/swarm
Fan out a parallel review across the workspace using Kimi's AgentSwarm tool, bounded by a hard wall-clock budget. Read-only by default (enforced by the same PreToolUse hook as review, applied to every subagent); --write fans out edits in a throwaway worktree and returns a patch.
$ npx -y skills add linxule/kimi-plugin-cc --agent claude-codeShips with kimi. Installing the plugin gets this command.
How it fires
How this command gets triggered: by you, by Claude, or both.
- Fires itselfClaude auto-loads it when your prompt matches the work.
- You can call itInvoke it directly when you want it.
- Slash command
/swarm
Context preview
What this command does when you run it.
Fan out a parallel review across the workspace using Kimi's AgentSwarm tool, bounded by a hard wall-clock budget. Read-only by default (enforced by the same PreToolUse hook as review, applied to every subagent); --write fans out edits in a throwaway worktree and returns a patch.
Command definition
swarm.mddescription: Fan out a parallel review across the workspace using Kimi's AgentSwarm tool, bounded by a hard wall-clock budget. Read-only by default (enforced by the same PreToolUse hook as review, applied to every subagent); --write fans out edits in a throwaway worktree and returns a patch.
argument-hint: "[--write] [--budget <30m|1h>] [--cap <N>] [--max-concurrency <N>] [-m <model>] <objective>"
disable-model-invocation: true
Run the companion with any user-supplied flags appended after `task swarm`:
`${CLAUDE_PLUGIN_ROOT}/scripts/companion.sh task swarm <args>`
`/kimi:swarm` is a **parallel fan-out**: Kimi uses the `AgentSwarm` tool to fan the work out across subagents (one per file/module/question), then consolidates into one markdown report. **By default it is read-only**, enforced by the same PreToolUse hook as `/kimi:review` — the hook runs under the `swarm` label (read-only tool set plus `AgentSwarm`), and **every spawned subagent inherits that label and fires the same hook** (policy index 0), so a subagent's write/edit/shell call is denied exactly like a single-turn review's.
**`--write`** turns it into a write-capable fan-out: the coordinator and `coder` subagents run inside an **ephemeral throwaway git worktree off your HEAD**, edit disjoint targets there, and the result is captured as a **reviewable patch** (written to a `.patch` file whose path is printed in the report). Writes are confined to that worktree by the `swarm-write` hook label (rescue-grade allowlist, scoped to a forge-proof trusted worktree root — not the payload cwd); git mutation and out-of-worktree writes are denied; **the plugin never applies or commits — you own the merge.** Your real working tree is never touched.
Supported flags:
- `--write` — fan out EDITS (not just review). Requires kimi-code **>= 0.18.0**, a git repo with a committed HEAD, and the PreToolUse hook. Bases the worktree on HEAD: **uncommitted changes are NOT included** (you'll get a warning) — commit or stash first if the swarm needs them. `--write` is also reachable via the model-invocable `kimi-swarm-write` subagent, with strict triggering (many disjoint write targets AND explicit fan-out intent); auto-dispatch widens no write surface and keeps every bound, and the slash command itself stays human-only.
- `--budget <duration>` — HARD wall-clock ceiling (e.g. `30m`, `1h`, `90s`; bare number = minutes). Default 30m. The always-on bound on cost/runaway.
- `--cap <N>` — SOFT cap on TOTAL subagent count: injected into the prompt as a model instruction. Advisory, not hook-enforced (the hook is stateless and can't count subagents), so the model may exceed it. Bounds lifetime total, not peak parallelism.
- `--max-concurrency <N>` — HARD ceiling on how many subagents run AT ONCE, on kimi-code **0.18.0+** (exported as `KIMI_CODE_AGENT_SWARM_MAX_CONCURRENCY`; older binaries ignore it). **Defaults to 4 for read, 1 for `--write`** (writes serialize by default since disjoint-target partitioning is prompt-only) — pass a value to widen or throttle. Distinct from `--cap`: concurrency (simultaneous) ≠ total count (lifetime).
- `-m`, `--model <name>`
Prototype limitations:
- **No `--background` flag** — the runtime has no detached-worker mode for swarm. That is separate from how the *caller* runs the shell command: for `--write` (30m default budget vs a 10-minute foreground Bash cap) detaching the call is expected; for a read-only fan-out the default stays foreground and detaching is a per-request choice. `--budget` and `--max-concurrency` stay finite either way, and budget expiry still captures the patch/report. To stop a run, just say so — the stop path is `companion.sh cancel` with no id (it targets the latest running job for this repo). For `--write`, prefer that over Esc: an interrupt can kill the run mid-teardown and lose the captured patch.
- Read-only swarm requires kimi-code **>= 0.12.0** (the `AgentSwarm` tool); `--write` requires **>= 0.18.0** (the hard concurrency cap). Both **refuse** without the `/kimi:setup` PreToolUse hook (a fan-out with no enforcement is an N-fold blast radius).
Return the companion stdout verbatim.
