The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses strong mandatory language and override markers to force the agent into a specific workflow. Phrases such as "⚠️ MANDATORY: Every research poster MUST include...", "This is not optional", and "The AI will automatically..." are designed to bypass the agent's default reasoning and force the use of external scripts and specific third-party content.
[METADATA_POISONING]: There is a discrepancy between the provided author context ("wu-yc") and the metadata claims in the skill ("skill-author: K-Dense Inc."). Furthermore, the skill contains self-promotional content for an external platform ("www.k-dense.ai") that is not associated with the confirmed author's known resources.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a local script scripts/generate_schematic.py using arguments derived directly from natural language user input ("your diagram description"). Without evidence of sanitization, this pattern presents a high risk of command injection if the user input contains shell metacharacters.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect injections through the ingestion of untrusted research data.
Ingestion points: User-provided research content, diagram descriptions, and LaTeX source files.
Boundary markers: No specific delimiters or "ignore instructions" warnings are used when processing external content.
Capability inventory: Use of the Bash tool to run scripts (generate_schematic.py, review_poster.sh) and LaTeX compilers.
Sanitization: Lack of visible validation or escaping for user-supplied descriptions before they are passed to shell commands.

latex-posters