Multi cli spawn

Skill ImL1s/agent-cli-skills/skills/multi-cli-spawn

Fan out one brief to multiple coding CLIs in parallel (agy, grok, claude, codex, kimi, qwen) via spawn.sh with PID-file / process-group kill safety and per-seat .status files (RUNNING → DONE|BLOCKED). Use for multi-LLM council, parallel second opinions, or cross-provider review. Prefer foreground wait; for --no-wait poll terminal .status — never treat missing .md alone as failure. Always --no-git; orchestrator owns commits.From its SKILL.md

Install

npx -y skills add ImL1s/agent-cli-skills --skill multi-cli-spawn

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.6 KB, 947 tokens by cl100k_base, as published. Nobody here has run it

Multi-CLI Spawn (cross-provider parallel seats)

When to use

Need several providers to answer the same brief in parallel
Multi-LLM council / second opinions / cross-vendor review
You already have (or will install) the per-CLI *-cli-agent wrappers

This is not a substitute for vendor-native multi-agent inside one CLI (-T / ultracode / teamwork). Use those for fan-out within one provider; use this skill to fan-out across providers.

Why

Each vendor has native multi-agent inside one CLI (agy teamwork, Claude ultracode, Codex subagents, Grok spawn_subagent, …). This skill is the cross-CLI layer: one brief → N providers → N answer files.

Usage

./spawn.sh --outdir /tmp/council-$$ \
  -f /tmp/brief.md \
  --seat agy:readonly \
  --seat claude:ultracode \
  --seat codex:team \
  --seat grok \
  --seat kimi:readonly \
  --seat qwen:readonly \
  -C /path/to/repo \
  -t 1200s

Seat forms: name, name:readonly, name:ultracode|team, name:readonly,team, name@label (unique output key when duplicating a CLI).

Seat modifiers (comma-separated after :): readonly/r, team/ultracode/T.

Outputs

File	Meaning
`<key>.log`	Live CLI stdout/stderr
`<key>.md`	Final answer, or `BLOCKED:…` stub (written when the seat finishes)
`<key>.pid`	Seat process-group id (kill target)
`<key>.rc`	Seat exit code
`<key>.status`	`RUNNING` → `DONE rc=0` \| `BLOCKED rc=N` (atomic)
`spawn.pid` / `spawn.status`	Parent: `RUNNING` → `DONE`, or `SPAWNED` if `--no-wait`
`brief.md`	Copied/normalized brief inside outdir

Waiting contract

Default: do not poll. Run spawn.sh in the foreground (wait mode). Its exit code and CLI_AGENT_RESULT: PASS|BLOCKED <seat> lines are the completion signal.
--no-wait / external watcher: wait until every seat .status is terminal (DONE / BLOCKED), e.g.:
```
while grep -qlE '^RUNNING' "$OUTDIR"/*.status 2>/dev/null; do sleep 15; done
```
Do not treat missing *.md or a dead *.pid alone as "no answer" — .md appears only when the seat finishes; pidfiles can look dead if the parent spawn was killed while seats continue.
Kill only via numeric PIDs in seat *.pid files (process groups). Skip spawn.pid when iterating kill targets unless you intend to stop the parent. Never pkill -f long patterns.

Hard rules

PID files only — never pkill -f with a pattern that also appears in the launcher argv
Quota fail-open — provider errors / empty output → BLOCKED stub; do not retry-loop
agy needs PTY — handled inside agy-exec.sh
Orchestrator owns git — spawn always passes --no-git
Parallel writers should use separate worktrees (-C per seat) to avoid collisions
Seat rc != 0 is BLOCKED (missing binary is not a council vote)

Synthesize

Read *.md only for seats whose .status is terminal. Strictest blockers win. Divergent findings are high value. Re-run only BLOCKED seats after the user refreshes quota/auth.

What ships with it: 4 files

19.8 KB alongside SKILL.md, 4 of them executable

lib/

common.shruns6.8 KB
parse_stream_json.pyruns3.5 KB
pty_run.pyruns1.7 KB

spawn.shruns7.7 KB

Gives 0 of the 12 instructions most context ai engineering skills give in 947 tokens

Counted across 1,328 of the 2,349 authors here whose files we hold, read 2026-09-06

Dispatch a fresh subagent for each taskin 76 of 1328, across 59 files
Perform spec compliance review before code quality reviewin 44 of 1328, across 34 files
Dispatch a final code reviewer after all tasksin 38 of 1328, across 26 files
Answer subagent questions before allowing implementationin 36 of 1328, across 26 files
Use the least powerful model capable of the taskin 33 of 1328, across 26 files
Create a TodoWrite list for all tasksin 32 of 1328, across 22 files
Perform a task review after each implementationin 31 of 1328, across 24 files
Extract all tasks and context from the planin 29 of 1328, across 20 files
Provide full task text to subagentsin 28 of 1328, across 20 files
Use git worktrees for isolated workspacesin 25 of 1328, across 20 files
Specify the model explicitly when dispatching a subagentin 23 of 1328, across 18 files
Execute all tasks from the plan without stoppingin 21 of 1328, across 16 files

Said here and by no other author read

Run spawn.sh in foreground by default
Use --no-git for all spawned agents
Use separate worktrees for parallel writers
Poll .status files for terminal state when using --no-wait
Read .md files only when .status is terminal
Treat seat exit code non-zero as BLOCKED

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.