The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on several CLI tools, including herdr, git, and gh, to manage workspaces, track repository state, and interact with terminal panes. These operations are executed via the provided pm.py and resolve_ids.py helper scripts.
[EXTERNAL_DOWNLOADS]: The documentation suggests the installation of sessionr from the author's repository (github.com/yigitkonur/cli-sessionr) to enable deep history analysis. As this is a vendor-owned resource, it is a standard extension of the skill's functionality.
[REMOTE_CODE_EXECUTION]: The skill's primary function is to orchestrate other agents by sending shell commands and slash-commands (e.g., /compact, /plan, /simplify) into their respective panes. It can also spawn parallel 'executor lanes' in git worktrees, which involves starting new agent processes.
[DATA_EXFILTRATION]: The agent reads local session transcripts and state data from ~/.claude/projects/ and ~/.local/state/herdr-pm/. To mitigate risk, the instructions explicitly mandate a 'Secret-scan' before any write operation to scrub API keys, tokens, or sensitive credentials from its output.
[PROMPT_INJECTION]: While the skill contains extensive instructions for managing other agents, these are operational guidelines (e.g., 'mission-grade instructions') rather than attempts to bypass the model's safety guardrails. The instructions focus on maintaining context and verifying task completion.

herdr-pm-agent