Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+22 moreClaude CodeCursorCodex CLIMuseDots
OpenClaw
Manus
Grok
Grok Bot+18 agentsWorking Mode
Switch Claude Code between five working modes (A-E), from solo work to up to 14 role subagents, with per-mode review, approval and model rules.
Free
Working Mode
Example session with this skill installed
/working-mode C sonnet --once
- Read your context and instructions
- Compiled the working mode
Mode changed for this session only (written to the session scratchpad; .claude/mode is untouched).
Mode change
| Before | After | |
|---|---|---|
| Mode | B - Selective | C - Full team |
| Subagents | analyst, test-writer, doc-writer | 9 role subagents: product-manager, analyst, architect, designer, developer, test-writer, qa, devops, doc-writer |
| Review | main session | qa subagent first pass, then the main session verifies critical findings |
| Approval | default = approval | choosing C = approval for these 9 subagents |
| Expected cost | 1.15-1.35x | 2.5-4x (estimate) |
Model line
Opus-defined subagents now run on
Sonnet. Subagents already defined on another model keep it. Every Agent call to an Opus subagent passes model: "sonnet", and the model that actually ran is reported.
Next steps
- Give me the task. I will list which subagents I am starting and why before starting them.
- At the end of the turn you get the number of subagents that ran and the estimated cost.
/working-mode C(without--once) makes C the project default;/working-mode Cwith no model word goes back to the defined models.
Connects securely to your tools. The creator never sees your data.
About this skill
Managing token usage and subagent orchestration across a complex project requires granular control over who works on what. This skill manages project-wide operating modes to balance speed, cost and depth of review. It gives you a structured way to move between solo execution and delegating to specialised Claude Code subagents, and to override the model those subagents run on.
Built on Claude Code subagents (the
Agenttool, plus theWorkflowtool in mode E). It does not use the experimental Agent Teams feature; the old team-based modes are archived and never offered.
The five modes
| Mode | Subagents | Review |
|---|---|---|
| A - Skill | none, everything in the main conversation | the main session |
| B - Selective (default) | analyst, test-writer, doc-writer | the main session |
| C - Full team | 9 role subagents (B plus product-manager, architect, designer, developer, qa, devops) | qa subagent, then the main session |
| D - Wide | 14 role subagents (C plus security, data, coverage-auditor, observability, e2e-writer) | qa + security/coverage-auditor, then the main session |
| E - Fan-out | D's 14 roles run in parallel through Workflow | qa subagent, then the main session |
The main session is always the orchestrator; there is no separate orchestrator subagent.
What it does
- Switch modes between five levels (A-E) that decide which subagents may be started and who reviews their output.
- Model overrides run every Opus-defined subagent on Sonnet or Haiku (
/working-mode C sonnet) to control token spend. - Session persistence:
--oncestores the mode in the session scratchpad (survives context compaction); without it the mode is written to.claude/modeand shared through the repo. - Subagent gating: choosing a mode is the approval for its subagent set; at the end of each turn the number of subagents that ran and the estimated cost are reported.
- Role handoffs follow
role-selection.md: who does what, ordering, audits and when to stop.
What's inside
SKILL.mdandmodes/(A-E definitions, role selection, autonomous runs)agents/: the 14 subagent definitions
How it works
- At session start it asks for the mode and the model override (applied for this session only).
/working-modeshows the current mode, its subagent set, reviewer and cost multiplier./working-mode <A-E> [model] [--once]changes the mode for the project or just this session.- The main session then follows that mode's rules for every subagent call.
Why this beats prompting it yourself
Instructions given in chat are lost after context compaction. This skill keeps the mode in a file, so the session never silently drifts back to a more expensive or less reviewed way of working.
Use cases
- Run routine work on Sonnet subagents instead of Opus.
- Bring in 14 role subagents (mode D) for a large cross-cutting change.
- Stay in mode A (no subagents) when you want to control every change yourself.
Known limitations
- Modes B-E need the bundled subagent definitions copied into
.claude/agents/(or~/.claude/agents/). - Model words are limited to sonnet, haiku, fable and opus.
- Cost multipliers are estimates.
- A mode change is not retroactive.
How to use
Skills on Agensi run inside your AI agent. Copy the prompt, paste it into your agent, and it connects to Agensi and installs this skill for you. If a payment is needed, your agent shows it to you first.
Reviews
1 people have installed this skill.
Trust & safety
Security scanned
Verified clean today
- Free to download with an account