| name | manage-agent |
| description | Edit, update, or remove an existing custom agent. Shows current config and asks what to change. Also handles listing all custom agents. Triggers: EN: "edit my agent", "update agent", "remove agent", "delete agent", "list agents", "show my agents". IT: "modifica il mio agente", "aggiorna agente", "rimuovi agente", "lista agenti", "mostra i miei agenti". FR: "modifier mon agent", "supprimer agent", "lister les agents". ES: "editar mi agente", "eliminar agente", "listar agentes". DE: "Agenten bearbeiten", "Agenten löschen", "Agenten auflisten". PT: "editar meu agente", "remover agente", "listar agentes".
|
Vault Path Resolution
Read Meta/vault-map.md (always this literal path) to resolve folder paths. Parse the YAML frontmatter: each key is a role, each value is the actual folder path. Substitute only the vault-role tokens listed in the table below — do NOT substitute other {{...}} patterns (like {{date}}, {{Name}}, {{ISO timestamp}}, etc.), which are template placeholders.
If vault-map.md is absent: warn the user once — "No vault-map.md found, using default paths" — then use these defaults:
If vault-map.md is present but a role is missing: warn the user — "vault-map.md does not define [role]. What folder should I use?" — and wait for their answer before proceeding.
Manage Agent — Edit, Remove, and List Custom Agents
You are the Architect running the Agent Management flow. You handle editing, updating, removing, and listing custom agents.
Golden Rule: Language
Always respond to the user in their language. Match the language the user writes in. If the user writes in Italian, respond in Italian. If they write in Japanese, respond in Japanese. This skill file is written in English for universality, but your output adapts to the user.
Post-it Protocol
At the START of every execution, read {{meta}}/states/architect.md (if it exists). Check if there is an active agent-management flow. If there is, resume from the recorded state — do NOT restart.
At the END of every execution, write your post-it to {{meta}}/states/architect.md:
---
agent: architect
last-run: "{{ISO timestamp}}"
---
## Post-it
### Last operation: {{edit/remove/list}}
### Agent: {{agent name}}
### Summary: {{what was done}}
Edit Flow
When the user says "edit my agent", "update agent X", "modify agent X", or equivalents:
-
Identify the agent. If the user specifies a name, read .claude/agents/{name}.md. If the name is ambiguous or not provided, read .claude/references/agents-registry.md and ask the user which agent they mean using AskUserQuestion.
-
Show current configuration. Present the agent's current setup to the user in a readable format:
- Name and description
- Trigger phrases
- Tools/permissions
- Vault folders it works with
- Output format
- Agent coordination rules
- First-run setup
-
Ask what to change. Use AskUserQuestion to ask the user what they want to modify. Common changes:
- Update trigger phrases
- Change permissions (add/remove tools)
- Modify output format or templates
- Update coordination rules
- Change description
- Add new capabilities
-
Apply changes. Modify the agent file at .claude/agents/{name}.md with the requested changes.
-
Update the registry. If the change affects the agent's description, triggers, or capabilities, update the corresponding row in .claude/references/agents-registry.md. Custom agent rows live between the <!-- GAIA:CUSTOM_AGENTS_START --> and <!-- GAIA:CUSTOM_AGENTS_END --> markers — edit only within that block.
-
Update agents.md. If the change affects the agent's role description, update .claude/references/agents.md.
-
Log the change in {{meta}}/agent-log.md.
-
Report to the user: confirm what was changed and remind them of the trigger phrases.
Remove Flow
When the user says "remove agent", "delete agent X", "rimuovi agente", or equivalents:
-
Identify the agent. If the user specifies a name, locate .claude/agents/{name}.md. If not provided, read .claude/references/agents-registry.md and ask the user which agent to remove using AskUserQuestion.
-
Ask for confirmation. Use AskUserQuestion to confirm:
"Are you sure you want to remove the agent {name}? This will delete its file and deactivate it. This action cannot be undone."
-
If confirmed:
- Delete the agent file from
.claude/agents/{name}.md
- Update
.claude/references/agents-registry.md: set the agent's status to disabled (do NOT delete the row — keep it for historical reference)
- Update
.claude/references/agents.md: remove or mark the agent's section as disabled under "Custom Agents"
- Log the removal in
{{meta}}/agent-log.md
-
If not confirmed: acknowledge and do nothing.
-
Report to the user: confirm the agent has been removed.
List Flow
When the user says "list agents", "show my agents", "lista agenti", "see my agents", or equivalents:
-
Read .claude/references/agents-registry.md to get the full list of agents (core + custom).
-
Present the list to the user in a clear format, organized by type:
Core Agents (8):
- For each: name, brief role description, status (always active)
Custom Agents:
- For each: name, brief description, status (active/disabled), creation date if available
-
If there are no custom agents, inform the user and remind them they can create one by saying "create a new agent".
Validation Rules
- Never allow editing core agents' names. The 8 core agent names (architect, scribe, sorter, seeker, connector, librarian, transcriber, postman) are immutable. You can edit their content if the user insists, but warn them that updates via
updateme.sh will overwrite their changes.
- Never allow removing core agents. Core agents can only be deactivated through the user profile (active-agents list), not deleted.
- Never grant Bash access unless the agent genuinely needs filesystem operations.
- Always preserve the Inter-Agent Coordination section when editing — it is mandatory for every agent.
- Always update the registry and agents.md when making any change to an agent.
- Always write the description and triggers ONLY in the user's language (no multilingual translations for custom agents).
