The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the subprocess module in its utility and testing scripts to execute external analysis and validation tools.
scripts/signoff_preflight.py runs a narrative linter to ensure that generated report text accurately reflects the underlying statistical data.
scripts/test_render_record_integration.py and other test files invoke Rscript to perform meta-analysis calculations and generate figures.
[DYNAMIC_EXECUTION]: Statistical validation and reporting depend on the dynamic invocation of local Python and R scripts.
Auxiliary scripts are called using sys.executable to ensure they run within the same Python environment as the main skill.
R-based plotting scripts are executed at runtime to produce meta-analysis visualizations such as forest and funnel plots.
[INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface through the ingestion and processing of JSON, YAML, and CSV files generated within the project workspace.
Scripts like scripts/audit_package.py and scripts/registration_reconcile.py parse these artifacts to create comprehensive audit summaries for user review.
To mitigate risks associated with untrusted data, the skill implements a verification mechanism using SHA-256 hashes (defined in scripts/decision_block_util.py) to confirm that the decision points presented to the user have not been altered prior to sign-off.

review-methodology-foundations