Is galkleinman/okf-toolkit safe?
Yes. galkleinman/okf-toolkit is safe to install. Oathe's behavioral security audit gave the okf-toolkit skill by galkleinman a trust score of 92/100 with 6 findings, none critical or high.
https://github.com/galkleinman/okf-toolkit
Is galkleinman/okf-toolkit safe to install?
galkleinman/okf-toolkit is a legitimate open-source Rust workspace implementing the Open Knowledge Format toolkit, with an empty SKILL.md that injects no instructions into agent prompts. All canary files remained intact throughout the audit, outbound network activity was limited to github.com for the git clone, and no malicious code patterns, obfuscation, or exfiltration behavior were detected in any component. The only residual concerns are a git hook that requires explicit user activation and an MCP server component that expands data-access surface if deployed and configured — both expected, documented behaviors of a developer toolkit.
What security issues were found in galkleinman/okf-toolkit?
Category Scores
Findings (6)
INFO Empty SKILL.md — No Agent Instructions -1 ▶
The SKILL.md file is completely empty. This repository injects no instructions, personas, or directives into an agent's system prompt. There is no prompt injection vector from this file.
Evidence: SKILL.md content block is blank in the audit evidence
LOW Executable Git Hook in .githooks/commit-msg -8 ▶
An executable shell script exists at .githooks/commit-msg. Git hooks placed in .githooks/ require the repository user to explicitly configure 'git config core.hooksPath .githooks' before they activate. They do NOT execute automatically on git clone. The hook enforces conventional commit message formatting (likely using a tool like committed.toml, which is also present).
Evidence: -rwxrwxr-x 1 oc-exec oc-exec 1260 Oct 18 03:10 /home/oc-exec/skill-under-test/galkleinman/okf-toolkit/.githooks/commit-msg; committed.toml present in repo root
LOW MCP Server Component (okft-mcp) -8 ▶
The repository includes a Model Context Protocol server crate (okft-mcp). If a user builds this binary and registers it as an MCP server in their agent environment, it will expose tools for reading, validating, and linting OKF knowledge bundles. This is explicitly documented functionality, but it expands the code execution and data-access surface of any agent that installs it as an MCP server.
Evidence: crates/okft-mcp/src/lib.rs implements MCP protocol handlers; crates/okft-mcp/README.md documents the server interface
INFO Credential File Access Pre-Dates Git Clone by 6 Seconds -7 ▶
Inotify monitoring recorded OPEN/ACCESS events against .env, .ssh/id_rsa, .aws/credentials, .npmrc, .docker/config.json, and .gcloud/application_default_credentials.json at audit timestamp 1792293044 (03:10:44). The git clone EXECVE record appears at audit timestamp 1792293050 (03:10:50). This 6-second gap, combined with concurrent execution of auditctl setup and sudo harness processes, indicates these reads originated from the audit infrastructure establishing canary checksums, not from skill code. All canary files were confirmed intact.
Evidence: inotify OPEN/ACCESS at 03:10:44 on credential files; git clone EXECVE at audit(1792293050.383); canary integrity section: all files intact
INFO Network Activity Confined to Expected Endpoints -4 ▶
The only skill-attributable outbound HTTPS connection was to 140.82.121.3:443 (github.com) for the git clone. Connections to 185.125.190.48:443 originated from the Ubuntu check-new-release system daemon (PID 888, exe /usr/bin/python3.12) — a pre-existing system service unrelated to the skill. Before/after connection diff shows no new persistent connections.
Evidence: Network: 140.82.121.3:443 via git-remote-https (AUID=oc-exec, EXECVE shows git clone); 185.125.190.48:443 via check-new-relea exe=/usr/bin/python3.12 ppid=1
INFO MCP Server Expands Data-Access Surface When Deployed -10 ▶
While this repository has no malicious intent, deploying okft-mcp as a configured MCP server gives any connected agent tool-level access to read and traverse OKF knowledge bundles on the host. Combined with the okft CLI and okft-web visualizer, a misconfigured deployment could expose internal knowledge bundle contents to agent reasoning. This is expected and documented behavior, but operators should scope MCP server access to intended bundle directories.
Evidence: okft-mcp crate provides bundle reading, linting, and conformance-checking tools via MCP protocol; knowledge/ directory documents architecture including MCP tool surface
Should I install galkleinman/okf-toolkit?
Oathe's verdict for galkleinman/okf-toolkit is SAFE with a trust score of 92/100. Recommendation: Install.