[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77MMGEZ1F59K59WH3978KTS":3,"skill-same:01M77MMGEZ1F59K59WH3978KTS":69},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":29,"commitSha":30,"skillDirectory":31,"forksCount":32,"repositoryHtmlUrl":33,"repositoryHomepageUrl":6,"ownerHtmlUrl":34,"ownerBlogUrl":35,"ownerDescription":36,"updatedAt":37,"install":38,"resources":50},"01M77MMGEZ1F59K59WH3978KTS","orchestrate",null,"Arm the session for orchestration-heavy work with seven standing imperatives (delegate and fan out, spec every spawn, fresh-context verify, and more), optionally exported as a worker brief. Use when: 'orchestrate', 'prime this session', or before heavy delegation.","melodic-software\u002Fclaude-code-plugins","melodic-software","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F58273638?v=4","01M77JVQTXNGKCE4MDFY2BC936","Melodic Software","01M77JVQVN9QDWKAH6W03ANP76",30,"MIT","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fsession-flow\u002Fskills\u002Forchestrate\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\ndescription: \"Arm the session for orchestration-heavy work with seven standing imperatives (delegate and fan out, spec every spawn, fresh-context verify, and more), optionally exported as a worker brief. Use when: 'orchestrate', 'prime this session', or before heavy delegation.\"\nargument-hint: \"[\u003Ctask>|handoff [compact]|worker [compact]|coordinator [\u003Carea>]]\"\nuser-invocable: true\ndisable-model-invocation: false\nmetadata:\n  workflow-stage: session\n  summary: Arm the session with proactive-orchestration imperatives\n---\n\n## Purpose\n\nInvoking this skill **arms the current session** for an orchestration-heavy task: it loads the\nexpanded seven-imperative operational form into active working context, and it declares deliberate\nintent to orchestrate the work about to start, so the triggers get evaluated actively rather than\nsitting passively in background rules. That is the default: no paste, no rails, just preloaded\ncontext.\n\nThe same imperatives also **export** as a self-contained, paste-ready brief for targets that LEAVE\nthe session and therefore inherit none of its context: a spawned subagent\u002Fteammate, a fresh session\nyou will `\u002Fclear` into, or a non-Claude-Code tool. Export is model- and tool-agnostic by\nconstruction. Nothing in the pasted text depends on a specific model, env var, or repo file.\n\nPointers behind each imperative: `context\u002Fsources.md`; observed failure modes:\n`context\u002Fgotchas.md`. Read gotchas before authoring a nested tree or trusting a worker's return.\n\n## Actions\n\n| Action | What it does |\n|---|---|\n| *(default. Optional `\u003Ctask>`)* | **Prime THIS session.** The standing instructions below are now active for the upcoming task; respond with a terse acknowledgment and, if `\u003Ctask>` is given, one line orienting to it. Do NOT re-emit the imperatives. Loading them IS the priming. |\n| `handoff [compact]` | **Export** the imperatives as a paste-ready dashed-rail brief framed for a fresh session. `compact` = headlines only. |\n| `worker [compact]` | **Export** framed for a spawned worker (prepends the did-not-inherit-context line). `compact` = headlines only. |\n| `coordinator [\u003Carea>]` | **Arm THIS session as the one coordinator for `\u003Carea>`** to delegate that area's work to workers. The coordinator keeps no store of its own: state lives in the tracker and the repository's durable files, so it creates no store file. It runs on whatever surface the user chooses, a plain CLI session included, so account rotation never strands it. Load `context\u002Fcoordinator.md` and follow its four rules for the rest of the session. Respond with a terse acknowledgment that names the area and where state lives. |\n\n## Orchestration imperatives. Standing instructions\n\nAt each decision boundary in this task, evaluate these and ACT on a match without waiting to be\ntold:\n\n1. DELEGATE \u002F FAN OUT. Start with one agent; delegate only when work would flood context, fans\n   across genuinely independent paths, or needs a tool-restricted specialist. Decompose by what\n   CONTEXT each piece needs, not by head-count or work-type. Sequential or shared-context steps stay\n   in one agent, and one feature is never split across agents. Our operating figure for a fan-out\n   is 3–10× one agent's tokens (returns cost context too), and it is a floor: budget a\n   research-shaped fan-out above it. Spend it on value + parallelism, not convenience. Pointer: for\n   multi-agent token cost, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fcosts#agent-team-token-costs>; no\n   docs page covers research fan-out sizing as of 2026-10-01 (correlate with\n   \u003Chttps:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fmulti-agent-research-system>). As of: 2026-10-01.