Use skills

Skills are reusable instruction sets the model loads only when a task matches — they don’t cost context on turns that don’t need them.

Where skills live

~/.octo/skills-default/ — the built-in set below, materialized from the binary on first run (octo skills update re-syncs it after an upgrade). Kept in its own directory so refreshing defaults never touches a skill you wrote yourself.
~/.octo/skills/<name>/SKILL.md — user-level, available across all projects. A same-named skill here overrides a default.
.octo/skills/<name>/SKILL.md — project-level, overrides both of the above on a name collision.

The format is identical to Claude Code’s, so you can symlink ~/.claude/skills to ~/.octo/skills and reuse everything you already have:

ln -s ~/.claude/skills ~/.octo/skills

Anatomy of a skill

Each SKILL.md is YAML frontmatter plus a markdown body:

---
name: review
description: Review the current diff for correctness and style
---
Walk the diff hunk by hunk and flag correctness bugs first, then style.

At session start, octo lists each skill’s name and description in the system prompt — a one-line manifest, not the full body. The model loads a skill’s full instructions on demand (via the skill tool) only when a task matches.

Using skills

octo skills list     # see what's discovered
octo skills path     # print the user/project/default skill directories
octo skills add      # install a skill from a source (guided)
octo skills update   # re-sync the built-in defaults after an upgrade

In the TUI, /skills lists them and /<name> (e.g. /review) runs one directly. In the Web UI, the composer’s / menu does the same. IM channels don’t support /<skill-name> triggering — any /text there is matched against the fixed slash command set, and an unrecognized one returns “Unknown command” rather than falling through to a skill; ask for the task in plain language instead and the model loads the matching skill itself.

Built-in skills

octo ships 21 skills out of the box. Every one triggers automatically when the model judges a task matches its description — you rarely need to invoke them by name.

Get started

Skill	What it does
`onboard`	First-run setup (name, personality, profile → `soul.md` + `user.md`); also handles narrower re-curation with `scope:soul`, `scope:user`, or a specific memory file path
`product-help`	Answers “how do I…” / “what is…” questions about octo itself by reading its own product docs
`config-setup`	Configures octo’s global settings through conversation — set up AI model endpoints (providers, API keys, models), adjust agent defaults (reasoning effort, permission mode, coauthor, workspace directory), and manage default/lite model assignments
`skill-creator`	Turns a repeatable task into a new `SKILL.md`, or edits/improves an existing one
`workflow-creator`	Chains existing skills and browser recordings, or composes fresh `agent`/`parallel`/`pipeline` orchestration, into one runnable, saved workflow

Build & ship code

Skill	What it does
`tech-design`	Produces a full backend technical design doc from a PRD or feature description
`grill-me`	Interviews you about a plan until every open decision is resolved — pairs with `tech-design`
`implement`	Decomposes a tech design into dependency-ordered slices, TDD-executes each, reviews each with a sub-agent, and checkpoints progress so it survives a restart
`code-review`	Reviews the current diff with an isolated sub-agent for unbiased correctness/convention/security feedback
`worktree-isolate`	Does a risky or experimental change in an isolated git worktree, then merge or discard

Automation & scheduling

Skill	What it does
`cron-task-creator`	Creates/inspects/edits/deletes recurring prompts that survive a restart, run by `octo serve`’s scheduler. See Schedule cron tasks

Connect things

Skill	What it does
`expert-agent-manager`	Manage octo’s agent profiles through conversation — create, modify, delete, list, and bind agents to IM chats by calling REST APIs
`mcp-creator`	Finds the right MCP server, writes the `mcp.json` entry, and verifies the connection
`channel-manager`	Walks you through setting up an IM platform (Feishu, WeChat, WeCom, DingTalk, Discord, Telegram) and writes `channels.yml`

Documents, media & research

Skill	What it does
`web-access`	Methodology + a cross-session experience library for hard web targets: login-gated or anti-bot sites, unknown page structure, cross-source verification
`artifact-design`	Design guidance for any self-contained HTML/Markdown page shown in the Artifacts panel — reports, dashboards, diagrams, generated UIs
`dataviz`	Chart/graph/dashboard rules — chart-type selection, categorical/sequential/diverging color systems, legend/axis/tooltip conventions
`office-xlsx`	Creates/reads/edits `.xlsx` spreadsheets — formulas, styling, merged cells, multiple sheets, charts, validation
`ppt-master`	Turns a document (PDF/DOCX/URL/Markdown) into an editable PowerPoint deck — SVG-authored slides with native charts/tables and speaker notes, exported to `.pptx`
`image-gen`	Generates images with an AI model (14 provider backends) or sources openly-licensed stock, saved to files — one-off or batch; other skills like `ppt-master` delegate to it
`deep-research`	Deep, multi-source, fact-checked research — fans out searches, reads primary sources, adversarially verifies claims, and synthesizes a cited report

Next: chaining several of these into one saved flow is exactly what Workflows are for.