The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses several local Bash scripts (scripts/review-package, scripts/sdd-workspace, scripts/task-brief) to automate development tasks. These scripts perform legitimate file system and Git operations, such as creating isolated workspaces, extracting task requirements from markdown plans, and generating diff packages for review. The scripts include safety checks, such as validating Git references before use.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process implementation plans and specifications, which are then interpolated into subagent prompt templates.
Ingestion points: External markdown plan files (PLAN_FILE) and technical specifications.
Boundary markers: The prompt templates (implementer-prompt.md, task-reviewer-prompt.md) use distinct Markdown headers and clear structural markers to delimit task descriptions and context from instructions.
Capability inventory: The skill has access to Git worktree management, local file system writes in a git-ignored directory, and the ability to dispatch subagents with specific roles.
Sanitization: Content is extracted using structured patterns via awk to ensure subagents only receive relevant task sections.
Risk mitigation: The workflow includes independent reviewer subagents for every task, a final whole-branch review, and a mandatory instruction to stop for any security-sensitive actions or destructive operations.

subagent-driven-development