The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions direct the agent to navigate and read file metadata in ~/.claude/projects/ to identify the current session ID from JSONL log files. Accessing this directory exposes the full history of the user's past interactions with the AI platform to the agent's context.
[COMMAND_EXECUTION]: The skill requires the execution of a bundled Python zipapp (wheypoint.pyz) and various Git CLI commands (such as git status, git rev-parse, and git-worktree) to perform state management and artifact creation. This involves executing code and shell commands shipped with the skill.
[PROMPT_INJECTION]: The skill processes the entire conversation history as its primary input to generate handoff documents. The instructions lack explicit boundary markers or delimiters for this untrusted data, which could allow instructions embedded within the conversation history to influence the agent's behavior during the compaction and checkpointing process. The skill does, however, include specific instructions for the agent to redact sensitive information such as API keys and tokens during this process.

wheypoint