boadij/pi-herdsman/agent-definitions
Claude Code and agent skill by boadij/pi-herdsman
- community
Checking that the SKILL.md file is still in the repository.
Manage Pi Herdsman Agent definitions and reserved role profiles. Use when inspecting, creating, customizing, enabling, disabling, or changing Agent-definition models, thinking, tools, skills, extensions, delegation, permissions, or instructions.
Install
npx vskill@latest install boadij/pi-herdsman/agent-definitionsSKILL.md
Pi Herdsman Agent definitions
Manage Agent definitions through Herdsman's existing Markdown configuration. The runtime definition engine and canonical documentation remain authoritative; this skill does not define a second schema or management API.
Workflow
- Inspect the Markdown definition layers before changing configuration.
- Identify a definition by its frontmatter
name, never by its filename.
Inspect bundled, trusted project, and global layers in precedence order (bundled < global < trusted project), including which files contribute to the effective definition. Project definitions apply only when Pi trusts the project.
- Resolve matching overlays and effective field values before choosing the
requested project or global edit scope. The Definitions UI edits global files only; compare its global editable values with effective project values and report masked changes accurately. list_agents is useful after an edit to invoke Herdsman's existing discovery and validation, but its enabled-definition capability summary does not expose source paths or complete configuration. It cannot establish effective configuration by itself.
- Disabled definitions, definitions with unavailable required children,
and reserved role profiles are not included in the runtime definition capability list; inspect their Markdown layers directly. Reserved role definitions have special restrictions. Chief is global-only; project-local Chief definitions are rejected.
- Read the canonical contract when exact field behavior matters:
../../docs/reference/agent-definition-schema.md../../docs/guides/agent-definitions.md../../docs/guides/customizing-agents.md
- Choose the configuration layer from the requested scope.
- Agent project policy:
<cwd>/.pi/agents/; global policy:
<agentDir>/agents/.
- Role project policy:
<cwd>/.pi/pi-herdsman/roles/; global
policy: <agentDir>/pi-herdsman/roles/. Pi's agent directory normally is ~/.pi/agent. Managed Lead project policy resolves from the trusted delegated worktree.
- Reserved role names are invalid in Agent directories; non-role names are
invalid in role-profile directories.
- If scope is genuinely ambiguous and materially changes the result, ask.
- Never edit bundled files under
dist/agents/or
dist/roles/. The Definitions UI writes global overrides only.
- Make the smallest correct Markdown change.
- Prefer a partial overlay over copying a complete lower-precedence
definition.
- Preserve unrelated frontmatter, body content, and existing source files.
- The frontmatter
nameis the definition identity; the filename is not. - Array fields replace the whole lower-precedence array. Before adding or
removing one array value, inspect the effective array and write the desired complete replacement.
- For additive prompt instructions, prefer an overlay body with
bodyMode: append; replace the body only when replacement is intended.
- Express capability restrictions structurally through definition fields
rather than relying on prompt prose.
- For ordinary Agents, make
descriptionexplain capabilities, suitable
tasks, and meaningful selection boundaries. Keep it accurate when changing effective capabilities; follow the Agent-definitions guide.
- Herdsman preserves
permission:for compatible extensions but does not
enforce its contents itself.
- Before editing a reserved role definition, read the canonical schema for
its restrictions and authority boundaries.
- Lead, Manager, and Chief runtime profiles are not launch definitions. They
support only name, description, body, bodyMode on overlays, tools, excludeTools, and agents as their direct-delegation allowlist. Do not configure model, thinking, prompt launch mode, enabled or permission policy, tools-default switches, skills, extensions, or context inheritance on these profiles. Their tool policy affects registered ordinary tools; Herdsman-owned tools cannot be activated through a profile and mandatory required coordination tools remain available. Manager defaults to no ordinary tools (tools: []); explicit overlays can select registered ordinary tools. The saved ordinary Lead tool baseline is restored when Manager mode ends. Chief also defaults to no ordinary tools and accepts global overrides only. Inspect effective definition layers, validate edits, and repeat /manager to refresh a running Manager.
- Validate through Herdsman's existing definition engine where available.
- After an edit in a Lead session, call
list_agentsagain. Treat any
discovery or validation failure as a failed edit and repair or restore the target file. Its returned summary can confirm discoverability and validation, not the complete effective configuration or role-profile validation. Validate role profiles through the relevant role resolution; Active role allowlist references must resolve against the Agent roster.
- If
list_agentsis unavailable, do not claim that the effective Herdsman
definition set was runtime-validated. The next normal definition discovery will validate it.
- Inspect the effective result using the definition layers and available
runtime evidence. Report the changed scope and file, effective change, and validation evidence; distinguish configuration inspection from runtime validation. Agent-definition changes affect future Agent assignments; they do not mutate an active Agent. Lead execution profiles resolve at session start, resume, or /lead. Manager and Chief profiles resolve at role activation or restoration; repeating /manager or /chief refreshes the active profile. managed-lead is launch policy: changes do not hot-reconfigure an assigned Lead, and taking effect requires a later managed Lead process start.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/agent-definitions/SKILL.md
Popularity
- Repository stars
- 209
- Repository forks
- 19
- 7-day trend score
- 29
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/boadij/pi-herdsman/agent-definitions)