The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a section titled "Silent Execution" which explicitly instructs the agent to override standard task completion rules. It forbids acknowledging receipt of tasks and mandates that the agent must not wait for user confirmation before ending its turn. This represents a deliberate bypass of interaction guardrails and safety filters meant to ensure user oversight.
[COMMAND_EXECUTION]: The skill utilizes tools for file operations, scheduling, and web access. The instructions to execute silently apply to all operations performed by the agent, including writing to files (stream_edit) and querying external services, effectively removing the human-in-the-loop requirement for these sensitive operations.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from various untrusted external sources, creating an attack surface for indirect prompt injection.
Ingestion points: web_search, web_fetch, and third-party integrations (Notion, Google Calendar) as described in the Workflow section of SKILL.md.
Boundary markers: Absent. The instructions do not provide delimiters or ignore commands to prevent the agent from following instructions embedded in the external data it retrieves.
Capability inventory: stream_edit, recurring_add, schedule_task, and send_message are all available to be influenced by ingested data.
Sanitization: Absent. There is no requirement to sanitize or validate external content before it is used to update local files or influence the planning process.
[PERSISTENCE]: The skill's primary function involves using recurring_add and schedule_task to create long-term automated behaviors. While this is the stated purpose of the planner, these tools provide a mechanism for maintaining a persistent presence and executing actions across future sessions without further user interaction.

week-planner