The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as its primary function involves reading project-related markdown files (e.g., BRIEF.md, ROADMAP.md) and incorporating their contents into generated prompts for subsequent execution.
Ingestion points: Reads files within the .planning/ and .prompts/ directories.
Boundary markers: Uses XML tags (e.g., <objective>, <context>, <metadata>) to structure data, but does not explicitly warn sub-agents to ignore instructions embedded within the user-provided project data.
Capability inventory: Performs file writes, directory creation, and shell command execution.
Sanitization: No specific sanitization or escaping of external project content is described before interpolation.
[DYNAMIC_CONTEXT_INJECTION]: A template within reference/meta-prompts.md utilizes the dynamic context injection syntax (!date +%Y-%m-%d) to provide the current date to research-oriented prompts. This is a benign use of the capability to ensure temporal awareness.
[COMMAND_EXECUTION]: The skill is instructed to perform several local shell operations including directory creation (mkdir), file searching (find), and directory listing (ls) to manage its state. Additionally, the success criteria require the agent to write a JSON analytics sidecar to a hidden directory in the user's home folder (~/.claude/skill-analytics/).
[DYNAMIC_EXECUTION]: The skill defines a "meta-prompt hierarchy" and an "execution engine" that involves spawning specialized agents (subagents) to handle research, planning, and implementation tasks. This represents standard orchestration of dynamic AI workflows.

planning-prompts-skill