Security audit
ontology-clawra
Security checks for vulnerabilities and agentic risk
Overview
The skill is plausibly a local ontology reasoning tool, but inconsistent file-path handling and mixed/inactive network-related code (plus mentions of GitHub pushes) create ambiguity about exactly where it will read/write and whether network actions could be enabled later.
What to check before installing or running this skill: - File-path inconsistency: inspect the code paths that define MEMORY_DIR. Some modules write to ~/.openclaw/skills/ontology-clawra/memory while scripts/ontology-clawra.py writes to BASE_DIR/memory (the skill package folder). Decide whether you are comfortable with the skill writing into its installation directory as well as your home directory, and request the author clarify/standardize the memory path. - Network code present but disabled: network_fetch.fetch_from_network is implemented to avoid external requests today, but the code and comments indicate network capability can be enabled with configuration/user consent. If you need strict offline behavior, either audit/modify network_fetch to permanently remove network code or test in an offline sandbox first. - GitHub push mention: SKILL.md/CHANGELOG mention GitHub push and self-repair/code generation. No push implementation is visible, but the presence of that language is a red flag you should ask the author to explain — specifically, whether the skill can modify remote repos, and whether any code-generation features can run subprocesses or network operations. - Test in a sandbox: run self_eval.py and the scripts in a controlled environment where filesystem writes are monitored to confirm where files are created and what they contain. Confirm that write operations indeed prompt for user confirmation as claimed. - Review persistent files: after a test run, inspect both ~/.openclaw/skills/ontology-clawra/memory and the skill package directory for unexpected files. If you operate in a multi-user environment, ensure file permissions are acceptable. - If you are not comfortable: ask the author to (1) standardize MEMORY_DIR usage to a single explicit path, (2) remove or clearly gate any network/git interactions, and (3) document exact prompts/confirmation flows for writes and for enabling network features. Overall: the skill looks like what it claims to be, but the inconsistent IO targeting and dormant network/git-related code justify caution and further clarification before trusting it with sensitive workspaces.
Static analysis
No suspicious patterns detected.