Barney241/loadout

The agent skills, hooks and tools I carry to every host and every project.
Claude Code is the main driver; Codex (GPT Sol) and Antigravity (Gemini)
review through T3 Code's
delegatetask. Built on tstack (see
NOTICE).
Everything here is generic. Anything true of only one repo lives in that
repo's .agents/loadout.md (template) or in a
project skill.
## Skills
Descriptions are triggers, not summaries. Say the words, chain them in one
sentence: "take over #123, sol loop, full stop".
| Skill | Say | What it does |
| full-send | "full send" | Takes a change to a merged PR with nobody watching: proves it, reviews it, files it, babysits it, merges it. |
| full-stop | "full stop" | Same as full-send, but stops when the PR is ready for you to merge. |
| sol-loop | "sol loop" / "soul loop" | GPT Sol reviews the change in a loop until both agree. 3 rounds, 5 for security or data loss. |
| council | "council" | Claude Opus, GPT Sol and Gemini Pro review; loops until they agree; scores go in the PR. |
| prove | "prove it", "verify" | Evidence the change works: reproduce, fail-on-base, the running app, screenshots. |
| precision-review | "precision review" | The review contract every reviewer works to: evidence, severity and effort, scope. |
| file-pr | "file a PR" | Short PR that leads with the problem. Repo preflight from .agents/loadout.md. |
| babysit-pr | "babysit" | Watches a PR through CI and review bots and fixes the real findings. |
| takeover | "take over" | Finishes someone else's PR, branch or agent thread and keeps their credit. |
| modernize | "modernize" | Rebases an old PR and decides whether it should still exist. |
| cut-release | "cut a release" | Version, gates that mirror prod, tag, notes, deploy, health checks. |
| retro | "retro" | Why did the agent do that? A better next prompt, and a patch to the skill. |
| harness-guards | (automatic) | What to do when the sandbox, a worktree guard or dcg rejects a command. |
| agent-report | (automatic) | Coverage block for investigation subagents. |
| leaving-pr-comments | (automatic) | Format for comments an agent posts on my behalf. |
Subagent: ci-digest (Haiku) reads one failed CI log and returns only the
failures.
Plugin loadout (Claude Code): a secret-file guard, per-project direnv
loading in worktrees, and a headless Playwright MCP server with a Nix
Chromium.
global/AGENTS.md holds the rules every agent on every
host follows, including the model roster. Claude imports it from
~/.claude/CLAUDE.md; Codex and Antigravity get it as ~/.codex/AGENTS.md
and ~/.gemini/GEMINI.md.
## How it reaches every host
edit skill ──► loadout push ──► GitHub ──► every host: loadout-sync.service
│ (timer every 15 min, and at login)
└──ssh──► other hosts: systemctl --user start loadout-sync
Each host keeps a live checkout (loadout path). loadout sync clones it
if missing, fast-forwards it (never over local edits or a diverged branch),
and links each skill into ~/.claude/skills and ~/.agents/skills. Links it
did not create are never touched, so claude.ai synced skills and other tools'