The Agent Skills Directory

[SAFE]: The skill's behavior is entirely consistent with its stated purpose of generating QA plans from project files.
[COMMAND_EXECUTION]: Utilizes standard file system tools (Read, Write, Glob, Grep) to process local data. These operations are transparently defined and scoped to the project directories for legitimate documentation analysis.
[PROMPT_INJECTION]: Instructions regarding the authority of the 'Type' field provided by users are functional constraints for the specific QA classification task and do not represent a safety bypass or malicious injection pattern.
[DATA_EXFILTRATION]: No network-capable tools are used, and all data processing is confined to the local project environment.
[PROMPT_INJECTION]: The skill processes untrusted project documentation, which creates a surface for indirect prompt injection. This is mitigated by the structured extraction process and mandatory human oversight.
Ingestion points: Reads story files and GDDs from directories such as production/sprints/ and design/gdd/.
Boundary markers: Extracted content is categorized into structured internal representations (e.g., Acceptance Criteria, Formulas) before being reported back.
Capability inventory: Has Write access to project directories for generating the QA plan and updating story files.
Sanitization: No explicit content sanitization is performed; the skill relies on structural parsing and human-in-the-loop confirmation via AskUserQuestion.

qa-plan