\n   Recheck trigger: that section moves or starts stating its own multiplier, or a docs page starts\n   covering fan-out sizing.\n   \"Would flood context\" is a measurement, not a hunch, when the instrument exists: with the\n   `context-usage` plugin enabled, resolve this session's zone word per its reader contract\n   before a fan-out decision\n   (the contract owns the snapshot path, staleness rule, and bands. Read them there; this\n   imperative consumes only the word, no band values). Never estimate your own remaining window,\n   that guess is the failure the seam replaces. A degraded or `unknown` zone shifts the balance\n   toward delegating context-heavy legs and shrinking what returns; a healthy zone is license to\n   keep sequential, shared-context work inline.\n2. SPEC EVERY SPAWN. Every brief states why the work is wanted, what done looks like, when to\n   stop and ask, and a deliberately chosen model tier, on top of the brief elements the\n   multi-agent post lists. Pointer: [Give the reason, not only the request](https:\u002F\u002Fplatform.claude.com\u002Fdocs\u002Fen\u002Fbuild-with-claude\u002Fprompt-engineering\u002Fprompting-claude-fable-5#give-the-reason-not-only-the-request);\n   correlate with \u003Chttps:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fmulti-agent-research-system> for the\n   brief elements, which no docs page covers. Name the model tier deliberately rather than leaving\n   it unspecified: what an omitted `model` resolves to, with or without a consumer-level\n   subagent-model override, is upstream behavior\n   (pointer: \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#choose-a-model>).\n   Holding only an objective, a worker resolves each ambiguity toward the sentence you wrote rather\n   than the outcome you wanted, and returns something well-formed and wrong.\n   A worker that measures is told which exact commits to compare and how to sample (how many\n   runs, in what order), and its return repeats both beside the numbers; a measurement whose\n   return leaves either out is not evidence. A brief for a worker that writes may state how long\n   the work should take, which imperative 4 uses.\n3. FRESH-CONTEXT VERIFY, after an edit batch or a finding set, hand it to a SEPARATE verifier;\n   never self-audit in the context that produced it. Give the verifier concrete pass\u002Ffail criteria\n   (\"run the full suite, report all failures\"), scope it to correctness\u002Frequirements (not style),\n   and judge the final STATE, not the process, an uncriteriaed verifier just rubber-stamps. When\n   the verdict is high-stakes, prefer a different-vendor advisor when one is set up and able to\n   judge this artifact, its blind spots are uncorrelated with yours, with the fresh-context\n   same-vendor verifier as the fallback. A `light` or `off` verification level, where the session resolved\n   one, skips this verifier. Scope it to what ships: a process record about the work\n   (ledger, checklist, status log) is not the work and gets no verifier, however many of them a\n   batch touched, and a record OF a verification is never itself verified, that loop feeds itself.\n4. RUN WORKERS WELL. Dispatch without blocking, and keep a worker across subtasks where the\n   runtime allows; watch each worker and step in when it drifts. Pointer:\n   [Parallel subagents](https:\u002F\u002Fplatform.claude.com\u002Fdocs\u002Fen\u002Fbuild-with-claude\u002Fprompt-engineering\u002Fprompting-claude-fable-5#parallel-subagents).