Agent Native — Parallel Worker Dispatch & Role Contract
A strict operating contract for dispatching independent work packets to parallel agent workers or named persistent roles, with disjoint write scopes and observed-only runtime evidence.
AutomationAdvanced★ 446⑂ 42AI score 7/10Last updated: Sep 26, 2026
What it does
- Defines how to dispatch independent work packets to parallel workers or caller-selected persistent role sessions.
- Separates authority by role: Orchestrator (passes intent, scope, evidence refs; names the final-review owner), Implementer (may touch only its packet's subject), Validator (exact candidate content in a fresh read-only context), Scribe (records runtime evidence, never judges). Reader/Writer are bounded cheap delegations (≤350-line slices; a Writer requires a reference file).
- Forces pre-launch validation of every packet: resolve symlinks, normalize paths, case-insensitive scope comparison so an alias cannot hide a collision; reject nonempty write-scope excludes.
- Names and blocks the prompt-send optimism failure mode — a delivered prompt proves transport, not engagement — and the anti-pattern of restarting an unresponsive worker before capturing its state.
- Requires observed session/context identity (a requested ID is not an observed ID), preservation of launch failures, and clean separation between runtime facts and verdicts.
Who it's for
- Teams actually running AgentOps-style workflows or Codex native roles (
bulk-reader,code-writer). - Engineers orchestrating several concurrent agents who have hit write collisions, duplicate dispatches, or ghost workers.
- Advanced users who want an auditable split between execution and validation.
Examples
- Build three implementation packets scoped to separate directories, run the disjointness/symlink check, then dispatch each exactly once and keep identity with each result.
- Delegate a large-file survey to a Reader in ≤350-line slices and get line-referenced bullets back without loading file bytes into the parent context.
- On a suspected stall, capture observable native state first, then choose a nudge or replacement within remaining authority instead of restarting.
· · · Install guide · · ·
Try it now, no install
Paste this into Claude to use the skill without installing anything.
Read the instructions in this file and follow them to help me: https://raw.githubusercontent.com/boshu2/agentops/HEAD/skills-codex/agent-native/SKILL.md What I want: (describe your task here)
If Claude can't open the link, open it yourself and paste the contents instead.
↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.
Install in the Claude app (no terminal)
- Download the ZIP with the button below.
- In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
- Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
Install in Claude Code
Let Claude do it — paste this into Claude Code
Install the skill I found on Claude Skill Mart. Copy the skills-codex/agent-native folder from the GitHub repo boshu2/agentops into my ~/.claude/skills/boshu2-agent-native/. When it's done, tell me in one line what this skill can do.
Install with a command instead
git clone https://github.com/boshu2/agentops.git && mkdir -p ~/.claude/skills && cp -r agentops/skills-codex/agent-native ~/.claude/skills/agent-native⚠ This is a third-party skill. Check the source repository before installing.
- Open a terminal.
- Clone the repo:
git clone https://github.com/boshu2/agentops.git - Create the skills directory:
mkdir -p ~/.claude/skills - Copy the skill:
cp -r agentops/skills-codex/agent-native ~/.claude/skills/agent-native - Verify the
references/files (model-dispatch.md, session-associations.md, judgment-receipts.md, RAW_SOURCE_READS.md) came along; re-copy them from the repo if missing. - Note that
scripts/swarm/dispatch_once.pyis a developer reference requiring an AgentOps source checkout and is not bundled with a standalone skill install — keep the clone if you need it. - Restart Claude Code and ask something like "split this into disjoint packets and dispatch parallel workers" to trigger the skill.
View source on GitHub ↗License: Apache-2.0