メニュー
Use `execute-code` to get "Technical Design" when you must design technical tasks, implement features or refactor code.
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。
SOC 職業分類に基づく
Use `author-caveman` to write Caveman English.
Use `author-agent` to get OpenCode Agent Authoring when writing or reviewing OpenCode agent markdown.
Use `author-command` to get OpenCode Command Authoring when writing or reviewing OpenCode command markdown.
Use `author-rules` to write `AGENTS.md` files.
Use `author-skill` to get Skill File Authoring when you need to review/write skill files for agents.
Use `design-prd` to get Product Requirements when planning any feature or to understand project business requirements, user roles, and success criteria.
| name | execute-code |
| description | Use `execute-code` to get "Technical Design" when you must design technical tasks, implement features or refactor code. |
OpenCode plugin that injects agents, commands, tools, and generated skills. Runtime merges repo config with user config, enforces sandbox/external-directory policies, and manages lifecycle jobs under .agents/jobs/.
src/skills/*/SKILL.md into generated skill paths.src/agents/index.ts): Agent prompt, permissions, model tier, and sandbox policy.src/config.ts): Tier map, external directory rules, sandbox sync settings.src/utils/jobs.ts): concepts → drafts → assist/executing/facilitate/review → shelved.src/skills/index.ts): Bundled skill frontmatter + body./job-concepts (src/commands/index.ts): Save concept markdown./job-design (src/commands/index.ts): Read concept, create plan./job-draft (src/commands/index.ts): Save plan to draft folder./job-execute-assist (src/commands/index.ts): Start assisted execution session./job-execute-auto (src/commands/index.ts): Start autonomous execution session./job-review-commit (src/commands/index.ts): Check criteria, shelve accepted work./job-shelve (src/commands/index.ts): Close job without acceptance.src/utils/tools.ts): Normalized failedAction, error, instruction payloads.src/utils/tools.ts): Same failure retries up to 5, then abort.src/config.ts): Malformed JSONC throws with file path.src/tools/*): Abort or retry based on tool-specific instruction text.Auth is delegated to OpenCode session/client context. Security control is mostly permission policy: agents default-deny, then selectively allow task/tool access, external directories, and sandbox tools. applyExternalDirectoryPolicy() and applySandboxPlatformPolicy() remove unsafe access when platform support is missing. No secrets are stored in repo config; use ${ENV_VAR} placeholders.
src/tools, src/utils/jobs.ts): Session create/update/prompt and job orchestration — SDKsrc/config.ts, src/utils/jobs.ts, src/skills/index.ts): Reads local/project/global config and job files — Node fssrc/tools/autocode_sandbox_*, src/agents/index.ts): Creates/cleans sandboxes and gates unsupported platforms — local processsrc/agents/): Primary prompts, rules, and agent definitions.src/commands/): Programmatic command templates.src/tools/): OpenCode tool implementations and tests.src/utils/): Shared config, job, sandbox, and error helpers.src/skills/): Bundled generated-skill Markdown..agents/jobs/): Concepts, drafts, execution, review, and shelving artifacts.src/skills/index.ts: Bundles managed skills into generated config path.src/config.ts: Loads global/local .opencode/autocode.jsonc overrides.src/plugin.ts: Final merge hook for agents, commands, tools, and skills.src/agents/prompts/execute_code.ts: Canonical execution prompt for code-writing sessions.ensureGeneratedSkills() clears stale generated skill dirs each load.IMPORTANT: Update .agents/skills/execute-code/SKILL.md whenever architecture, APIs, data models, security, or integrations change.