The Agent Skills Directory

[SAFE]: The skill implements a robust, multi-step facilitator workflow for generating technology comparison reports. The use of 'micro-file architecture' (split across discovery, expectations, research, and scoring steps) ensures that operations are sequential and require user confirmation before proceeding to the next phase.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent surface for indirect prompt injection as it is designed to scan local workspace files and ingest external research data from the web. However, this is consistent with its primary purpose as a technical research facilitator. The risk is mitigated by explicit 'MANDATORY EXECUTION RULES' that prevent autonomous action without user confirmation at each stage, and the use of structured output templates.
[EXTERNAL_DOWNLOADS]: Technical research steps utilize web search to gather current data on technology stacks, architecture patterns, and benchmarks. These operations are transparently disclosed in the workflow and serve the legitimate purpose of providing evidence-based scoring for the trade-off matrix.
[DYNAMIC_EXECUTION]: The workflow allows for the invocation of a secondary skill (bmad-bmm-technical-research) or a bundled fallback to conduct deep-dives. This modular approach is a standard feature of the agent platform and is used here to compartmentalize complex research tasks.
[DATA_EXPOSURE_SAFE]: While the skill scans the user's workspace for markdown files and PRDs, it does so to identify relevant context for the requested analysis. Discovered documents are presented to the user for confirmation before being used as input, preventing unauthorized data access or exfiltration.

trade-off-analysis