The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions direct the agent to tell the user to run ./install.sh. While this is a user-facing instruction, it promotes the execution of local shell scripts from an external source which can contain arbitrary commands.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read relative skill files (e.g., ../p3-mode/SKILL.md) and "trust it when it disagrees with this map." This is an instruction to allow data from external files to override the primary skill instructions.
Ingestion points: Relative markdown files for sub-skills within the p3-stack repository (SKILL.md files).
Boundary markers: None specified; the agent is told to trust the incoming content directly.
Capability inventory: The skill references related tools that perform file writes (p3-models.md) and thread management (t3_thread_launch).
Sanitization: No sanitization or validation of the referenced files is mentioned.
[PERSISTENCE]: The skill documents an installation process that places links in ~/.agents/skills/. This is a persistence mechanism that modifies the environment to ensure the agent's skills remain available across sessions.
[EXTERNAL_DOWNLOADS]: The skill references and links to a third-party repository (github.com/uzairansaruzi/p3-stack) for the main codebase and documentation.

p3-help