メニュー
How to operate as a foci agent — the tools you call, how your prompts and turns are built, the databases behind your state, and the config that shapes you. Read the relevant subfile before doing related work.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Debug and investigate foci platform internals. API logs, payload logs, session files, CC backend transcripts, cache diagnosis, service logs, and common investigation patterns. Read the relevant subfile before investigating.
Security scanner for untrusted code and AI agent skills. Detects credential theft, code injection, privacy violations, deception, resource abuse, and ethical issues before you read or activate untrusted content.
Finding text on disk — use ack, not grep (grep is only for filtering the piped output of other commands). Conventions for fast, correct file-content search across a codebase.
Drive a live browser via accessibility-tree snapshots — navigate, click, fill forms, select, press keys, screenshot. Use when a task needs interacting with a real web page (logins, JS-rendered content, multi-step flows), not a static read (that's web_fetch).
Developing the foci platform itself (Go server + backends). Architecture, the CC/opencode backends, the routing/delivery model, and the turn/steer/ask lifecycle — the internals you need when CODING foci, not when operating as an agent. Read the relevant subfile before changing that area.
Generate images via OpenRouter (GPT-5 Image, GPT-5 Image Mini, Gemini Pro/Flash).
SOC 職業分類に基づく
| name | foci-usage |
| description | How to operate as a foci agent — the tools you call, how your prompts and turns are built, the databases behind your state, and the config that shapes you. Read the relevant subfile before doing related work. |
You are an agent running on foci, a platform that bridges messaging channels (Telegram, Discord, an HTTP API, voice) to an LLM backend and gives you a set of tools, a persistent memory, and a turn lifecycle. This skill is the operating manual: what foci is from your point of view, and where to look when you need detail.
This
SKILL.mdis yours to customise (seed-if-missing — override it, add your own sibling files). The content files it lists below ship with foci and are overwritten on restart — edit those in the foci repo (shared/skills/foci-usage/), not the deployed~/shared/skills/copy.
| Subfile | Read it when you need… |
|---|---|
| tools-api.md | You're an API-loop agent (backend = ""/"api"): a self-contained manual for calling tools as formal JSON tool-calls, including foci's own file/shell/spawn/browser tools. |
| tools-backend.md | You're a Claude Code (shell) agent (most agents): a self-contained manual for the foci_* shell functions, plus CC-native tools and deferred tools/ToolSearch. |
| prompts.md | Where foci's prompt templates live and how to customise them; the [meta] header, nudges, injections; [[NO_RESPONSE]]; compaction. |
| scheduled-tasks.md | The periodic tasks foci runs for you (keepalive, reflection, consolidation, log rotation) and how to create your own durable scheduled turns via the crontab. |
| databases.md | The SQLite stores behind todos, reminders, scratchpad, memory, sessions, cost. |
| config.md | foci.toml structure, smart defaults derived from agent ID, secrets, models, cron. |
For debugging foci (reading logs, diagnosing cache busts, tracing turns), see the companion foci-debugging skill — this skill is about normal operation, that one is about investigation.
foci_send_to_chat — that tool is for attachments or piping command output, always to your own chat (it has no chat-targeting param). To reach a different chat, use foci_send_to_session.[[NO_RESPONSE]] (and nothing else) to complete a turn without messaging anyone.{{secret:NAME}} in foci_http_request; never paste, echo, or commit them. See config.md.