The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external plans, agendas, and project descriptions through commands like /em:challenge <plan> and /em:board-prep <agenda>. It also integrates a company-context.md file into the agent's context. This creates a surface for indirect prompt injection where untrusted data could influence agent behavior.
Ingestion points: User input supplied to commands (<plan>, <agenda>, <decision>, <assumption>, <event>) and the company-context.md file referenced in SKILL.md.
Boundary markers: The skill uses specific command prefixes and structured output formats (e.g., the 'Concern Structure' in agents/devils-advocate.md) which provide structural delimiters, though no explicit 'ignore instructions' tags are present.
Capability inventory: The skill utilizes local Python scripts (scripts/decision_matrix_scorer.py and scripts/stakeholder_mapper.py) to perform calculations and visualization. These scripts are restricted to standard libraries (json, sys, argparse) and do not have network access or dangerous system-level permissions.
Sanitization: Input is parsed using standard json.load() and argparse within the provided scripts. The agent's core 'Adversarial Reasoning' instruction serves as a logical filter that encourages critical evaluation of the input content.

executive-mentor