dabingfy-dot/solo-to-team-skills

What this is

Solo to Team is a set of Agent Skills for the case where one person coordinates several long-running AI assistants. It targets four recurring failures: tasks that nobody closes out, "done" claims without evidence, instructions that only their author can follow, and the same mistake repeated twice. The package contains four skills, a task-register template, a walkthrough example, and desensitized case notes, released under MIT. The headline method is a single loop: set a goal, split work, hand off, check, then decide — with a register line recording who is doing the task, who receives it, and who checks it on schedule README introduction.

Two skills are adapted from earlier work under MIT (noted in NOTICE); two — the beginner-facing ops writing guide and the ops case notes — are original to this project skill list.

Quick start

The skills are plain directories with a SKILL.md; there is no app, and hooks: {} means no built-in scheduling or wake-up. Pick the path that fits you.

Without an agent host, start from the template and a table: copy the dispatch register line from templates/task-register-line.md, write one "to be received" line before dispatching, check at the time you set, then record "result received" and only mark "accepted" after checking evidence. The walkthrough in examples/ runs a small meeting-notes task from input to evidence to acceptance, including one rejection.

With an agent host, install the four skill directories into that host's skills folder and call them by name. Cursor is a compact example — copy the skill directories into ~/.cursor/skills/ and call them from the Agent or CLI. Codex uses the bundled marketplace manifest:

cd /path/to/solo-to-team-skills
codex plugin marketplace add "$(pwd)"
codex plugin add solo-to-team-skills@solo-to-team-skills

Then open a new session and ask it, for example, to use dispatch-accept-verify to register a pending task Codex setup. Other hosts have their own paths and call styles: WorkBuddy, DeepSeek Harness, CodeBuddy, Kimi Work, QwenWork, and Grok Bot (manual save of the skill text as a private skill). Claude Code has plugin structure prepared but is not listed as supported yet host table.

When copying by hand, copy the whole skill folder, not just SKILL.md: dispatch-accept-verify needs its templates/ directory alongside it.

Constraints

• Human-in-the-loop by design. Spending money, sending messages, deleting files, logging in, and confirming payment are written as "do this yourself"; the skills do not act on your behalf. Accept or reject decisions are yours.
• No automation included. Scheduling, reminders, cross-assistant messaging, and wake-ups depend on the host; where the host lacks them, a person must check at the set time. Do not treat the package as auto-dispatching or auto-acknowledging.
• Host support is narrow and per-product. "Supported" means one specific install-and-call route has been recorded, not that the same vendor's other clients work. CodeBuddy ≠ WorkBuddy, Kimi Work ≠ Kimi Code CLI, QwenWork ≠ Tongyi Lingma. Other tools are untested.
• Performance claims are bounded. The method ran on an internal daily-update pipeline for roughly two-plus weeks with three reviews; there is no controlled experiment showing lower team error rates or fewer reminders. The eight hosts each have one limited acceptance record, not a stability guarantee.
• Draft status. The README marks the package v0.9-draft and subject to change before public release.

HotRepo's take

The project entered HotRepo's new-project list on 2026-11-07 with 106 stars and 10 forks gained in about 10.7 hours of observed growth. That is a fast first-day pickup for a documentation-and-skills repository with no programming-language code, which fits the current interest in agent skill packs rather than any measured adoption. What distinguishes it is the explicit refusal to overclaim: the README states that the eight hosts each have one limited acceptance record, that the meeting-notes walkthrough is a teaching sample rather than an observed run, and that a ~10 minute walkthrough was timed only in Cursor CLI from an already-prepared environment at about 30 seconds. Readers should evaluate it as a method template with case notes, not as a proven team-automation product.