The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an automated workflow that processes repository-managed Markdown files, creating an attack surface for indirect prompt injection.\n
Ingestion points: Data enters the system via feature_records/*.md files which are scanned by the generated scripts/check-change-contracts.sh utility.\n
Boundary markers: The template in references/contract-system-implementation-details.md uses standard Markdown headers for sections like Proposed Behavior and Motivation, but provides no guidance on isolating these sections from instructions that might be embedded by an adversary.\n
Capability inventory: The generated scripts have the ability to fail validation checks (influencing process flow) and modify the local filesystem by moving records and updating state fields.\n
Sanitization: The implementation focuses on structural validation of fields and lifecycle states; it does not include sanitization or escaping mechanisms for the natural language content within the feature records.\n- [DYNAMIC_EXECUTION]: The skill generates and implements shell scripts (scripts/check-change-contracts.sh and scripts/set-feature-record-lifecycle.sh) to automate the contract system workflow. While these are generated from established patterns and intended for repository maintenance, the creation of executable content at runtime is a monitored behavior.

development-contract-system