The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill aggregates session history and project requirements into a checkpoint.md file designed to be read by the agent in future sessions. This pattern presents an indirect prompt injection surface where data from project files could influence the agent's behavior.
Ingestion points: Conversation history and local PRD files (e.g., 00-master-plan.md) located in the .dev/ directory.
Boundary markers: The generated checkpoint.md uses specific XML-like tags (e.g., <context>, <decisions>, <blockers>) to structure and delimit data.
Capability inventory: The skill uses Bash, git, and Write tools to manage project state and session logs.
Sanitization: Instructions advise the agent to use relative paths and avoid hardcoding secrets, but the automated scripts do not provide content sanitization for ingested text.

dev-checkpoint