The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data in the form of user-provided drafts, rubrics, and source material (existing_draft, rubric_or_constraint, source_data). This creates a surface for indirect prompt injection, where malicious instructions hidden within the processed documents could attempt to override the agent's behavior.
Ingestion points: External data enters the agent context via the mode inputs, specifically when reading the rubric_or_constraint, existing_draft, and source_data files as specified in the Entry flow.
Boundary markers: The skill implements a 'rubric-quote gate', which explicitly refuses to apply a rule until the literal text of the constraint is quoted from the source, providing a verification checkpoint.
Capability inventory: The skill is limited to READ, WRITE, and EDIT operations on local files and NOTIFY primitives for inter-skill communication. It lacks high-risk capabilities such as arbitrary shell command execution, network access to unknown domains, or dynamic code evaluation.
Sanitization: No automated sanitization of input text is documented; however, the skill follows a strict canonical workflow and iterative audit process that enforces specific stylistic protocols.

oma-academic-writer