Menú
Use when executing task lists where each task requires isolated execution
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
Use when running autonomous unattended loops — cross-session execution of DAG tasks without human intervention
Use when converting design documents to structured specs, when spec quality is below threshold, or when requirements need formal acceptance criteria
Use when you need to verify work is complete before making completion claims
Use when maintaining ArcForge itself by creating, editing, or verifying ArcForge skills before deployment
Use when optional learning is enabled and observations should become reviewable candidates, inactive drafts, and explicitly activated artifacts.
Use when user asks about behavioral patterns, requests instinct status, or wants to confirm/contradict a detected pattern
| name | arc-agent-driven |
| description | Use when executing task lists where each task requires isolated execution |
Execute plan by dispatching fresh subagent per task, with two-stage review after each: spec compliance review first, then code quality review.
Core principle: Fresh subagent per task + two-stage review (spec then quality) = high quality, fast iteration
digraph when_to_use {
"Have task list?" [shape=diamond];
"Tasks mostly independent?" [shape=diamond];
"Want automated execution?" [shape=diamond];
"arc-agent-driven" [shape=box];
"arc-executing-tasks" [shape=box];
"Use writing-tasks first" [shape=box];
"Have task list?" -> "Tasks mostly independent?" [label="yes"];
"Have task list?" -> "Use writing-tasks first" [label="no"];
"Tasks mostly independent?" -> "Want automated execution?" [label="yes"];
"Tasks mostly independent?" -> "arc-executing-tasks" [label="no - need human checkpoints"];
"Want automated execution?" -> "arc-agent-driven" [label="yes"];
"Want automated execution?" -> "arc-executing-tasks" [label="no - want control"];
}
vs. arc-executing-tasks:
digraph process {
rankdir=TB;
subgraph cluster_per_task {
label="Per Task";
"Dispatch implementer subagent" [shape=box];
"Implementer asks questions?" [shape=diamond];
"Answer questions" [shape=box];
"Implementer implements, tests, commits, self-reviews" [shape=box];
"Dispatch spec reviewer subagent" [shape=box];
"Spec compliant?" [shape=diamond];
"Implementer fixes spec gaps" [shape=box];
"Dispatch code quality reviewer subagent" [shape=box];
"Quality approved?" [shape=diamond];
"Implementer fixes quality issues" [shape=box];
"Mark task complete" [shape=box];
}
"Read tasks, create TodoWrite" [shape=box];
"More tasks?" [shape=diamond];
"Multiple independent issues?" [shape=diamond];
"Use arc-dispatching-parallel for fixes" [shape=box];
"Dispatch final code reviewer" [shape=box];
"Use arc-finishing" [shape=box style=filled fillcolor=lightgreen];
"Read tasks, create TodoWrite" -> "Dispatch implementer subagent";
"Dispatch implementer subagent" -> "Implementer asks questions?";
"Implementer asks questions?" -> "Answer questions" [label="yes"];
"Answer questions" -> "Dispatch implementer subagent";
"Implementer asks questions?" -> "Implementer implements, tests, commits, self-reviews" [label="no"];
"Implementer implements, tests, commits, self-reviews" -> "Dispatch spec reviewer subagent";
"Dispatch spec reviewer subagent" -> "Spec compliant?";
"Spec compliant?" -> "Implementer fixes spec gaps" [label="no"];
"Implementer fixes spec gaps" -> "Dispatch spec reviewer subagent";
"Spec compliant?" -> "Dispatch code quality reviewer subagent" [label="yes"];
"Dispatch code quality reviewer subagent" -> "Quality approved?";
"Quality approved?" -> "Multiple independent issues?" [label="no"];
"Multiple independent issues?" -> "Use arc-dispatching-parallel for fixes" [label="yes"];
"Multiple independent issues?" -> "Implementer fixes quality issues" [label="no"];
"Use arc-dispatching-parallel for fixes" -> "Dispatch code quality reviewer subagent";
"Implementer fixes quality issues" -> "Dispatch code quality reviewer subagent";
"Quality approved?" -> "Mark task complete" [label="yes"];
"Mark task complete" -> "More tasks?";
"More tasks?" -> "Dispatch implementer subagent" [label="yes"];
"More tasks?" -> "Dispatch final code reviewer" [label="no"];
"Dispatch final code reviewer" -> "Use arc-finishing";
}
Max review cycles: 3 per reviewer. If not converging, escalate to human with summary of unresolved issues.
Two ways to dispatch each role, depending on what your platform supports:
Pre-built agents (when your platform supports named subagents — e.g. Claude
Code's agents/): they bundle tool isolation and methodology.
implementer — TDD implementation with full write accessspec-reviewer — Spec compliance verification (read-only)quality-reviewer — Code quality assessment (read-only + test runner)Templates (cross-platform fallback — for custom prompts or when named agents aren't available):
./implementer-prompt.md - Implementer prompt with placeholders./spec-reviewer-prompt.md - Spec compliance review prompt./code-quality-reviewer-prompt.md - Code quality review prompt (references arc-requesting-review)This skill is platform-agnostic: it needs only the ability to run a fresh subagent per task. The named agents above are a Claude Code convenience; on any platform (Codex, Gemini CLI, OpenCode, or Claude Code) you can dispatch each role from the templates instead, using whatever subagent mechanism your harness provides. The templates carry the full role prompt, so the workflow — fresh implementer per task, then spec review, then quality review — is identical regardless of how the subagent is launched.
You: I'm using arc-agent-driven to execute these tasks.
[Read task file: docs/tasks/sync-command-tasks.md]
[Create TodoWrite with all 5 tasks]
Task 1: Add SyncResult dataclass
[Dispatch implementer subagent with full task text + context]
Implementer: "Before I begin - should SyncResult be in models.py or a new file?"
You: "In models.py with other dataclasses"
Implementer:
- Implemented SyncResult dataclass
- Added tests, 3/3 passing
- Self-review: All good
- Committed: abc1234 "feat(models): add SyncResult dataclass"
[Dispatch spec compliance reviewer]
Spec reviewer: ✅ Spec compliant - all fields present, nothing extra
[Dispatch code quality reviewer]
Code reviewer: Strengths: Clean, typed. Issues: None. Approved.
[Mark Task 1 complete]
Task 2: Add sync CLI command
...
[After all tasks]
[Dispatch final code reviewer for entire implementation]
Final reviewer: All requirements met, architecture solid
Done! Completion pipeline:
1. Run arc-verifying — confirm all requirements met and tests pass
2. Use arc-finishing (Step 0 discriminates on .arcforge-epic) to decide merge/PR/keep/discard
The full agent roster for arc-agent-driven workflows:
| Agent | Role | Model | Access |
|---|---|---|---|
| implementer | TDD implementation | sonnet | Read, Write, Edit, Bash, Grep |
| spec-reviewer | Spec compliance check | sonnet | Read, Grep, Glob |
| quality-reviewer | Code quality review | sonnet | Read, Grep, Glob, Bash |
vs. Manual execution:
vs. arc-executing-tasks:
Quality gates:
Never:
If subagent asks questions:
If reviewer finds issues:
Required workflow skills:
.arcforge-epic) - Complete development after all tasksSubagents should use:
Alternative workflow: