skills/mukul975/anthropic-cybersecurity-skills/building-devsecops-pipeline-with-gitlab-ci/Gen Agent Trust Hub
building-devsecops-pipeline-with-gitlab-ci
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECREDENTIALS_UNSAFEMETADATA_POISONINGINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external GitLab API endpoints in scripts/process.py, including project paths and vulnerability findings. Ingestion points: GitLab API projects and vulnerabilities endpoints. Boundary markers: No delimiters or ignore instructions are present in the generated report. Capability inventory: The skill writes data to local JSON files and performs network requests via urllib and requests. Sanitization: There is no evidence of filtering or validation for ingested external content, which could allow malicious metadata from a GitLab instance to influence downstream agent processing.
- [CREDENTIALS_UNSAFE]: The scripts/agent.py script is configured to receive sensitive GitLab private tokens through command-line arguments. This method of credential handling is discouraged because arguments are often exposed in system process listings and shell command histories.
- [METADATA_POISONING]: The skill manifest contains conflicting author information, identifying 'mahipal' in the YAML frontmatter but attributing copyright to 'mukul975' in the LICENSE file. This discrepancy is deceptive regarding the skill's origin.
- [EXTERNAL_DOWNLOADS]: The configuration examples in SKILL.md define semgrep rulesets located at semgrep.dev, which is a well-known service for security policy definitions.
Audit Metadata