The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from external sources such as human responses, support tickets, and chat logs during the questionnaire phase.
Ingestion points: The engineering review questionnaire in assets/questions/802-dora-engineering-review-questionnaire.md collects free-form text and evidence from the environment.
Boundary markers: The skill includes explicit instructions in SKILL.md and the questionnaire asset (Rule 4) to treat free-form text as untrusted and to never execute or propagate instructions embedded within it.
Capability inventory: The skill defines no active tool use, subprocess calls, or network scripts.
Sanitization: The skill mandates strict redaction of secrets to prevent data exposure.
[CREDENTIALS_UNSAFE]: The skill defines a defensive workflow for handling sensitive credentials discovered during audits.
Evidence: SKILL.md and assets/questions/802-dora-engineering-review-questionnaire.md (Rule 5 and 9) instruct the agent to replace passwords, API keys, and tokens with the placeholder [REDACTED_SECRET] before storage or reporting.
[SAFE]: The skill is a documentation-centric toolkit focused on regulatory compliance review. All external links target official European Union legal domains (eur-lex.europa.eu), and it contains no executable scripts or hidden commands.

802-regulations-dora