software-factory-mode-2026aki

Pass

Audited by Gen Agent Trust Hub on Nov 3, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to perform extensive shell operations, including environment verification using command -v codex, complex git operations (worktree management, commits, and merges), and Docker container orchestration (managing networks, volumes, and images).
  • [DATA_EXFILTRATION]: The workflow requires the agent to export 'conversation context' (session history) to the local filesystem at .dev/contexts/ and encourages sharing these files or their paths in Pull Request comments. This constitutes an exposure of session data and internal reasoning to the repository environment.
  • [DYNAMIC_EXECUTION]: The skill uses dynamic execution patterns by instructing the agent to spawn sub-agents (e.g., for sanity-review) with specific instructional strings. It also involves executing dynamically generated prompts through a CLI tool (codex exec --ephemeral) to verify model tiers and environment capabilities.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted data from the host repository, such as PR templates, issue descriptions, and ticket documentation, to automate the creation of branch names, commit messages, and PR summaries.
  • Ingestion points: Repository source code, PR templates, and external issue/ticket descriptions.
  • Boundary markers: The instructions do not define explicit delimiters or 'ignore' instructions for content retrieved from these sources.
  • Capability inventory: The agent has broad capabilities including file system write access, git command execution, network operations (pushing/merging), and container management.
  • Sanitization: There is no explicit requirement for the agent to sanitize or validate input from external issue trackers or documentation before using it to generate shell commands or commit metadata.
  • [EXTERNAL_DOWNLOADS]: The workflow references the installation of project dependencies (npm install, pip install) and the synchronization of local branches with remote repositories, which involves fetching external code at runtime.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 3, 2026, 11:59 PM