The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content such as card titles, descriptions, and comments retrieved from external project management platforms. This content is used to inform triage recommendations and generate status reports, which could be exploited to manipulate the agent's behavior.
Ingestion points: Project card data is pulled from Notion, GitHub, Plane, Linear, and Obsidian via their respective adapters in references/adapters.md.
Boundary markers: The instructions lack specific delimiters or guardrails to ensure that the agent ignores any embedded instructions found within the project data it reads.
Capability inventory: The skill possesses the capability to execute shell commands (curl, gh), write to local configuration files (e.g., boards.md), and access sensitive environment variables for authentication.
Sanitization: No sanitization, validation, or filtering of the external data is performed before the agent processes it.
[COMMAND_EXECUTION]: The skill facilitates the execution of shell commands using curl and the gh CLI to interact with various third-party APIs. While these are documented integrations for project management, the use of command execution to handle data from external sources increases the potential impact of an injection attack.

pm