feature-dev
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes shell commands using
gitandgh(GitHub CLI) to perform version control tasks. Specifically, it stages files, commits changes, and creates pull requests as part of the final development phase. These actions are aligned with the skill's primary purpose and are controlled through user approval gates.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes potentially untrusted data from the user and the codebase, which is shared across a team of agents via a blackboard mechanism.\n - Ingestion points: Data enters the system from
$ARGUMENTS(feature requirements), project source code analyzed by exploration agents, and various sub-agent reports.\n - Boundary markers: The workflow uses a key-value
blackboardto isolate different context types, although it does not document specific escaping of the content within these keys.\n - Capability inventory: The skill can spawn sub-agents, read and write files (via implementation agents), and execute shell commands for git operations.\n
- Sanitization: The skill incorporates five distinct interactive gates (Phase 1, 3, 4, 5, and 6) that require the user to validate findings, select architectures, and approve the implementation and review outcomes, providing strong human oversight.
Audit Metadata