The Agent Skills Directory

[SAFE]: The skill demonstrates safe operational patterns by managing project state through local markdown files and utilizing a specialized, tool-restricted subagent for research tasks.
[PROMPT_INJECTION]: The skill ingests external data that creates a standard indirect prompt injection surface, though no malicious exploitation was found. It processes user-supplied product descriptions and optionally fetches external web content to verify project references.
Ingestion points: The skill ingests data from user input (product ideas), external web content via the researcher subagent's WebFetch tool, and local project source code during brownfield repository scans.
Boundary markers: While the skill uses structured markdown templates for output, it does not specify explicit delimiters or instruction-bypass warnings when processing external data.
Capability inventory: The skill possesses significant local capabilities, including the Bash tool for repository interaction and Write/Edit tools for file system modification.
Sanitization: No explicit sanitization or filtering logic is documented for the processing of external web content or source code metadata.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform routine development tasks, such as checking git branch status and ahead/behind counts. These commands are executed within a clear context of repository reconciliation and orientation.
[EXTERNAL_DOWNLOADS]: The skill includes functionality to fetch content from the internet for reference verification. This behavior is properly implemented as an optional feature that requires explicit user consent through the 'References' decision panel.

scope