subagent-driven-development
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes shell commands for repository management, workspace cleanup, and execution of local helper scripts. Evidence includes the usage of
git rev-parse,git clean -fdx,rm -rf <workspace>, and local scripts such asscripts/sdd-workspace,scripts/task-brief, andscripts/review-packagedescribed inSKILL.mdandEXTENDED.md. - [INDIRECT_PROMPT_INJECTION]: The skill processes implementation plans and specifications provided as external files, which serves as an attack surface for instructions intended to influence subagent behavior.
- Ingestion points: The
SKILL.mdfile defines a process for reading aPLAN_FILEand aSpecto drive the development workflow. - Boundary markers: The process creates 'task briefs' to isolate requirements for individual subagents, providing a context boundary that mitigates accidental obedience to embedded instructions.
- Capability inventory: The skill has the capability to execute shell commands (git, rm), modify the file system via worktrees, and dispatch additional subagents via prompt templates.
- Sanitization: The workflow involves manual extraction of plan text into briefs, but no automated sanitization or escaping of external content is specified.
Audit Metadata