The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes forceful instructional language to steer the model's reasoning process, such as 'Each member MUST disagree' and 'If everyone agrees, the Council has failed.' These are functional requirements for the simulation but represent a form of behavioral override.
[PROMPT_INJECTION]: Vulnerability to indirect prompt injection via user questions.
Ingestion points: User input provided during the deliberation phases described in SKILL.md and custom archetype definitions in references/advanced-configurations.md.
Boundary markers: Absent. There are no explicit delimiters or warnings to ignore embedded instructions within user-provided context.
Capability inventory: The agent is granted Read and Write tool permissions. Maliciously crafted user input could potentially influence the agent to perform unauthorized file operations during the synthesis or recommendation stages.
Sanitization: Absent. No filtering or escaping mechanisms are described for handling external user data.
[NO_CODE]: The skill is entirely instructional and does not include any Python, Node.js, or other executable scripts.

consciousness-council