Menú
Dispatch scoped coding personas through Agent Fleet review pipelines — parallel runs from Hermes with pluggable execution backends
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
| name | coding-fleet |
| description | Dispatch scoped coding personas through Agent Fleet review pipelines — parallel runs from Hermes with pluggable execution backends |
| metadata | {"hermes":{"category":"autonomous-ai-agents","requires_toolsets":["coding_fleet"]}} |
When the user mentions coding fleet, PR analyzer, or gives persona + pipeline for a repo path: call the appropriate tool in your first turn.
For repos with pr_loop.enabled: true in .agent-fleet.yaml:
Poll all open fleet/* PRs once:
{
"workspace": "/absolute/path/to/repo",
"mode": "once"
}
Run full lifecycle for one PR (address review findings → wait CI → merge):
{
"workspace": "/absolute/path/to/repo",
"mode": "pr",
"pr_number": 42,
"skip_review_wait": true
}
Tool: coding_fleet_pr_loop
Requires gh authenticated, CURSOR_API_KEY, and pr_loop + pr_review in repo config. PRs touching .github/workflows/ auto-park for human review.
Product + technical discovery before dispatch:
{
"workspace": "/absolute/path/to/repo",
"depth": "light",
"product_context": "optional business context"
}
Tool: coding_fleet_scout
Then: coding_fleet_scope → coding_fleet_dispatch.
Rank dispatchable tasks from open issues + large-file hotspots:
{
"workspace": "/absolute/path/to/repo",
"github_repo": "owner/repo",
"issue_limit": 20
}
Tool: coding_fleet_scope
Then dispatch top-ranked tasks with coding_fleet_dispatch.
PR reviews automatically include a quality pass using the bundled thermo-nuclear-code-quality-review skill (maintainability, 1k-line guardrails, code-judo).
{
"goal": "<what to change>",
"workspace": "/absolute/path/to/repo",
"persona": "coder",
"pipeline": "code_review",
"context": "<constraints, file hints, errors>"
}
Repos with pr_review.use_in_code_review: true in .agent-fleet.yaml automatically use the two-pass PR analyzer (Composer 2.5) for the review phase instead of the generic reviewer.
When pr_loop.enabled: true, code_review dispatch also auto-fixes review/verify failures, pushes the branch, and runs coding_fleet_pr_loop on the new PR.
Use when reviewing an existing branch or worktree diff:
{
"workspace": "/absolute/path/to/repo",
"base_branch": "main",
"output_format": "json"
}
Tool: coding_fleet_pr_review
Or dispatch with pipeline pr_review:
{
"goal": "Analyze current branch diff",
"workspace": "/absolute/path/to/repo",
"persona": "pr-analyzer",
"pipeline": "pr_review"
}
default_backend | Key required | Model default |
|---|---|---|
cursor (default) | CURSOR_API_KEY | composer-2.5 |
kimi (optional) | KIMI_API_KEY | kimi-for-coding |
simple — implement onlycode_review — implement → scope → verify? → PR analyzer review (when repo configured)pr_review — analyze diff only (two-pass backend/security + frontend)full — plan → research → implement → verify → reviewAdd to .agent-fleet.yaml:
pr_review:
enabled: true
use_in_code_review: true
overlay: agents/pr_review_overlay.md
area_prefixes:
frontend: [frontend/, web/]
backend: [src/, api/, packages/]
See docs/NEW-REPO.md in the agent-fleet repo for full setup.
~/.agent-fleet/fleet.yamlCURSOR_API_KEY in the environmentKIMI_API_KEY + kimi-cli on PATH, default_backend: kimipip install -e /path/to/agent-fleetDesigns or reviews CLIs so coding agents can run them reliably: non-interactive flags, layered --help with examples, stdin/pipelines, fast actionable errors, idempotency, dry-run, and predictable structure. Use when building a CLI, adding commands, writing --help, or when the user mentions agents, terminals, or automation-friendly CLIs.
Sketch types, signatures, and module structure before code, then stay in the loop while implementation fills in. Use for /architect, 'architect this', 'design this', or non-trivial work where jumping to code would lock in the wrong shape.
Spawn N parallel candidates at the same task, pick a base, graft the strongest parts of the losers into it. Use for /arena, 'arena this', 'throw it in the arena', or when one attempt at a non-trivial artifact would lock in the wrong shape.
Use for "how does X work", code walkthroughs before changing something, and placement / ownership / layering questions ("where should this live", "which package owns this", "is this the right layer"). Explains subsystem architecture, runtime flow, onboarding mental models. Can critique architecture. Use why for motivation.
Use for "interrogate", "adversarial review", "multi-model review", "challenge this", "stress test this code", "find blind spots", or "tear this apart". Four LLM reviewers challenge changes from independent angles.
poteto's agent style for concise, detailed responses, deliberate subagents, unslopped prose, simple code, and verified work. Use for poteto, /poteto-mode, or requests to work in this style.