The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/run.py determines the repository root directory at runtime by searching parent folders for a marker file (AGENTS.md). It subsequently modifies sys.path to enable the import of local modules from the tooling package. This dynamic path computation for loading executable code presents a security risk if the filesystem structure is manipulated or untrusted.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted content from output/DRAFT.md and outline/transitions.md to identify voice artifacts.
Ingestion points: The text content of the draft and transition files are read into the analysis context in scripts/run.py and are presented to the agent via the Role Prompt in SKILL.md.
Boundary markers: The skill does not implement delimiters or explicit instructions (such as "ignore embedded instructions") within the script logic or the agent role prompt to prevent instructions contained within the analyzed documents from influencing agent behavior.
Capability inventory: The analysis process results in file system writes to output/POST_MERGE_VOICE_REPORT.md and output/QUALITY_GATE.md via internal workspace utilities.
Sanitization: No sanitization, escaping, or validation of the ingested draft text is performed before it is processed by the script or the agent.

post-merge-voice-gate