The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses files outside its immediate directory (e.g., ../../tw_edu_concept_alignment.md). This indicates a file system access scope that extends to parent directories, which is used here for concept alignment across related academic skills.
[COMMAND_EXECUTION]: The skill configuration allows access to the Bash tool. While the instructions are focused on generating academic critiques and reports, the presence of a shell tool provides a high-capability environment for the agent to operate within.
[PROMPT_INJECTION]: The skill presents an Indirect Prompt Injection surface. It is designed to ingest large amounts of untrusted academic text (research papers or drafts) and process them using high-capability tools like Bash and Write (Cat 8).
Ingestion points: User-provided academic views in Step 1, Step 2b, and full paper drafts in SE-Step 1.
Boundary markers: Absent; there are no specific delimiters or instructions to ignore embedded commands within the ingested academic text.
Capability inventory: Includes Bash, Write, Read, and WebSearch tools.
Sanitization: No explicit sanitization or validation of the ingested text is performed before it is processed by the agent's logic.

tw-research-dialectics