The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to install external companions using npx skills add mattpocock/skills. This method downloads and executes code from a repository not listed as a trusted vendor, posing a significant remote execution risk.
[COMMAND_EXECUTION]: The skill incorporates shell commands (npx) to expand its capabilities. Executing such commands programmatically without specific user confirmation for the source repository is a security concern.
[DATA_EXPOSURE]: The skill is designed to read .waypoint/local.yaml to obtain identity metadata (actor_id). This involves direct access to local configuration files, which could contain environment-specific or sensitive information.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from requirement pools, incident reports, and product evidence. It uses this data to generate persistent milestone and feature documentation without adequate sanitization or clear boundary markers to prevent embedded instructions from influencing the agent's behavior.
Ingestion points: requirement pool, incidents, product evidence, repository instructions (SKILL.md).
Boundary markers: None present.
Capability inventory: File system materialization (writing feature/milestone files), package installation/execution (npx).
Sanitization: No sanitization or validation of external content is described.

milestone-planning