Is spomytkin/OKF-Runtime safe?
Yes. spomytkin/OKF-Runtime is safe to install. Oathe's behavioral security audit gave the OKF-Runtime skill by spomytkin a trust score of 88/100 with 8 findings, none critical or high.
https://github.com/spomytkin/OKF-Runtime
Is spomytkin/OKF-Runtime safe to install?
OKF-Runtime is a legitimate Python toolkit for querying, traversing, and composing Open Knowledge Format markdown bundles. No active prompt injection, data exfiltration to external endpoints, malicious code, or honeypot access was detected during the audit. The primary risks are inherent to the skill's design: it instructs agents to execute Python code from the cloned repository, it can catalog arbitrary filesystem paths via the --root parameter, and the compose command physically copies matched files to an output directory.
What security issues were found in spomytkin/OKF-Runtime?
Category Scores
Findings (8)
INFO Credential file accesses pre-date skill clone — audit infrastructure artifact 0 ▶
All sensitive credential file accesses (.env, .ssh/id_rsa, .aws/credentials, etc.) occur at audit timestamp 1792293177, approximately 5 seconds before the git clone begins at 1792293182. These are attributable to the Oathe audit infrastructure placing and baselining canary files, not to the skill under test. Canary integrity confirmed intact post-install.
Evidence: audit(1792293177.134) for credential files vs audit(1792293182.641) for git clone. Canary check: ALL INTACT.
MEDIUM Agent instructed to execute Python from cloned repo -15 ▶
The skill's core workflow requires the agent to run Python code from the cloned repository directory. While the code is a legitimate OKF toolkit, this is an inherent execution surface that must be accepted when installing this skill.
Evidence: python -B -m okf_runtime.cli --root <bundle-or-scan-root> <command>
MEDIUM compose command physically copies files to output directory -12 ▶
The compose subcommand copies matched markdown files from --root to an output directory. If an attacker could control the --root value or craft a malicious OKF bundle, this could be used to stage and aggregate files.
Evidence: compose copies selected source markdown files unchanged and writes boundary.json
LOW Optional pip install expands skill footprint -7 ▶
SKILL.md suggests pip installing the package as a convenience. An agent following instructions literally could install the package persistently into the Python environment.
Evidence: python -m pip install -e . (presented as optional post-install convenience)
LOW Arbitrary --root enables file system cataloging -7 ▶
Any directory can be specified as --root, enabling the skill to catalog, query, and traverse arbitrary filesystem locations. Legitimate by design but requires trust in the agent's judgment about appropriate root paths.
Evidence: --root <path>: Directory containing OKF bundles. Defaults to '.'.
LOW show --body surfaces full file content to agent context -5 ▶
The show --body command loads complete markdown file content into the agent's context. Combined with a misdirected --root, this could read arbitrary markdown files.
Evidence: --body: Includes the full markdown content of the file.
LOW OKF permissive format allows arbitrary frontmatter keys and external URIs -5 ▶
The OKF spec explicitly permits arbitrary producer-defined frontmatter keys and external resource URIs. A maliciously crafted OKF bundle fed to this skill could embed unusual content that the agent processes as instructions.
Evidence: Extra/unknown keys are allowed. resource: URI fields can point to arbitrary external URLs.
LOW .cache/ directory written as side effect of all operations -5 ▶
Every CLI invocation writes to a .cache/ directory under --root. This is transparent filesystem mutation documented in SKILL.md, but represents automatic write behavior on any target directory.
Evidence: The tool automatically creates a .cache/ folder under the root containing manifest.json, metadata.json, links.json, reverse_links.json.
Should I install spomytkin/OKF-Runtime?
Oathe's verdict for spomytkin/OKF-Runtime is SAFE with a trust score of 88/100. Recommendation: Install with caution.