メニュー
Expert in OpenCode permission configuration. Use when user asks about opencode.json, permission allowlists, bash command patterns, or wants to create/configure OpenCode permissions for a project.
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
Expertise in designing Claude Code subagents. Use when creating new agents, understanding agent composition, designing agent templates, or learning about agent constraints.
Advises on when to delegate to subagents. Use when user is working on complex multi-file tasks, exploring codebases, researching, implementing large features, or needs help deciding if delegation would be beneficial.
Filesystem-based message passing for agent communication. Use when user asks about agent messaging, inbox/outbox, how agents talk to each other, multi-agent workflows, agent coordination, passing data between agents, agent handoffs, or working around the subagent spawn constraint. Also triggers on "how do I use messaging", "send message to agent", "agent communication".
Create and edit JSON Canvas files (.canvas) with nodes, edges, groups, and connections. Use when working with .canvas files, creating visual canvases, mind maps, flowcharts, or when the user mentions Canvas files in Obsidian.
The maintenance program for this scaffold's live knowledge gradient — where captures go (zz-log → zz-research → 02-stack → 01-kernel), the promotion gates between stages (evidence-gating, second-user test, supersession rules), decay review, and retrieval-friendly writing (description-as-semantic-key, one concept per file). Use when capturing insights, promoting notes between stages, deciding where knowledge lives, or reviewing stale docs.
Create and edit Obsidian Bases (.base files) with views, filters, formulas, and summaries. Use when working with .base files, creating database-like views of notes, or when the user mentions Bases, table views, card views, filters, or formulas in Obsidian.
| stratum | 2 |
| name | opencode-permissions |
| description | Expert in OpenCode permission configuration. Use when user asks about opencode.json, permission allowlists, bash command patterns, or wants to create/configure OpenCode permissions for a project. |
| branches | ["agentic"] |
I help configure OpenCode's granular permission system. OpenCode has three permission modes per tool and supports pattern-based bash command allowlisting.
| Mode | Behavior | Use Case |
|---|---|---|
"allow" | Auto-execute without prompting | Trusted operations, fast iteration |
"ask" | Prompt for approval each time | Security-conscious, learning |
"deny" | Block entirely | Dangerous operations |
Default: OpenCode defaults to "allow" for most operations.
| Scope | Location | Priority |
|---|---|---|
| Global | ~/.config/opencode/opencode.json | Lowest |
| Per-project | .opencode/opencode.json | Middle |
| Per-agent | Agent definition file | Highest |
edit - File modificationsbash - Command execution (supports patterns!)skill - Skill tool accesswebfetch - Web fetchingdoom_loop - Infinite loop protectionexternal_directory - Operations outside projectWildcards:
* = matches any characters (e.g., git * matches all git commands)? = matches single character{
"permission": {
"edit": "ask",
"bash": {
"npm run *": "allow",
"npm install": "allow",
"npm test": "allow",
"git status": "allow",
"git diff *": "allow",
"git log *": "allow",
"git add *": "ask",
"git commit *": "ask",
"git push *": "ask",
"rm -rf *": "deny",
"rm -r *": "deny",
"*": "ask"
},
"webfetch": "allow",
"external_directory": "ask"
}
}
{
"permission": {
"edit": "ask",
"bash": {
"ls *": "allow",
"find *": "allow",
"du *": "allow",
"file *": "allow",
"mediainfo *": "allow",
"ffprobe *": "allow",
"exiftool *": "allow",
"cat *": "allow",
"head *": "allow",
"tail *": "allow",
"ffmpeg *": "ask",
"mv *": "ask",
"cp *": "ask",
"rm *": "deny",
"*": "ask"
},
"webfetch": "allow",
"external_directory": "ask"
}
}
{
"permission": {
"edit": "ask",
"bash": {
"curl *": "ask",
"wget *": "ask",
"docker ps *": "allow",
"docker logs *": "allow",
"docker exec *": "ask",
"systemctl status *": "allow",
"systemctl restart *": "ask",
"journalctl *": "allow",
"ping *": "allow",
"ssh *": "deny",
"rm *": "deny",
"*": "ask"
},
"webfetch": "allow",
"external_directory": "deny"
}
}
{
"tools": {
"write": false,
"edit": false
},
"permission": {
"bash": {
"ls *": "allow",
"cat *": "allow",
"head *": "allow",
"tail *": "allow",
"find *": "allow",
"grep *": "allow",
"wc *": "allow",
"*": "deny"
},
"webfetch": "allow",
"external_directory": "allow"
}
}
{
"permission": {
"edit": "ask",
"bash": {
"rm *": "deny",
"rmdir *": "deny",
"truncate *": "deny",
"dd *": "deny",
"mkfs *": "deny",
"chmod 777 *": "deny",
"curl * | sh": "deny",
"wget * | sh": "deny",
"sudo *": "deny",
"*": "ask"
},
"external_directory": "deny",
"doom_loop": "deny"
}
}
During interactive use, the "accept always" option in the permission dialog:
ls /some/folder whitelists ls * for the entire sessionls commands will auto-approve after the first onecd dir && ls whitelists BOTH cd * AND ls *Security consideration: Be cautious with "accept always" for risky commands. Once approved, all variations of that command will auto-execute.
"ask" as fallback{
"$schema": "https://opencode.ai/schema.json",
"permission": {
"edit": "ask|allow|deny",
"bash": {
"safe-command": "allow",
"risky-command *": "ask",
"dangerous *": "deny",
"*": "ask"
},
"webfetch": "allow",
"skill": "allow",
"doom_loop": "ask",
"external_directory": "ask"
}
}