Menú
Use only when user explicitly invokes $loop.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Basado en la clasificación ocupacional SOC
Plan and run Codex-native dynamic workflows for complex tasks that benefit from explicit orchestration, goal mode, subagents or simulated work packets, approval gates, integration, verification, and reusable workflow artifacts. Use when the user invokes this skill, asks for dynamic workflows, subagents, parallel agents, swarm-like work, Goal Maker orchestration, large audits, migrations, multi-track research plus implementation, or Claude Code-style fan-out/fan-in workflows.
Direct browser control via CDP. Use when the user wants to automate, scrape, test, or interact with web pages. Connects to the user's already-running Chrome.
This skill should be used for browser automation tasks using Chrome DevTools Protocol (CDP). Triggers when users need to launch Chrome with remote debugging, navigate pages, execute JavaScript in browser context, capture screenshots, or interactively select DOM elements. No MCP server required.
Use Orca's computer-use CLI to inspect and control local desktop apps through accessibility trees, screenshots, and safe UI actions. Use when an agent needs to list desktop apps, get an app state, read visible UI, click, type, press keys, scroll, drag, set values, or perform app accessibility actions. Triggers include "computer use", "orca computer", "list apps", "get app state", "read Spotify", "read Slack", "click app", "type text", "press key", "set value", "scroll app", "drag app", and desktop app interaction tasks.
Use the `orca` CLI to drive a running Orca editor — manage Orca worktrees; create and manage scheduled automations; create, read, and run shell commands in Orca-managed terminals; and automate Orca's built-in browser (snapshot/click/fill/screenshot/tabs). Use this instead of raw `git worktree`, ad hoc shell PTYs, or Playwright whenever the task touches Orca state. Coding agents inside an Orca worktree should also use it to keep the worktree comment fresh at meaningful checkpoints. Boundary with `orchestration`: if the recipient of a terminal write is another AI agent (Claude Code, Gemini, Codex, a worker), use `orchestration` — it is the only correct way to send messages, nudges, replies, or task hand-offs to agents. orca-cli writes are for non-agent terminals (shells, build/test commands); reading or `wait`ing on any terminal — including agent terminals — stays in orca-cli.
Use Orca orchestration for structured multi-agent coordination: threaded messages, blocking ask/reply flows, task dispatch, worker_done/escalation waits, task DAGs, decision gates, coordinator loops, or decomposing work across agents. Use `orca-cli` instead for ordinary terminal control, lightweight terminal prompts, shell commands, Orca worktree management, reading or waiting on terminals, and automation of the browser embedded inside Orca. Use Computer Use for browser windows, webviews, Orca app UI, or desktop UI outside Orca's embedded browser.
| name | loop |
| description | Use only when user explicitly invokes $loop. |
This is a control protocol for running subagents in a loop pattern to "reconcile" repo to fulfill the user request, which may be a complex task or target state.
When working in this pattern, subagents own all task execution, you are the orchestrator, you must:
Build a concise initial_prompt for the next LLM. Include user's request and context from previous conversation; do not inspect the repository to enrich it.
Structure initial_prompt with these sections:
## Original User Request
<The user's original request, unchanged.>
## Important Context, Constraints, and User Preferences
<Supplied from previous conversation, if user's request is not self-contained.>
## Critical Data, Examples, and References
<Supplied from previous conversation, if user's request is not self-contained.>
Keep it concise, structured, and focused on helping the next LLM seamlessly continue the work.
$loop control flow instructions MUST BE EXCLUDED in initial_prompt, otherwise subagent will loop again.
Other skills like $xxx should be kept as is if they are part of the user request.
2.1. Resolve the current git commit with git rev-parse HEAD, skip if not in a git repository.
2.2. Create handoff file:
.codexpotter/projects_v2/{yyyy}_{mm}_{dd}_{slug}.md
where:
{yyyy} {mm} {dd} is current local date{slug} is a short descriptive name generated from the user request (e.g., "add_login_feature", "fix_bug_123").Do not overwrite an existing project directory.
2.3. Write handoff file in this shape:
---
status: initial
finite_incantatem: false
git_commit: <current_git_commit> <-- leave empty if not in a git repository
---
# Overall Goal
<initial_prompt>
## In Progress
## Todo
## Done
Run at most 6 rounds by default (user can change via --rounds N). In each round, do the following:
Start one subagent using the potter_worker agent.
Prompt the subagent with handoff file path only:
Work according to this handoff file: <path to handoff md file>
Wait for the subagent to finish.
Report last subagent message, keep it unchanged.
Read handoff file, stop if the frontmatter contains the literal value finite_incantatem: true
If potter_worker agent is not available: This is a critical issue. Fail fast, do not use other alternative agents. Ask user to follow https://github.com/breezewish/CodexPotter to setup.
If handoff file cannot be read: Stop immediately and report the control-state problem. Do not repair it in the parent agent.
If finite_incantatem is not true after a round and there are more rounds left:
Start the next round (even if subagent reports all tasks done).
After the loop stops, report these to user (using bullet points):
Do not add implementation analysis, code review, extra verification, or recommendations beyond the loop outcome.
If what user passed in is an existing handoff file to resume (iterate more rounds), you should:
open, finite_incantatem to false.If current loop was paused / interrupted and user wants to continue, you could simply send continue prompt to the subagent.