skills.rest
GLinnik21
GLinnik21COMMUNITY
Explore all skills by this author →
GitHub

fleet-plan

Plans parallel agent work across git worktrees with one shared television.

39|7|Updated Jul 3, 2026
Search Tags:#parallel agents#git worktree#webos#fleet planning#worker prompts#television lock#build disk management
One-click install
npx skills add https://github.com/GLinnik21/plx-native --skill fleet-plan-glinnik21
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: fleet-plan
Source: https://github.com/GLinnik21/plx-native/tree/main/.agents/skills/fleet-plan
Command: npx skills add https://github.com/GLinnik21/plx-native --skill fleet-plan-glinnik21

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running multiple AI agents in parallel on one repository fails in predictable ways: two lanes fight over the single physical TV, a shared git stash stack hands one lane another lane's work, worktrees get cut from the wrong base, and full cross-builds fill the disk. This Skill decides, before any worker launches, whether a fleet is worth it and how to partition it safely. ## Core Features & Use Cases - Fleet go/no-go decision: A checklist for when fanning out pays (disjoint files, independent compilation, host-verifiable work) versus doing it yourself. - TV lane assignment: At most one lane gets the physical television; all others verify on the simulator with their own SIM_DIR. - Worktree and disk hygiene: Cut each worktree from a named base, seed gitignored files like src/config.local.h, keep build trees outside the worktree, and run make check only. - Worker-prompt block: A paste-ready rules block covering base verification, device access, disk limits, and the no-stash rule, since workflow agents cannot be reached mid-run. - Use Case: You want three agents to work on separate UI modules of a webOS app at once. Use this Skill to cut three worktrees from origin/main, assign the TV to none of them, and paste the worker-prompt block into each lane. ## Quick Start Ask the agent to plan a parallel fleet for this repo, splitting the work into independent lanes with one lane allowed to use the television.

Frequently Asked Questions about fleet-plan

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run multiple AI agents in parallel on one git repository?▼

Cut one git worktree per lane from a named base commit, give each lane its own build directories via CARGO_TARGET_DIR and SIM_TDIR, and paste a rules block into every worker prompt. Only fan out when lanes touch disjoint files and verify independently.

When should I not fan out work across parallel agents?▼

Skip the fleet when lanes edit the same files, when one lane depends on symbols another is writing, or when all verification needs the single television. TV-bound lanes queue on the lock and finish no faster than one lane while paying worktree and merge overhead.

Why is git stash dangerous with multiple worktrees?▼

The stash stack is a repo-wide ref shared across all worktrees, so one lane's pop takes whatever is on top, including another lane's work. Commit with git add -A instead, or pin stashes to refs/rescue/<lane> using git stash create.

Can two agents share the physical television for testing?▼

No. The TV has one hardware video plane and one decoder, so two installs cannot play at once. At most one lane gets device access; all other lanes verify on the simulator, each with its own SIM_DIR.

How do I prevent parallel worktrees from filling the disk?▼

Restrict workers to make check, which never cross-builds FFmpeg, and move build trees outside the worktree with CARGO_TARGET_DIR. Run make disk before launching and tools/build-gc.sh --orphans after teardown to reclaim dead lane trees.

What files must a new worktree be seeded with?▼

Copy src/config.local.h for the PMS host and token, and tests/manifest.local.json only if using tests/run.py --server. Do not copy .tv-host, .tv-mac, or the vendor directory; the FFmpeg build now comes from a machine-wide cache.

Related Skills

Explore More Software Engineering Skills→
tranhieutttranhieutt
70

fork-join

Orchestrate parallel git worktrees with subagents and sequential merges.

Community
Advanced
wan-huiyanwan-huiyan
3

subagent-driven-branch-ref-froze-stranded-commits

Diagnose and recover commits stranded when a worktree HEAD advances past a frozen branch ref.

Community
Advanced
anthony-chaudharyanthony-chaudhary
18

dos-supervise-loop

Supervise DOS dispatch-loop worker fleets by replacing stalled leases.

Community
Advanced
bjconlanbjconlan

dispatching-parallel-agents

Dispatch separate agents to investigate independent problems in parallel.

Community
Intermediate