The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple sources including configuration files, user iteration feedback, and external content referenced via URL placeholders. This data is used to build the final agenda and is passed to subsequent editorial tools without explicit sanitization or boundary markers to prevent embedded instructions from influencing agent behavior.
Ingestion points: Project configuration files (DISCOVERY.json, PROJECT.json), user-provided iteration feedback, and external web content referenced via the [Source](url) convention.
Boundary markers: None present in the instructions for processing external content or local configuration data.
Capability inventory: File system access (reading, writing, and deleting files), cross-skill tool invocation, and downstream content fetching.
Sanitization: No explicit sanitization or validation of the ingested data is described.
[COMMAND_EXECUTION]: The skill performs legitimate file system operations as part of its core functionality, including checking directory writability, reading/writing project state files, and executing file deletion routines (via the RESTART-GUARD.md logic) to manage project cleanup.

structure-agenda