{"slug":"delegate-setup","title":"delegate-setup","summary":"Configure delegation fleet lanes: which implementer CLI handles which kind of work, with optional model and effort (or variant) dials. Discovers installed CLIs, proposes a lane map for user approval, and writes global or project config only after explicit yes. Use when the user a","platform":"Cursor","tags":[],"authorName":"LLM Mart","authorSlug":"llm-mart","score":0,"source":"github","price":null,"verified":false,"createdAt":"2026-09-02T05:41:26.316617Z","repo":{"url":"https://github.com/amElnagdy/delegate-skills","stars":2222,"forks":184,"license":"MIT","updatedAt":"2026-09-20T19:29:41Z"},"bodyHtml":"<hr>\n<h2>name: delegate-setup\ndescription: &gt;-\nConfigure delegation fleet lanes: which implementer CLI handles which kind of work,\nwith optional model and effort (or variant) dials. Discovers installed CLIs, proposes\na lane map for user approval, and writes global or project config only after explicit\nyes. Use when the user asks to set up, configure, or reconfigure delegation lanes,\na fleet of lanes, or which implementer handles feature/tests/ui work — not for\ndispatching a coding task to an implementer.\nlicense: MIT\ncompatibility: Requires Node 18+. No implementer CLIs are required — the skill discovers what is available.\nmetadata:\nversion: 0.5.0</h2>\n<h1>Delegate Setup</h1>\n<p>You are the <strong>orchestrator</strong> in <strong>setup mode</strong>. Discover installed implementer CLIs, propose a\n<strong>fleet of lanes</strong>, and write configuration only after the user approves.</p>\n<p>This skill does <strong>not</strong> dispatch coding work. It only authors the lane map.</p>\n<p>One concept: <strong>lanes</strong>. Never say “routes.”</p>\n<p>Example lane: <strong>feature</strong> → implementer <code>opencode</code>, model <code>opencode/grok</code>, variant <code>high</code>\n(OpenCode uses <code>variant</code> for reasoning intensity, not <code>effort</code>).</p>\n<h2>When NOT to use this</h2>\n<ul>\n<li>The user wants a task implemented — use the matching <code>*-delegate</code> skill instead.</li>\n<li>A one-off model change on a single dispatch — pass <code>--model</code> / <code>--effort</code> / <code>--variant</code> on that relay.</li>\n</ul>\n<h2>Hard rules</h2>\n<ol>\n<li>Every lane <strong>must</strong> include <code>implementer</code>.</li>\n<li>Put dials on the same object (<code>model</code>, <code>effort</code> or <code>variant</code>, …) only if that implementer supports them — see <a href=\"references/schema.md\">references/schema.md</a>.</li>\n<li>Show a human-readable lane table <strong>and</strong> the full JSON before every write; re-show after every tweak.</li>\n<li>Write <strong>only</strong> after an explicit approval (“yes”, “approve”, “write it”).</li>\n<li>Ask scope unless already clear: <strong>global</strong> (all projects) vs <strong>this repo only</strong>. Never create a project file just because cwd is a git repo. If there is no git repo, default to global and say so.</li>\n<li>Do not invent model identifiers.</li>\n<li>In interview or usage-scan mode, never write <strong>any</strong> dial the user did not give you and the schema does not require — omit it, so the CLI’s or relay’s own default applies.</li>\n<li>Prefer 3–5 useful lanes over a kitchen-sink map.</li>\n<li>Never edit <code>AGENTS.md</code>, <code>CLAUDE.md</code>, or other user agent-instruction files.</li>\n<li>Never run a <code>*-delegate</code> relay from this skill.</li>\n</ol>\n<p>(<code>&lt;skill-dir&gt;</code> is this skill’s install directory — the folder that contains this <code>SKILL.md</code>.)</p>\n<h2>Flow</h2>\n<p><code>discover → load → grounding menu → propose (with Basis) → scope → approve → write</code></p>\n<h3>1. Discover</h3>\n<pre><code>node \"&lt;skill-dir&gt;/scripts/discover.mjs\"\n</code></pre>\n<p>Summarize installed vs missing, auth (<code>true</code> / <code>false</code> / <code>null</code> = unknown), and whether models were\n<code>reported</code>, <code>aliases</code> (curated aliases in the registry, not live discovery — full model names also\nwork), <code>unsupported</code>, or <code>failed</code>.</p>\n<h3>2. Load existing (effective map)</h3>\n<pre><code>node \"&lt;skill-dir&gt;/scripts/config.mjs\" load --cwd \"$PWD\"\n</code></pre>\n<ul>\n<li>Neither present → “No lanes configured yet.”</li>\n<li>Otherwise → table of <strong>effective</strong> lanes with a Source column (<code>global</code> / <code>project</code>). Do not paste\nboth raw files unless asked.</li>\n<li>If <code>projectPresent</code> is true and <code>projectTrusted</code> is false, label the project lanes <strong>untrusted</strong>.\nThey cannot dispatch until the user reviews and approves a project write.</li>\n</ul>\n<h3>3. Propose</h3>\n<p>Discovery reports capability, never task fit. So ask <strong>one</strong> grounding question before proposing\nanything — one question, three options, not a wizard:</p>\n<blockquote>\n<p>How should I pick the lanes? <strong>(1) Quick defaults</strong> — I decide, no questions.