The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies a clear surface for indirect prompt injection by ingesting and acting upon raw, untrusted text data.
Ingestion points: The skill reads elicitation_log.md in Step 1, which contains raw requirements and evidence gathered from various stakeholders.
Boundary markers: The instructions do not define delimiters or explicit "ignore embedded instructions" warnings for the agent when processing these raw inputs.
Capability inventory: The skill has capabilities to read multiple project files and write a comprehensive report to requirements_analysis_report.md.
Sanitization: There is no evidence of text sanitization, validation, or escaping logic to prevent malicious instructions embedded in the raw requirements from influencing the agent's behavior.
[METADATA_POISONING]: The SKILL.md file contains documentation that references a specific local Windows file path (C:\Users\Peter\Downloads\ECC-main\skills\product-capability\SKILL.md).
Evidence: This hardcoded path reveals the author's local username ('Peter') and file structure, which is a potential information disclosure. Additionally, including environment-specific paths in a skill marked as portable: true in its frontmatter is a misleading metadata practice.

04-requirements-analysis