The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill performs local filesystem scanning of specific directories (e.g., docs/, spec/, design/) to identify project artifacts. This access is restricted to common documentation formats and is consistent with the skill's purpose as a project guide. No evidence of unauthorized data transmission or network exfiltration was found.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests content from Markdown and YAML files within the user's repository as defined in references/workflow-map.yaml.
Boundary markers: The skill prioritizes TRACEABILITY-METADATA blocks and specific document headers to distinguish structured data from free-form text.
Capability inventory: The skill is limited to recommending workflow-specific slash commands (e.g., /prd-writer, /test-reviewer). It does not possess capabilities for arbitrary shell execution, file modification, or network requests.
Sanitization: The skill uses strict artifact identification rules (defined in references/artifact-detection.md) to minimize the impact of misleading content in processed files.
[COMMAND_EXECUTION]: The skill identifies and recommends canonical slash commands for the agent to use. These commands are mapped to specific project nodes within the TestAny-io ecosystem. There is no evidence of the skill attempting to execute arbitrary system commands or bypass security restrictions.

guide