The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The framework includes logic to automatically bootstrap its own environment by installing Python dependencies at runtime.
_cli/cli/dependencies.py and _shared/tools/yaml_compat.py use subprocess.run to call pip install for required libraries.
[EXTERNAL_DOWNLOADS]: The skill fetches software packages and system utilities from external repositories.
_shared/tools/dependency-utils.sh contains routines to install system packages using brew, apt-get, apk, dnf, or yum based on the detected OS.
[COMMAND_EXECUTION]: The core functionality of the skill relies on executing system-level commands and external binaries.
_cli/cli/orchestrator/agent_caller.py executes the claude CLI to invoke sub-agents, passing them generated prompts and controlling their tool access.
_scripts/hooks/workflow_hooks.py runs project-specific test suites using pnpm test or pytest via subprocess calls.
_shared/tools/dependency-utils.sh uses sudo on Linux distributions to execute package management commands with root privileges.
[PROMPT_INJECTION]: The skill creates a significant surface for Indirect Prompt Injection due to its data ingestion and tool capability mix.
It ingests untrusted data from the web (via WebFetch) and from local codebase files (via Read).
This data is processed by agents that possess powerful tools including Bash (arbitrary command execution) and Write (filesystem modification), creating a risk that malicious instructions in analyzed data could influence the agent's actions.

tasks