The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes scripts (scripts/bootstrap-upstreams.sh) that clone external GitHub repositories (msitarzewski/agency-agents and pbakaus/impeccable) to provide specialist functionality and methodology overlays.
[COMMAND_EXECUTION]: The skill performs various system-level operations, including cloning repositories and using the OpenViking CLI tool to index files for memory-aware execution.
[REMOTE_CODE_EXECUTION]: The .superset/setup.sh script automatically detects and runs various package managers (npm, yarn, pnpm, bun, pip, uv) to install dependencies within worker workspaces.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted project documentation (e.g., DesignSpec.md, ProjectArchitecture.md, and .context/ files) to generate delivery plans and GitHub issues.
Ingestion points: SKILL.md defines several input files including DesignSpec.md and the .context/ directory.
Boundary markers: The templates and instructions do not specify the use of boundary markers or 'ignore' instructions for external data.
Capability inventory: The skill can execute shell commands, clone repositories, and perform network operations via the OpenViking CLI.
Sanitization: No explicit sanitization or validation of the ingested data is documented.
[DYNAMIC_EXECUTION]: Shell scripts within the skill (scripts/bootstrap-openviking.sh, scripts/bootstrap-upstreams.sh, etc.) utilize Python heredocs to dynamically process configuration files and JSON data at runtime.
[DATA_EXFILTRATION]: The setup process (.superset/setup.sh) copies .env files from the repository root to workspace-local directories to ensure environment consistency, which could facilitate the exposure of credentials across different execution contexts.

Swarm Architect