The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several local shell scripts (scripts/review-package, scripts/sdd-workspace, scripts/task-brief) to manage implementation tasks. These scripts use standard tools such as git, awk, and mkdir to extract requirements from plan files and generate diff packages for review. All operations are confined to the local repository structure and the working tree.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external implementation plans and subagent reports, creating a surface for potential instruction injection.
Ingestion points: The scripts/task-brief utility reads user-provided plan files, and the agent processes implementation reports.
Boundary markers: The agent prompts use clear Markdown section headers to maintain instruction hierarchy.
Capability inventory: The workflow includes the ability to write code and execute local Git commands, though these are gated by review loops.
Sanitization: The skill relies on subagent isolation and mandatory reviewer checkpoints to validate content. This surface is consistent with the primary purpose of a coding assistant workflow.

subagent-driven-development