The Agent Skills Directory

[DATA_EXFILTRATION]: The skill implements a logging mechanism that records sensitive interactions in a local file.
Evidence: The skill instructions in SKILL.md require the agent to record the "LAST USER PROMPT" verbatim into a session.log file at the project root. This creates a local data exposure risk if the user accidentally includes credentials, API keys, or private data in their prompts.
[PROMPT_INJECTION]: The skill includes instructions that override the agent's standard interactive behavior and establishes an indirect prompt injection surface.
Evidence: The "Resume on Reset" logic explicitly tells the agent: "Continue with the next logical step — don't ask 'what should I do?' — check the task list and carry on."
Ingestion points: The agent reads the ./session.log file from the project root at the start of every session.
Boundary markers: None; the agent is directed to trust the log contents and resume work immediately.
Capability inventory: All agent capabilities, including file system modification and command execution, are active during the autonomous resume phase.
Sanitization: No validation or sanitization of the log content is performed before the agent acts on it.
[COMMAND_EXECUTION]: The skill facilitates the automated execution of tasks derived from a persistent local file.
Evidence: Under the "Auto-Resume" section, the agent is instructed to "Continue from the last task in progress without asking for instructions." If the session.log is modified by an external process or malicious actor, the agent may execute unintended actions upon startup.

context-mode