Read more
description: Fan out a parallel review across the workspace using Kimi's AgentSwarm tool, bounded by a hard wall-clock budget. Read-only by default (enforced by the same PreToolUse hook as review, applied to every subagent); --write fans out edits in a throwaway worktree and returns a patch. argument-hint: "[--write] [--budget <30m|1h>] [--cap <N>] [--max-concurrency <N>] [-m <model>] <objective>" disable-model-invocation: true
Run the companion with any user-supplied flags appended after `task swarm`:
`${CLAUDE_PLUGIN_ROOT}/scripts/companion.sh task swarm <args>`
`/kimi:swarm` is a **parallel fan-out**: Kimi uses the `AgentSwarm` tool to fan the work out across subagents (one per file/module/question), then consolidates into one markdown report. **By default it is read-only**, enforced by the same PreToolUse hook as `/kimi:review` — the hook runs under the `swarm` label (read-only tool set plus `AgentSwarm`), and **every spawned subagent inherits that label and fires the same hook** (policy index 0), so a subagent's write/edit/shell call is denied exactly like a single-turn review's.
**`--write`** turns it into a write-capable fan-out: the coordinator and `coder` subagents run inside an **ephemeral throwaway git worktree off your HEAD**, edit disjoint targets there, and the result is captured as a **reviewable patch** (written to a `.patch` file whose path is printed in the report). Writes are confined to that worktree by the `swarm-write` hook label (rescue-grade allowlist, scoped to a forge-proof trusted worktree root — not the payload cwd); git mutation and out-of-worktree writes are denied; **the plugin never applies or commits — you own the merge.** Your real working tree is never touched.
Supported flags:
- `--write` — fan out EDITS (not just review). Requires kimi-code **>= 0.18.0**, a git repo with a committed HEAD, and the PreToolUse hook. Bases the worktree on HEAD: **uncommitted changes are NOT included** (you'll get a warning) — commit or stash first if the swarm needs them. `--write` is also reachable via the model-invocable `kimi-swarm-write` subagent, with strict triggering (many disjoint write targets AND explicit fan-out intent); auto-dispatch widens no write surface and keeps every bound, and the slash command itself stays human-only.
- `--budget <duration>` — HARD wall-clock ceiling (e.g. `30m`, `1h`, `90s`; bare number = minutes). Default 30m. The always-on bound on cost/runaway.
- `--cap <N>` — SOFT cap on TOTAL subagent count: injected into the prompt as a model instruction. Advisory, not hook-enforced (the hook is stateless and can't count subagents), so the model may exceed it. Bounds lifetime total, not peak parallelism.
- `--max-concurrency <N>` — HARD ceiling on how many subagents run AT ONCE, on kimi-code **0.18.0+** (exported as `KIMI_CODE_AGENT_SWARM_MAX_CONCURRENCY`; older binaries ignore it). **Defaults to 4 for read, 1 for `--write`** (writes serialize by default since disjoint-target partitioning is prompt-only) — pass a value to widen or throttle. Distinct from `--cap`: concurrency (simultaneous) ≠ total count (lifetime).
- `-m`, `--model <name>`
Prototype limitations:
- **No `--background` flag** — the runtime has no detached-worker mode for swarm. That is separate from how the *caller* runs the shell command: for `--write` (30m default budget vs a 10-minute foreground Bash cap) detaching the call is expected; for a read-only fan-out the default stays foreground and detaching is a per-request choice. `--budget` and `--max-concurrency` stay finite either way, and budget expiry still captures the patch/report. To stop a run, just say so — the stop path is `companion.sh cancel` with no id (it targets the latest running job for this repo). For `--write`, prefer that over Esc: an interrupt can kill the run mid-teardown and lose the captured patch.
- Read-only swarm requires kimi-code **>= 0.12.0** (the `AgentSwarm` tool); `--write` requires **>= 0.18.0** (the hard concurrency cap). Both **refuse** without the `/kimi:setup` PreToolUse hook (a fan-out with no enforcement is an N-fold blast radius).
Return the companion stdout verbatim.
Use Kimi as Claude Code's second reviewer, independent thinker, and delegated worker — without building your own multi-agent stack. This is a Claude Code plugin that drives the kimi-code CLI (the Node.js successor to Kimi CLI) as a subprocess.
Repo: linxule/kimi-plugin-cc
Other commands on kimi.
- /README
Claude Code slash command markdown. Each file is thin — it routes through `scripts/companion.sh` to the Node runtime under `runtime/`.
Open command - /ask
Ask Kimi a read-only question in free-form prose mode.
Open command - /cancel
Cancel an active plugin-managed Kimi job for the current repository.
Open command - /challenge
Run a read-only Kimi challenge review that challenges assumptions and surfaces safer alternatives.
Open command - /pursue
Hand Kimi an objective to pursue AUTONOMOUSLY across multiple turns (experimental goal mode), bounded by a hard wall-clock budget. Write-capable, gated by the same PreToolUse hook as rescue.
Open command - /replay
Re-render a stored Wire event log for a completed plugin-managed Kimi job.
Open command

