The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill renders untrusted workflow data obtained from the environment into its user-facing output. If this metadata contains malicious instructions, the agent could potentially be influenced when processing the candidate table or during the confirmation prompt.
Ingestion points: Data is ingested from the output of exarchos:exarchos_orchestrate and exarchos:exarchos_view tools.
Boundary markers: No boundary delimiters or instructions to ignore embedded content are present in the rendering logic for the candidate table.
Capability inventory: The skill is capable of performing state-modifying actions through the exarchos:exarchos_orchestrate tool.
Sanitization: The instructions do not mandate sanitization or escaping of the dynamically loaded workflow identifiers.
[COMMAND_EXECUTION]: The skill's operational logic depends on the backend server executing shell commands via the GitHub CLI (gh) and git for safeguard evaluations, as specified in the 'Prerequisites' and 'Safeguards Explained' sections of the skill file.

prune