The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill executes remote scripts by piping content from external URLs directly into the shell or PowerShell during installation processes.
[EXTERNAL_DOWNLOADS]: The skill downloads and executes installation scripts from the external domain herdr.dev to set up its required environment.
[PRIVILEGE_ESCALATION]: The skill bypasses platform safety mechanisms by using PowerShell's ExecutionPolicy Bypass and enabling agent-specific flags like --dangerously-skip-permissions and --dangerously-bypass-approvals-and-sandbox, which disable all user confirmation prompts and security sandboxing for autonomous operation.
[COMMAND_EXECUTION]: The skill makes extensive use of the herdr command-line tool to manage terminal panes, agent lifecycles, and cross-agent communications.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an intermediary for instructions and reports between multiple AI agents, creating an attack surface for indirect prompt injection.
Ingestion points: Data is ingested from external files such as loop-state.md and audit-register.md, as well as relayed messages from the user.
Boundary markers: Uses quoted heredocs with unique, randomized delimiters to prevent prompt data from being interpreted as shell commands during relay.
Capability inventory: Broad permissions including file editing, shell command execution, and control over other agents with bypassed security settings.
Sanitization: Employs shell-safe quoting and heredoc delimiters to ensure relayed strings are treated as literal data by the coordinator's execution environment.
[DYNAMIC_EXECUTION]: The skill dynamically modifies command syntax, such as skill prefixes, based on the target agent's specific requirements at runtime.
[TIME_DELAYED_CONDITIONAL]: The quota management system pauses agent execution based on specific rate-limiting messages detected in agent output, waiting for calculated reset periods.

dev-harness