sdlc-agentic-pipeline

Warn

Audited by Gen Agent Trust Hub on Sep 23, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATIONPERSISTENCE
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill uses utility scripts (apply-tool-selections.ps1 and apply-tool-selections.sh) that dynamically rewrite the YAML frontmatter of agent definition files (pm-agent.md, backend-agent.md, etc.). This process modifies the active instruction set and tool permissions of the agents at runtime based on the contents of .codeartsdoer/tool-selections.json.
  • [DYNAMIC_EXECUTION]: The apply-tool-selections.sh script employs a Python heredoc (PYEOF) to execute dynamically generated logic for parsing and rewriting agent instructions, which is a form of runtime code assembly.
  • [CREDENTIALS_UNSAFE]: Multiple templates in references/templates/sprint-scripts/ (e.g., sprint-start.sh, sprint-close.ps1) require users to populate them with sensitive values like <AUTH_HEADER>. The skill's documentation explicitly notes that these scripts will contain active auth tokens and advises users to delete them after execution, acknowledging a window where credentials reside in plain text on the file system.
  • [INDIRECT_PROMPT_INJECTION]: The skill uses Jira comments as an asynchronous message bus for inter-agent communication (documented in references/branch-strategy.md).
  • Ingestion points: Agents fetch instructions and status updates via atlassian-rovo-mcp_searchJiraIssuesUsingJql in references/agents/pm-agent.md and references/agents/shared/developer-agent-base.md.
  • Boundary markers: The skill lacks explicit boundary markers or instructions to ignore potential adversarial injections embedded in Jira task descriptions or comments.
  • Capability inventory: Agents have extensive capabilities including shell execution (bash), file system writes (write, edit), and network operations via GitHub and Huawei Cloud MCPs.
  • Sanitization: There is no evidence of sanitization or validation of the content retrieved from Jira before it is processed by the agents to determine next steps.
  • [PRIVILEGE_ESCALATION]: PowerShell templates (sprint-start.ps1, sprint-close.ps1) use the -ExecutionPolicy Bypass flag. While common for automation, this explicitly instructs the system to ignore local security policies for script execution.
  • [PERSISTENCE]: The references/templates/add_ssh_key.py script appends a public key to the ~/.ssh/authorized_keys file on a remote Huawei Cloud ECS instance. While intended for deployment setup, this is a standard persistence mechanism that maintains access across sessions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 23, 2026, 02:59 AM