The Agent Skills Directory

[PROMPT_INJECTION]: Multiple skills within the suite, such as meeting-analyzer, scrum-master, and senior-pm, are designed to process external, untrusted data including meeting transcripts and project-related JSON files. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the data to manipulate the agent's behavior during analysis.
Ingestion points: meeting-analyzer/SKILL.md (transcripts in .txt, .vtt, .srt, .docx, .json), scrum-master/SKILL.md (sprint data), senior-pm/SKILL.md (portfolio data).
Capability inventory: The skill set uses python3 for local script execution and Atlassian MCP tools for live API interactions with Jira and Confluence.
Boundary markers: The instructions lack explicit delimitation or warnings to ignore commands potentially embedded within the ingested data.
Sanitization: No sanitization or validation of the input content is mentioned prior to processing by the agent or the Python scripts.
[COMMAND_EXECUTION]: The skill architecture relies on the execution of local Python scripts (senior-pm/scripts/*.py, scrum-master/scripts/*.py) to perform complex analysis. While the instructions state these scripts utilize only the standard library, the execution of local scripts on data provided by the user is a capability that requires monitoring.

pm-skills