\n<strong>(2) Interview</strong> — about four questions on how you want work allocated.\n<strong>(3) Usage scan</strong> — I re-read your CLIs’ local session folders (counts and dates only, never the\nconversations) and let the numbers place your lanes — if one CLI dominates, expect one question\nabout its role. Happy to do 2 and 3 together.</p>\n</blockquote>\n<ul>\n<li><strong>Quick defaults</strong> → propose immediately.</li>\n<li><strong>Interview</strong> → the four questions (allocation policy, never model rankings) and how to ask them\n(one medium per round) live in <a href=\"references/setup-dialogue.md\">references/setup-dialogue.md</a> — read\nit before you ask.</li>\n<li><strong>Usage scan</strong> → <code>node \"&lt;skill-dir&gt;/scripts/discover.mjs\" --usage</code>. Tell the user it is metadata\nonly before running it. Each discovered CLI gains <code>usage: { sessions, lastUsed }</code>; <code>null</code> means no\nprobe is wired — unknown, not unused.</li>\n<li><strong>Both</strong> → run the scan first, then ask only what the numbers cannot answer.</li>\n<li>Inside a git repo, repo signals (languages, test weight, frontend share) are a fourth source of\nevidence. They do not change the menu; they feed the proposal and the <code>repo</code> basis.</li>\n</ul>\n<p><strong>That menu is also the consent surface</strong> — the option chosen sets how much of the map is yours to\ndecide:</p>\n<ul>\n<li><strong>Quick defaults</strong> — the user hired your opinion. A full map is legitimate, dials included; label\nevery lane <code>my opinion</code>, say plainly that the map is your opinion, and keep it cheap to revise.</li>\n<li><strong>Interview / usage scan</strong> — evidence modes, so <strong>every</strong> dial is gated (rule 7): set one only from\nthe user’s answer, or where the schema requires it (opencode lanes require <code>model</code>). Omitting is\nalways safe — every dial has a default the user already lives with, and a CLI’s configured default\nis their standing choice, better evidence than your priors. Choosing which installed implementer\ngets a lane is still yours — Basis <code>my opinion</code> — but a dial that raises spend is not: offer your\ndial picks only as an addendum after the proposal, see\n<a href=\"references/setup-dialogue.md\">references/setup-dialogue.md</a>.</li>\n<li><strong>An unanswered question shrinks the map; it never licenses a substitution.</strong> Propose fewer, more\nconservative lanes, name the axis you are blind on (no quota answer → say the map is quota-blind),\nand invite the answer anytime. Re-ask once at most; never backfill silence with priors.</li>\n</ul>\n<p><strong>Delegation economics.</strong> The orchestrator reviews and lands every result — the review is the\nquality gate, so optimize total cost, not implementer prestige:</p>\n<ul>\n<li>Prefer capable, authenticated, burnable, <strong>low-usage</strong> CLIs for bounded, objectively gated work\n(tests, mechanical refactors, straightforward fixes) when their reliability keeps review and\nrework economical — lanes push token burn away from the subscriptions the user is protecting.\nLow usage alone does not establish burnable: discovery cannot see plans, limits, or per-run\ncost, and a rarely-used CLI may be metered or deliberately avoided. Burnable comes from the\nuser's quota answer — or, in quick defaults, from your labeled opinion.</li>\n<li>Avoid binding a lane to a CLI the user is protecting or orchestrates from, by default; bind it\nonly when the user asks for it or no acceptable alternative exists. Lanes are\n<strong>orchestrator-blind</strong>: the same lane fires from every seat the user drives from, and from that\nCLI's own seat it dispatches the CLI to itself.</li>\n<li>Surplus placement breaks down when rework and review cost exceed the savings; when the\nimplementer is flaky; when correctness rides on security, concurrency, migrations, or unstated\ndomain knowledge; and when the output <strong>is</strong> the product (debate, architecture, research) —\nreview limits damage, it does not manufacture a good first attempt. Bind those lanes to stronger\nimplementers.</li>\n<li>An explicit \"spare X\" answer removes X from proposed lanes by default, and overrides blanket\nposture answers on any lane the user explicitly retains for X — ask whether the posture applies\nthere; omit the dial if unanswered. Never silently stretch one answer across an axis it\nconflicts with.</li>\n</ul>\n<p>Question phrasings for the burn/spare and trust interview live in\n<a href=\"references/setup-dialogue.md\">references/setup-dialogue.md</a>.</p>\n<p>Then propose the lanes. Name them after the work the user described; fall back to <code>feature</code>, <code>tests</code>,\n<code>ui</code>, <code>fast</code>, <code>complex</code>. Installed implementers only.