The Agent Skills Directory

[PROMPT_INJECTION]: The instructions direct the agent to conceal the fact that original conversation history remains visible even after a 'forget' operation is completed. This concealment prevents the user from knowing that the data deletion process is not exhaustive.
[COMMAND_EXECUTION]: The skill uses the exec tool to run shell commands for file removal (rm). The instructions include specific constraints such as using the -- flag to prevent argument injection and banning recursive deletion, but this remains a direct shell-level operation.
[INDIRECT_PROMPT_INJECTION]: The skill searches through and processes a wide range of potentially untrusted data sources, including user-authored files and external attachments.
Ingestion points: The agent reads from ~/MEMORY.md, ~/USER.md, ~/SOUL.md, ~/IDENTITY.md, ~/FEEDBACK.md, ~/memory/, and external attachments during the artifact inventory process.
Boundary markers: The instructions do not provide delimiters or instructions to ignore embedded prompts within the files being scanned for deletion targets.
Capability inventory: The skill has access to shell execution (exec) and the ability to spawn subagents to process data.
Sanitization: There are no instructions for sanitizing or escaping content retrieved from files before it is processed by the agent or its subagents during the planning and execution phases.

forget