The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts (scripts/review-package, scripts/task-brief, scripts/sdd-workspace) to manage development tasks. These scripts use standard system utilities like git, awk, and mkdir to automate task isolation and diff generation within the local repository.
[PROMPT_INJECTION]: The skill is designed to ingest and process external "implementation plans," creating a surface for indirect prompt injection (Category 8). Malicious instructions embedded in a plan could attempt to override the subagent's implementation or review logic.
Ingestion points: Implementation plan files (parsed via scripts/task-brief) and global constraint definitions.
Boundary markers: Subagent templates use structured Markdown headers (e.g., ## Task Description, ## What Was Requested) to isolate external content.
Capability inventory: Implementer subagents have capabilities for file modification, git operations, and test execution; reviewers have read access to diffs and the codebase.
Sanitization: No explicit content sanitization is performed; the skill relies on the structural isolation provided by prompt headers.
[SAFE]: No remote downloads, hardcoded credentials, or obfuscated segments were identified. The workflow is restricted to the local filesystem and standard developer tooling.

subagent-driven-development