The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands using the herdr binary for workspace, tab, and agent management. It also uses the gh (GitHub) CLI for issue tracking and references various coding agent runtimes including codex, claude, agy, cursor, opencode, and copilot as valid PATH-resident binaries for worker tabs.\n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to instructions embedded in external data that could influence the behavior of fanned-out worker agents.\n
Ingestion points: Untrusted data enters the context through issue titles and bodies fetched from GitHub or Linear APIs (references/tracker-map.md), and via the stdout/stderr read back from worker tabs (references/herdr-commands.md).\n
Boundary markers: The worker-prompt.md template uses headers like ISSUE: and ISSUE_URL: to structure data, but it does not provide explicit directives for the worker agent to ignore malicious instructions potentially residing in the issue body.\n
Capability inventory: The orchestrator can create/delete tabs, launch coding agents with user-specified flags, and update issue trackers with labels and comments (references/tracker-map.md).\n
Sanitization: The skill does not perform validation or sanitization of ingested issue content before interpolating it into the prompts used to drive worker agents.

orchestrate-herdr