The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes content from remote task trackers, including task bodies and comments, which are untrusted external data sources. Maliciously crafted instructions embedded within a remote task could potentially influence the agent's behavior during the decomposition process.
Ingestion points: The skill fetches the full body and comments of referenced remote tasks and reads remote backlog entries as specified in SKILL.md and references/project-documents.md.
Boundary markers: Instructions are absent regarding the use of delimiters or 'ignore' directives to separate fetched external content from the agent's core logic.
Capability inventory: The agent can read local source code, create or update markdown files across the repository (e.g., in documentation/ subdirectories), modify the AGENTS.md index file, and publish new tasks to remote trackers.
Sanitization: The skill does not provide procedures for sanitizing, escaping, or validating the text retrieved from remote sources before it is used to generate local documentation or influence task decomposition.

taskify