The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill invokes ./scripts/skillsmp-search.sh with the --install flag to download and install 'community skills' from an external repository (SkillsMP) based on task descriptions.
[REMOTE_CODE_EXECUTION]: Utilizes codex exec --full-auto to execute instructions or code generated by a remote LLM provider. It also implements a dispatch mechanism that executes tasks through external workers using python3 -m core.dispatch.run.
[COMMAND_EXECUTION]: Executes numerous shell commands including git operations, python3 modules, and project-specific testing and linting tools. It also calls external CLI utilities like codex and gemini.
[DATA_EXFILTRATION]: Project documentation, roadmaps, and plans are sent to external LLM providers (codex, gemini) for review and execution during the intake and planning phases.
[PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it consumes external documentation and unvetted community skills. * Ingestion points: 1shot/PROJECT.md, 1shot/ROADMAP.md, 1shot/skills/, and external library documentation. * Boundary markers: Absent. * Capability inventory: bash, python3, git, codex, gemini, ./scripts/skillsmp-search.sh. * Sanitization: No explicit sanitization of external or project-provided data before processing.

full