The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute local Python scripts (scripts/integrity_audit.py, scripts/structured_review.py, and scripts/revision_audit.py) to automate integrity checks, logic audits, and comparative reviews between original and revised drafts. These commands are triggered at the pre-LaTeX gate and during the logic transfer audit phase.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data, specifically user manuscripts, draft sections, and notes, which are then used to generate substantive rewrites. This data ingestion creates an attack surface for indirect prompt injection if the user content contains adversarial instructions.
Ingestion points: External data enters the agent context through the draft_path, user draft text, and configuration files like paper_rewriting_output/research_dossier.md.
Boundary markers: The skill uses markdown headers and structured table formats to separate user data from instructions, which provides some structural delimitation.
Capability inventory: The agent has the capability to execute shell commands (running Python scripts) and interact with other sub-agents (via the dispatch command), which increases the potential impact of a successful injection.
Sanitization: There is no evidence of automated sanitization, filtering, or escaping of the user-provided draft content before it is processed or used in writing rationale matrices.

paper-spine-rewrite