afkbot-io

aider-cli

Aider CLI orchestration through `bash.exec`. Use when the user explicitly wants to run aider from shell, install aider, choose provider-specific models, or execute one-shot `aider --message` edits against selected files or a repo.

automation

Automation management via `automation.*` tools. Use when the user explicitly wants to create, inspect, update, list, or delete cron or webhook automations.

claude-code

Anthropic Claude Code orchestration through `bash.exec`. Use when the user explicitly wants to run Claude Code from shell, install or authenticate the `claude` CLI, choose model aliases such as `sonnet` or `opus`, or execute non-interactive `claude -p` workflows.

codex-cli

OpenAI Codex CLI orchestration through `bash.exec`. Use when the user explicitly wants to run Codex from shell, install or authenticate Codex CLI, choose a Codex/OpenAI model, or execute one-shot `codex exec` or `codex review` workflows against a repo.

gemini-cli

Google Gemini CLI orchestration through `bash.exec`. Use when the user explicitly wants to run Gemini from shell, install or authenticate the `gemini` CLI, choose a Gemini model, or execute one-shot `gemini -p` workflows against a repo or prompt.

imap

Mailbox search via `app.run` for `imap`. Use when searching email messages in IMAP mailboxes and resolving the needed credential profile.

partyflow

PartyFlow bot operations through `app.run`. Use when sending messages to PartyFlow, polling bot events, checking bot identity, or joining a PartyFlow conversation with stored credentials.

skill-creator

Profile skill management via `skill.profile.*` and `skill.marketplace.*`. Use when listing, inspecting, creating, editing, deleting, or installing custom skills. Every skill must keep `name` and `description` in YAML frontmatter because AgentLoop summaries are built from that metadata.

smtp

Outbound email via `app.run` for `smtp`. Use when sending email and resolving SMTP credential profiles securely.

taskflow

Task Flow backlog management via `task.*` tools. Use when the user explicitly wants durable tasks, TODO/backlog management, kanban-like flows, decomposition into multiple tasks, dependencies, ownership changes, or human handoff.

Telegram Bot API operations via `app.run` for `telegram`. Use when sending messages, fetching bot metadata, or reading updates with stored bot credentials.

bash-exec

Shell command execution via `bash.exec`. Use when a task requires running terminal commands and inspecting bounded stdout or stderr under policy controls.

credentials

Secret and credential lifecycle management via `credentials.*` tools. Use when listing, requesting, creating, rotating, or deleting integration credentials without exposing plaintext secrets.

diffs

Render readable unified or HTML diffs from inline text or local files. Use after file edits when you need to inspect or show exact changes, optionally as persisted artifacts.

file-ops

Workspace filesystem operations via `file.*` tools. Use when listing, reading, searching, writing, or editing files inside the allowed workspace.

mcp-manager

Profile MCP management via `mcp.profile.*` tools and `afk mcp`. Use when the user wants to connect, inspect, validate, update, or remove MCP servers by URL for one profile.

sysadmin

System administration via `bash.exec` for host inspection, package management, and service control. Use when tasks mention apt, dnf, yum, brew, package install or update, nginx setup, systemctl, or service status.

memory

Scoped semantic memory management via `memory.*` tools. Use when saving, searching, listing, deleting, or promoting durable facts in local chat memory or profile-global memory.

browser-control

Real browser automation via `browser.control`. Use when a task needs navigation, clicking, typing, screenshots, or page content from a live browser rather than plain HTTP.

subagent-manager

Profile subagent management via `subagent.profile.*` and `subagent.run`. Use when listing, inspecting, creating, editing, deleting, or executing custom subagents.

http-request

Outbound HTTP calls via `http.request` with secure credential reuse. Use when querying or mutating remote HTTP APIs from the agent.

security-secrets

Mandatory secret-handling policy skill. Use on every task involving credentials, API tokens, passwords, or other secrets and rely only on `request_secure_field` plus `credentials.*` tools.