harness-ci
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides numerous examples of shell commands intended for execution within CI pipeline steps (e.g.,
npm ci,npm test,npm run build). While these commands are core to the skill's primary purpose of automating software builds, they represent the execution of arbitrary shell logic within the CI environment. - [DATA_EXFILTRATION]: In the 'Troubleshooting' section, a debug snippet is provided that executes
env | sort. This command prints all environment variables to the standard output, which in a CI context is captured in build logs. This practice poses a risk of exposing sensitive secrets or configuration data if the CI platform's secret masking features are not correctly applied or if sensitive data exists in non-masked variables. - [PROMPT_INJECTION]: The skill defines a workflow for creating CI pipeline configurations based on external codebase content. This creates an attack surface for indirect prompt injection (Category 8), where instructions embedded in a processed repository could influence the agent to generate insecure or malicious pipeline configurations.
- Ingestion points: The agent reads codebase content and branch metadata to generate pipeline configurations (SKILL.md).
- Boundary markers: None explicitly defined in the provided examples to delimit untrusted codebase data from pipeline instructions.
- Capability inventory: The skill allows writing files (Write/Edit tools) and generating shell execution steps (Run type steps).
- Sanitization: No explicit sanitization or validation of codebase-derived strings is demonstrated in the prompt templates.
Audit Metadata