skills.rest
MP281X
MP281XCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

workflow

Maintains the reusable pair-thread agent workflow across Codex and Claude Code harnesses.

1|Updated Jan 29, 2026
Search Tags:#claude code#codex#agent workflow#multi-agent#pair programming#probing#harness configuration
One-click install
npx skills add https://github.com/MP281X/deslop --skill workflow-mp281x
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: workflow
Source: https://github.com/MP281X/deslop/tree/main/.agents/skills/workflow
Command: npx skills add https://github.com/MP281X/deslop --skill workflow-mp281x

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Changing a multi-agent workflow that runs on both Codex and Claude Code risks re-deriving decisions, losing measured facts, and drifting between the two harness copies. This Skill holds the settled decisions, measured harness behavior, and editing rules so every change to the pair text, roles, configurations, or installer stays consistent. ## Core Features & Use Cases - Decision ledger: Records settled workflow decisions with the measured facts behind them, so nothing is re-derived or re-litigated. - Harness facts: Documents exactly which files reach which thread on Codex and Claude Code, with token counts and configuration switches measured on specific versions. - Probing and eval playbooks: Defines how to run clean-clone probes and skill evals, including scratch homes, prompt style, and per-run cost. - Use Case: When editing the pair text or a role body, consult this Skill to apply the change in both harness copies, respect ownership boundaries between roles, and avoid re-running measurements already settled. ## Quick Start Ask the agent to update the pair-thread text or a role configuration using the workflow skill so the change is applied consistently to both the Codex and Claude Code copies.

Frequently Asked Questions about workflow

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I change the pair-thread text for both Codex and Claude Code?▼

Edit the pair text in both harness copies: tools/workflow/src/agents/codex/config.toml as developer_instructions and tools/workflow/src/agents/claude/CLAUDE.md. There is no renderer; the two files are maintained by hand and copied as-is into the agent homes.

How are Codex and Claude Code agent roles configured?▼

Codex roles live in agents/*.toml files with developer_instructions and model keys, while Claude Code roles live in agents/*.md files with model, effort, and tools frontmatter. A same-name role overrides the built-in explorer, worker, or default agents on both harnesses.

Does Codex spawn subagents from developer instructions alone?▼

No. Measured runs showed developer instructions alone produced zero spawns; Codex authorizes spawning only when the user, an AGENTS.md file, or a skill asks for it. The workflow ships a one-line AGENTS.md naming the roles for this reason.

How do I install the workflow into Codex and Claude Code homes?▼

Run vp run build in tools/workflow, then node dist/install.js. CODEX_HOME and CLAUDE_CONFIG_DIR select the target homes, and the installer removes every managed entry before copying, so hand edits to installed files are lost.

Why do Claude probes need to run outside the home directory?▼

An ancestor .claude/CLAUDE.md under the home directory is loaded as a project file and suppresses the project's AGENTS.md, contaminating measurements. Claude probes run under /tmp with a scratch CLAUDE_CONFIG_DIR holding symlinked credentials.

What are the limitations of the question tool in headless runs?▼

claude -p and codex exec have no question tool, so decisions must be presented as a table in the final message instead. AskUserQuestion exists only when a host provides the permission callback, such as the Agent SDK with canUseTool.

Related Skills

Explore More Software Engineering Skills→
coseto6125coseto6125

validate-prompt-rules

A/B tests prompt rules against isolated no-rule baselines to measure actual behavior change.

Community
Advanced
Comfy-OrgComfy-Org
2.0k

publishing-a-new-package

Publishes monorepo packages to npm and verifies consumer installation end to end.

Official
Advanced
NVIDIA-TAONVIDIA-TAO
83

paidf-anomalygen

Generate synthetic anomaly images for manufacturing defect detection workflows.

Official
Advanced
madebywildmadebywild
12

harness-prompt

Generate CLAUDE.md, AGENTS.md, and copilot-instructions.md from a canonical system prompt.

Official
Intermediate