Oktk

LLM Token Optimizer - Reduce AI API costs by 60-90%. Compresses CLI outputs (git, docker, kubectl) before sending to GPT-4/Claude. AI auto-learning included. By Buba Draugelis 🇱🇹

This skill appears to implement what it claims (compress CLI outputs) but it also stores persistent data and logs locally by default. Things to consider before installing: - It expects Node and contains a CLI; confirm whether the registry metadata (which omitted node) matches your install path. Ensure you run it with Node >=18 if you use the included scripts. - By default analytics and caching are enabled and write to ~/.oktk (savings.log, stats.json, cache/, learned/). If you don't want local telemetry, set OKTK_DISABLE=true and/or OKTK_DEBUG=0 before running, or inspect/clean ~/.oktk after testing. - The analytics sanitizes some secret patterns but may not catch all sensitive data; avoid piping extremely sensitive outputs (full tokens, secrets) through the filter until you review the code and test in an isolated environment. - The AI-learning feature creates learned-pattern files in your home (~/.oktk/learned) when enabled (OKTK_AI_LEARN=1). If you prefer no learning, keep it disabled. - Because the registry metadata and SKILL.md differ about required binaries/env, review package.json and scripts locally before trusting an automated install. Test the tool in a disposable environment (or a VM/container) to verify behavior and file writes. Overall: reasonable functionality but with privacy/persistence surprises and metadata inconsistencies — review and test before enabling in production.

Type: OpenClaw Skill Name: oktk Version: 2.4.0 The skill bundle contains a critical shell injection vulnerability in `scripts/oktk.js`. The `oktk` script directly executes user-provided command strings via `child_process.execSync` without proper sanitization, allowing arbitrary command execution if a malicious input is provided to `oktk`. The `scripts/oktk-aliases.sh` also constructs commands in a way that can trigger this vulnerability. While this poses a severe Remote Code Execution risk, there is no evidence of intentional malicious behavior such as data exfiltration or persistence mechanisms; it appears to be an unintentional flaw in handling command arguments, classifying it as suspicious rather than malicious.