without sanitization, allowing for arbitrary code execution via a crafted license key. Additionally, the `show_trend` command in `scripts/analyzer.sh` performs automated `git checkout` operations on historical commits, which is an invasive operation that could trigger unexpected side effects or git hooks in certain repository environments.