Skip to content
New

DeepSeek V4.1 Flash free for 7 days: 10M tokens a day after a top-up of $10 or more. Until Nov 30.

How it works
aeondave/opencode-agent-creatorReporting and compliance

opencode-agent-creator

⁨Create, revise, or debug OpenCode agent definitions and teams: Markdown/JSON configuration, permissions, model routing, discovery, and native Task delegation.⁩

v2026.11.13-21cd5682 installs+2 in 7 days39 StarsMITsynced 11 hours ago

Install

No NotHoly CLI yet? Install it first.

$ curl -fsSL https://notholy.com/install.sh | sh
$ notholy skills install aeondave/opencode-agent-creator

Writes to ⁦~/.notholy/skills⁩ after checking the signature and the SHA-256.

SignatureSigned, ed25519
BNQdmrMTss…8ydRyzBw==The CLI checks the SHA-256 and the signature before it writes a file.
Package9d26f4ef…b1daa8e4
9d26f4eff368…f7a0b1daa8e413 files · ⁦10.0 KB⁩
ScanMedium2 findings
  • injection.no_confirmationreferences/agent-config-reference.md:67
  • Links to outside domains
See the full scan
Sourcegithub.com/AeonDave/malskill/tree/main/harness-dev/opencode-agent-creator@21cd568

Create the smallest agent definition or roster that handles the requested work. Prefer project-local .opencode/agents/ unless another scope is requested. Preserve existing agent identities, permissions, and user decisions.

Use agent-md-creator for repository instructions, skill-creator for portable skills, and claude-agents-creator for Claude Code subagents. Use opencode-plugin-creator for custom tool or plugin implementations.

Workflow

  1. Run opencode --version; inspect the existing project configuration and requested agent scope. These instructions target stable v1.18.35. For a different release or V2 target, verify its actual schema and runtime first.
  2. Choose one bounded specialist, or a primary supervisor plus the needed specialists. Reuse an available built-in when it already fits. Load multi-agent-topologies.md when choosing a team shape.
  3. Give each custom agent a concise routing description, explicit mode, task-specific prompt, and appropriate permissions. Keep one definition per agent. Load agent-config-reference.md when writing fields, resolving discovery, or configuring nested delegation.
  4. Start from agent-template.md or supervisor-template.md. The examples form a read-only review team; add a real write-capable worker when implementation is required.
  5. Leave model unset to inherit, or pin a model verified by opencode models. Load model-tiering.md when choosing model policies or investigating a failed dispatch. Task has no per-dispatch model parameter.
  6. Give workers the target, constraints, relevant context, and expected result in each dispatch. A fresh child does not receive the parent's conversation; a resumed child retains its own history. Load orchestrator-pattern.md for packets, task resumption, and experimental background work.
  7. Enforce capabilities with permissions. hidden controls autocomplete visibility, not access. Keep leaf workers at task: deny; configure depth and task permissions explicitly for intentional nested delegation. A read-only supervisor can dispatch a write-capable worker, so constrain each worker separately.
  8. Reload the definitions and inspect opencode agent list. Test routing, the promised permission boundary, the returned result, and any resume/background behavior used. Load triggering-and-testing.md before completion.

Report the files changed, target OpenCode version, checks actually run, and any untested model or delegation behavior. File/schema validation alone does not prove model routing or tool enforcement.

Load plugins-and-tools.md when an agent needs a custom tool, plugin, or launch-time secret; it routes implementation work to the plugin skill.

Files in this version

Permissions

PermissionDeclaredFound by the scan
Reads filesNoNo
Writes filesNoNo
Runs shell commandsYesYes
Uses the networkYesYes
Uses connectorsNoNo
Drives a browserNoNo

Scan findings

Reviewed on Nov 13, 2026

  1. MediumPrompt injection · injection.no_confirmationreferences/agent-config-reference.md:67
  2. InfoLink · Links to outside domains

A clean scan lowers the risk but does not prove a package is safe. Read what you install.

VersionPublishedCommitSHA-256SignaturePermission change
v2026.11.13-21cd568Nov 13, 202621cd5689d26f4efSigned–

Ask anything your scope allows.

Create an account, add credits and ask the first question in the chat or the terminal.

  1. 1/3

    Create the account

    Individual or company. You can switch to a company later without losing balance or keys.

    IndividualCompany
  2. 2/3

    Add credits

    By card. Pay per token, no subscription.

    $5.00minimum deposit
  3. 3/3

    Ask

    In the chat, or install the CLI and run notholy login in your terminal.

    $ notholy login