The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted sprint data, such as story titles and descriptions, from JSON, CSV, and YAML files. This content is interpolated into the agent's context when generating reports, creating a surface for indirect prompt injection where malicious instructions in the data could influence the agent's behavior.
Ingestion points: parse_input.py processes user-supplied sprint exports.
Boundary markers: Absent; the data is mapped into internal objects and directly formatted into output strings.
Capability inventory: The skill has network capabilities (POST requests) via notify_channels.py and file-read capabilities through the agent's environment.
Sanitization: Absent; the skill normalizes the data structure but does not sanitize text content for potential injection sequences.
[DATA_EXFILTRATION]: The notify_channels.py module allows the agent to send sprint metrics and risk summaries to external Slack and Microsoft Teams webhooks. While these notification channels are user-configured via environment variables or a configuration file, the skill performs network POST operations to these external services.
[EXTERNAL_DOWNLOADS]: The skill imports the yaml (PyYAML) library in parse_input.py and notify_channels.py, which is not part of the standard Python library. The skill does not provide a requirements.txt or similar dependency manifest, although it does include a fallback mechanism to JSON in case the library is missing.

scrum-master-agent