The Agent Skills Directory

[Prompt Injection] (SAFE): No evidence of instructions designed to bypass agent safety filters or override system-level constraints. The instructional language is focused on task decomposition and workflow management.
[Data Exposure & Exfiltration] (SAFE): The skill references local task management tools (Sibyl) and standard development utilities (pnpm, curl). It does not access sensitive system files (~/.ssh, ~/.aws) or attempt to exfiltrate data to external domains.
[Indirect Prompt Injection] (LOW): The skill is designed to ingest and process information from the local codebase and external task lists to generate implementation plans. This creates a surface for indirect prompt injection if project files contain adversarial instructions.
Ingestion points: Phase 1 and 2 involve searching Sibyl and reading codebase files.
Boundary markers: Absent; the skill does not explicitly instruct the agent to ignore embedded instructions in the files it reads.
Capability inventory: File reading, task creation via sibyl CLI, and local execution of test/build scripts via pnpm.
Sanitization: Not present; the skill assumes the codebase content is trusted for planning purposes.
[Unverifiable Dependencies] (SAFE): References standard Node.js environment tools (pnpm) for verification tasks (linting, testing) but does not trigger the installation of arbitrary or external packages.
[Dynamic Execution] (SAFE): No use of eval(), exec(), or runtime code generation. Mentions of pnpm build or pnpm test are within the context of a developer agent verifying its own work on a local repository.

plan