The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess for system interaction. scripts/manage_execution.py executes git commands to infer branch-related IDs and context. lib/workflow_state/parity.py runs npx skills ls -g --json to list globally installed skill metadata for environment discovery.\n- [EXTERNAL_DOWNLOADS]: The use of npx in lib/workflow_state/parity.py to execute the skills package involves a dependency on the npm registry, which may download and run code at runtime if the package is not cached locally.\n- [REMOTE_CODE_EXECUTION]: Implements dynamic module loading via importlib.util on computed paths. lib/workflow_state/inventory.py resolves paths to internal scripts (e.g., manage_proposals.py) and executes them. scripts/manage_execution.py similarly loads scope_runtime.py. This pattern allows for the execution of Python code from paths calculated at runtime.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8).\n
Ingestion points: Data is ingested from .slice-meta.json, registry.json, and slice-traceability.md (markdown tables).\n
Boundary markers: None identified; the skill trusts the structure and metadata of these files to guide its logic without explicit delimiters.\n
Capability inventory: The skill has the capability to delete directories (shutil.rmtree), move files (shutil.move), and execute system commands (subprocess.run) based on state transitions derived from input data.\n
Sanitization: Inputs are checked for basic format (e.g., SLICE_ID_PATTERN), but the logic lacks robust sanitization against instructions embedded in markdown content that could influence decision-making.

guide-execution