The Agent Skills Directory

[SAFE]: The skill implements a robust framework for managing product data and decisions within a local workspace.
[COMMAND_EXECUTION]: The skill includes a shell script (scripts/init-product.sh) designed to initialize the workspace directory structure. The script uses standard commands like mkdir and cat to create placeholders for product documentation and does not perform any dangerous operations.
[DATA_EXPOSURE]: The skill emphasizes 'number provenance' and strict tracking of evidence sources through IDs (e.g., [S#], FB#). This promotes transparency and prevents the use of unverified data. No patterns for unauthorized data access or exfiltration were found.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as customer feedback and market research. It includes explicit 'Habits' and 'Non-negotiables' that require the agent to use verbatim quotes, cite source IDs, and verify quote integrity, which serves as a mitigation against indirect prompt injection by maintaining clear boundaries between external data and agent instructions. Ingestion points include discovery/feedback-log.md and discovery/sources.md, while capabilities are limited to reading and writing within the designated product workspace.

product-manager