harness-code-integration
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines workflows that ingest external, potentially untrusted data from Harness webhook payloads such as branch names, tag references, and pull request metadata.
- Ingestion points: Trigger configuration in SKILL.md ingests data from webhook payloads via expressions like '<+trigger.payload.ref>'.
- Boundary markers: The provided examples do not demonstrate the use of explicit delimiters or instructions to ignore potential commands embedded in git references or PR titles.
- Capability inventory: The skill utilizes tools such as Bash, Write, and WebFetch, which could be leveraged if untrusted data is processed without sufficient isolation.
- Sanitization: Sanitization is limited to basic string manipulation for tag references and does not include robust validation of all incoming payload fields.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill follows security best practices for secret management by using Harness platform secret references (e.g., '<+secrets.getValue("harness_code_user")>') rather than hardcoding credentials in the connector configuration. It also identifies a standard repository structure that includes a Keycloak realm export for identity management.
- [DYNAMIC_EXECUTION]: The skill documents the execution of shell commands within CI stages for routine development tasks including Helm chart linting, security scanning with Trivy and Checkov, and unit testing. These commands are executed within the user's defined Harness infrastructure.
Audit Metadata