Skip to content

opencode-commands-skills-tools.md

Latest commit

 

History

History
83 lines (59 loc) · 3.28 KB

opencode-commands-skills-tools.md

File metadata and controls

83 lines (59 loc) · 3.28 KB

OpenCode Commands, Skills, and Tools

This page documents repository-specific slash commands, the skill layout, and custom tools shipped in this repo.

Slash commands

Command Purpose
/research <task> RPI phase 1 - explores the codebase and writes a research doc to docs/thoughts/
/deep-research <task> Same as /research but uses Claude Opus for harder problems
/plan <artifact-folder> RPI phase 2 - turns a research doc into a numbered implementation plan
/implement <artifact-folder> RPI phase 3 - executes the plan step by step and runs the quality gate
/skill <skill-name> "<description hint>" Scaffolds a new skill folder
/security-profile [init|refresh] [project-name] Creates or refreshes a project-specific defensive security-profile skill from security-core

The RPI workflow is research -> plan -> implement:

  • research reads and maps the codebase only
  • plan converts findings into a step-by-step plan only
  • implement executes the plan and runs verification gates

Per-project security profile

For each project, run both commands so the repository gets and then maintains its security profile:

/security-profile init
/security-profile refresh
  • init scaffolds .opencode/skills/security-profile/ with SKILL.md, GOTCHAS.md, HISTORY.md, and config.json.
  • refresh updates inferred metadata while preserving user-owned settings where possible.

The profile is defensive-only and intended to complement .opencode/skills/security-core/.

Skills

Skills are domain-knowledge folders loaded on demand. Each skill lives under .opencode/skills/<skill-name>/ and uses this structure:

File / Dir Purpose
SKILL.md Reference facts, file map, config and memory guidance
GOTCHAS.md Accumulated failure points and fixes - never deleted
HISTORY.md Append-only change log - one entry per session that modified the domain
assets/ Templates, static files, and output scaffolds
scripts/ Helper scripts and libraries the agent can run or compose

Creating a skill

/skill <skill-name> "<short description hint>"

Example:

/skill billing-lib "Internal billing library - edge cases, footguns, charge flow"

After creation:

  1. Fill in SKILL.md -> ## Key Facts with what the model needs to know (not do): file paths, data shapes, naming rules, API shapes.
  2. Add reusable scripts to scripts/ and reference them in ## File & Directory Map.
  3. Add output templates or static assets to assets/.
  4. If the skill needs per-user config, keep ## Configuration and define config.json shape.
  5. If the skill benefits from memory, keep ## Memory and choose a persistent log format.
  6. Remove ## Configuration and ## Memory if not needed.

Custom tools

The custom tool .opencode/tools/pdftotext.ts exposes pdftotext with these arguments:

  • filePath (required): PDF path, absolute or relative to the current directory
  • firstPage / lastPage (optional): 1-based page range
  • preserveLayout (optional): uses pdftotext -layout
  • rawOrder (optional): uses pdftotext -raw
  • maxChars (optional): maximum characters returned (default 40000)

Example prompt:

Read from docs/some-pdf-file.pdf, pages 2-4, preserving layout.