The Agent Skills Directory

[PROMPT_INJECTION]: No prompt injection patterns were detected. The instructions do not attempt to override agent safety protocols or bypass system constraints.
[DATA_EXPOSURE_AND_EXFILTRATION]: No unauthorized data access or exfiltration patterns were found. The skill uses curl for local API assertions and environment verification, which is consistent with its stated purpose. Access to source code and specs is limited to local retrieval tasks.
[OBFUSCATION]: No obfuscation, multi-layer encoding, or hidden characters were detected in any of the skill files.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill does not download or execute remote code from untrusted sources. It relies on pre-installed local tools like rg, git grep, curl, and agent-browser for verification tasks.
[PRIVILEGE_ESCALATION]: No privilege escalation attempts were found. The skill does not use sudo or modify system permissions.
[PERSISTENCE_MECHANISMS]: No persistence mechanisms, such as modifications to shell profiles or cron jobs, were detected.
[METADATA_POISONING]: Metadata fields are used correctly and do not contain deceptive instructions or hidden payloads.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Reads PRD specs and codebase files (SKILL.md).
Boundary markers: None explicitly defined for untrusted data input.
Capability inventory: File system writes (contract artifact), search operations (rg, git grep), and verification executions (curl, agent-browser).
Sanitization: No explicit sanitization or validation of the content of the specs or code is described.
[TIME_DELAYED_OR_CONDITIONAL_ATTACKS]: No time-based or environment-specific triggers for malicious behavior were found.
[DYNAMIC_EXECUTION]: The skill generates and executes smoke tests via curl and agent-browser based on the integration contract. This behavior is standard for the skill's purpose and does not involve executing arbitrary code from untrusted network sources.

integration-contract