sdlc-agentic-pipeline
Warn
Audited by Gen Agent Trust Hub on Jul 28, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill utilizes several template files, such as
sprint-start.ps1,sprint-close.ps1,add_ssh_key.py, andset-secrets.js, which contain placeholders for highly sensitive credentials including Jira authorization headers, ECS passwords, and GitHub Personal Access Tokens. The documentation instructs agents to populate these files and run them locally, which results in temporary storage of secrets in the file system. - [COMMAND_EXECUTION]: The
add_ssh_key.pyscript usessubprocess.runto execute SSH commands. Additionally, the skill includes several shell and PowerShell scripts (apply-tool-selections.sh,sprint-start.sh) that perform system-level tasks and interact with external APIs. - [REMOTE_CODE_EXECUTION]: During the deployment phase, the
add_ssh_key.pyscript establishes SSH connections to remote ECS instances and executes commands to configure SSH keys. This provides a pathway for remote command execution as part of the DevOps workflow. - [EXTERNAL_DOWNLOADS]: The onboarding documentation (
service-onboarding.md) provides commands to download and install a helper skill (skill-installer) from a GitHub repository associated with the author's organization (CodeArtsAgent). - [PROMPT_INJECTION]: The orchestration logic uses Jira comments as a primary message bus for inter-agent communication. Since agents are instructed to follow directives found in these comments, it creates a surface for indirect prompt injection. An external party capable of commenting on the monitored Jira issues could potentially inject malicious instructions to redirect the pipeline's behavior.
Audit Metadata