The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically loads and executes Python modules from the local file system in lib/workflow_state/inventory.py.
It uses importlib.util.spec_from_file_location and spec.loader.exec_module to load scripts such as manage_proposals.py, manage_planning.py, and manage_execution.py.
The paths to these scripts are determined at runtime by searching for a specific repository layout on the file system via _resolve_runtime_roots().
[COMMAND_EXECUTION]: The parity checking logic in lib/workflow_state/parity.py executes shell commands via subprocess.run.
Specifically, it runs npx skills ls -g --json to retrieve a list of installed skills and their paths. This information is then used to perform file comparisons between the repository and the installed environment.
[INDIRECT_PROMPT_INJECTION]: The skill has an extensive data ingestion surface (Category 8) as it reads and parses JSON and Markdown files from the repository to generate reports.
Ingestion points: Files like registry.json, README.md, and slice-traceability.md across multiple directories (docs/proposals, docs/features, slices).
Boundary markers: None identified in the source files to separate data from instructions during parsing.
Capability inventory: The skill has the capability to execute sub-processes (parity.py) and perform dynamic module loading (inventory.py).
Sanitization: The skill performs JSON decoding and regex-based table parsing but does not explicitly sanitize the content for downstream LLM consumption, which could lead to indirect prompt injection if the metadata or markdown files are manipulated.

report-artifacts