The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions that attempt to override the agent's or analyzer's reporting logic by pre-emptively dismissing potential security findings.\n
Evidence: resources/error-playbook.md explicitly commands the agent to 'Retract the finding' and 'Do NOT raise this as critical / security / medium / low at any severity' if it detects issues with the project's non-standard proxy.ts convention.\n
Evidence: The skill uses aggressive, coercive language in SKILL.md ('touch it and you die') to enforce specific architectural patterns and discourage reporting of deviations by the agent.\n- [INDIRECT_PROMPT_INJECTION]: The skill performs automated reading and writing of frontend codebase files, which constitutes a potential vector for indirect injection from untrusted source code or project assets.\n
Ingestion points: The agent is designed to inspect existing patterns in components, routes, hooks, and styles across the codebase (SKILL.md).\n
Capability inventory: The skill has the authority to write new code (WRITE), execute shell commands for linting and testing (CALL_TOOL), and interact with network registries (NETWORK).\n
Boundary markers: While the skill establishes structural flows, it lacks explicit negative constraints to prevent the execution of instructions found within the data it processes (e.g., instructions hidden in code comments).\n
Sanitization: The skill recommends using zod for API contracts, which validates data structure but does not necessarily sanitize content against prompt injection.\n- [EXTERNAL_DOWNLOADS]: The skill integrates with external component registries and library repositories.\n
Evidence: Mentions calling the shadcn registry tools to fetch and install UI components (SKILL.md). These are well-known developer services and are treated as safe within the context of frontend development.

oma-frontend