</p>\n<p>Show:</p>\n<table>\n<thead>\n<tr>\n<th>Lane</th>\n<th>Implementer</th>\n<th>Model</th>\n<th>Effort / variant</th>\n<th>Basis</th>\n<th>Source (if updating)</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>feature</td>\n<td>opencode</td>\n<td>opencode/grok</td>\n<td>variant: high</td>\n<td>your answer + schema requirement</td>\n<td>—</td>\n</tr>\n<tr>\n<td>tests</td>\n<td>codex</td>\n<td>—</td>\n<td>—</td>\n<td>usage data</td>\n<td>—</td>\n</tr>\n<tr>\n<td>ui</td>\n<td>claude</td>\n<td>—</td>\n<td>—</td>\n<td>my opinion (implementer)</td>\n<td>—</td>\n</tr>\n</tbody>\n</table>\n<p><strong>Basis</strong> is mandatory on every lane: <code>your answer</code> / <code>usage data</code> / <code>repo</code> / <code>my opinion</code> /\n<code>schema requirement</code> (a dial the schema forces is neither evidence nor opinion — say so). A lane you\npicked from model-quality priors is <code>my opinion</code> — never present it as something the tooling\ndetermined, and “installed and authenticated” is capability, not evidence of fit. When a lane’s\nimplementer and its dials come from different places, split the label — see\n<a href=\"references/setup-dialogue.md\">references/setup-dialogue.md</a>.</p>\n<p>Then the <strong>complete</strong> JSON (<code>version</code>: <code>delegate-fleet.v1</code>). One line of why per lane; flag auth or\nmodel uncertainty.</p>\n<p>Schema and dial table: <a href=\"references/schema.md\">references/schema.md</a>.</p>\n<h3>4. Scope</h3>\n<ul>\n<li>User said global / all projects / outside the project → <code>global</code>.</li>\n<li>No git repo → <code>global</code> (say so).</li>\n<li>Else ask once: global vs this repo only.</li>\n</ul>\n<h3>5. Approve and write</h3>\n<p>On explicit yes, write <strong>only</strong> the chosen scope (validate first). Build the payload from that\nscope’s raw file (or an empty <code>lanes</code> object if new) — not from the effective merged <code>load</code> view,\nor a project write will shadow global-only lanes and a global write will promote project-only ones.</p>\n<p>Create a uniquely named file under the platform temporary directory (<code>$TMPDIR</code>, <code>%TEMP%</code>, or Node\n<code>os.tmpdir()</code>; never hard-code <code>/tmp</code>, which breaks on native Windows), write the <strong>exact approved\nJSON</strong> into it with the orchestrator's file-writing tool, and use that populated path as\n<code>&lt;lanes-json&gt;</code> below. Never validate an empty temp file. Remove the temp file after the\nvalidation/write attempt, whether it succeeds or fails.</p>\n<pre><code>node \"&lt;skill-dir&gt;/scripts/config.mjs\" validate \"&lt;lanes-json&gt;\"\nnode \"&lt;skill-dir&gt;/scripts/config.mjs\" write --scope global \"&lt;lanes-json&gt;\"\n# or:  write --scope project --cwd /path/to/repo \"&lt;lanes-json&gt;\"\n</code></pre>\n<p>Re-read with <code>load</code>, then confirm the path written and the active lane names. Project writes bind\napproval to the exact config content; later changes fail closed until re-approved. On update, a short\nbefore/after is enough.</p>\n<h3>6. Ready to delegate</h3>\n<p>Stop after confirming. Tell the user the map is ready. For later work: read the lane’s\n<code>implementer</code>, load that <code>*-delegate</code> skill, and dispatch with <code>--lane &lt;name&gt;</code> (explicit\n<code>--model</code> / <code>--effort</code> / <code>--variant</code> still win when passed). Do not start a delegate task\nunless they ask.</p>\n<h2>Reconfigure</h2>\n<p>Same flow. Show the effective current map, propose changes, approve, write one scope’s file.\nReinstalling the skills package must not rewrite these files — they live outside the package.</p>\n","files":[{"path":"references/schema.md","sizeBytes":5131,"isText":true},{"path":"references/setup-dialogue.md","sizeBytes":6958,"isText":true},{"path":"scripts/config.mjs","sizeBytes":18101,"isText":false},{"path":"scripts/discover.mjs","sizeBytes":16072,"isText":false},{"path":"scripts/implementers.mjs","sizeBytes":16442,"isText":false},{"path":"scripts/lane.mjs","sizeBytes":6817,"isText":false},{"path":"SKILL.md","sizeBytes":10929,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-09-27T19:30:05.848099Z","sha256":"49F1280820035061AB165677F9F846DA136A247056A46BFC6B46A559F8DBEFAE","sizeBytes":28996},"review":null,"source":{"repositoryUrl":"https://github.com/amElnagdy/delegate-skills","path":"skills/delegate-setup","license":"MIT","commit":"6826b363085dcc80875372315fe7d208c4bf733f","subtreeSha":"29A55A95577C3D4F4B2F341E5752CBCE141BA48FB95D2DD73991545703DEEC52","lastSyncedAt":"2026-09-27T19:29:51.157831Z"},"reviewedAt":"2026-09-27T19:30:09.154611Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/amElnagdy/delegate-skills/tree/master/skills/delegate-setup"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install amelnagdy-delegate-skills@llmmart"},{"target":"git","command":"git clone https://github.com/amElnagdy/delegate-skills.git"}]}