\n   A worker that must wait on an external result polls it in the foreground with a bounded loop, or returns what\n   it has and lets the parent re-dispatch. A background command or watch the worker started is not\n   a wait: the runtime may stop it when the worker returns, a watch expires at its deadline, and no\n   sentinel file appears unless something the worker launched writes it. The runtime may also leave\n   it running: a hung background shell has been observed to keep a worker that already reported\n   listed as active (`context\u002Fsources.md`, \"A worker's own background work is not a wait\"). So\n   retiring a finished worker includes checking for its still-running background tasks and\n   surfacing each one; stopping one is gated like any kill (`\u002Fsession-flow:reconcile` step 3).\n   Liveness covers workers that write (edit files, commit, or run a process for the run), never\n   read-only ones such as a verifier. Read a writing worker's progress from its side effects\n   (new commits, written files, its task status), never by sending it a message to ask. Past the\n   runtime its brief states with no side effect, it is stuck: stop it (gated like any kill) and\n   respawn it with consolidated scope. With no stated runtime, confirm at each decision boundary\n   that its task is still running and still producing output. Every resume message restates the\n   brief's scope fence and standing constraints; after an interrupt, respawn the worker instead\n   of resuming it. Landing: \"land\" means a local commit on the run's own branch, never a push or\n   a merge. Land each unit as soon as it passes verification instead of holding a batch, and\n   apply no cutoff based on remaining budget, since imperative 1 forbids estimating the window.\n   Where a configured batched drain decides when returns are read, a unit lands once its return\n   has been read and verified. A worker runs where its parent session runs; cloud placement is\n   chosen per work item by the execution-target contract\n   (`docs\u002Fcontracts\u002Fexecution-target\u002FREADME.md` in the marketplace repository).\n5. NESTED SUBAGENTS, a worker may spawn its own workers when a delegated task itself subdivides\n   AND the depth is non-load-bearing. Reliability\n   degrades with depth and platforms cap it (for the cap, see\n   \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#let-subagents-spawn-their-own-subagents>), so never author a tree that needs a specific or deep\n   nesting level.\n6. SURFACE DRIFT, the moment you notice a stale reference, broken citation, or convention\n   conflict adjacent to your task, flag it in one line; don't fix it silently, don't deep-dive.\n7. CALIBRATE TO CONDITIONS. Size the whole orchestration (whether to delegate at all, fan-out\n   width, nesting depth) to the conditions in play, never a fixed recipe: the active model's\n   capability (a stronger model reaches further single-agent; a weaker one needs more decomposition\n   and tighter specs), whether a capable advisor\u002Fverifier is on hand, current context pressure\n   (delegate to protect a filling window; stay inline when it is roomy). Usage or rate-limit headroom\n   never lowers model tier, effort, review depth or worker count; pacing is the user's call. Back\n   off only on this session's own rate-limit errors or on live sibling-automation 429s already\n   visible to the session (for example review-lane infra comments classifying\n   `api_error_status: 429`); scale back up only after those reactive signals stop, never on a\n   guessed recovery. Sizing is\n   small\u002Fmedium\u002Flarge, a small ask stays single-agent, a medium one fans out a few, only a large\n   genuinely-independent surface earns a wide or nested tree. Single-agent is the floor, not the\n   fallback. Per-worker tier is part of sizing and follows the work: mechanical work runs on the\n   cheaper tier, which a wide mechanical fleet inherits as its DEFAULT because volume multiplies\n   every notch of over-provisioning, and the standing exception is an explicitly hard stage\n   (verify, judge\u002Fadjudicate, judgment-heavy synthesis), which keeps the parent tier. Tier is not only the\n   model: match the reasoning depth (effort) to the subtask too, not the parent session,\n   high-volume mechanical work (search, extraction, per-item transforms, formatting) runs cheaper\n   on both. Work that changes code, verifies a change, or is likely to hit edge cases is excluded\n   from the lower effort: pick its level from model-config's effort table\n   (\u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmodel-config#choose-an-effort-level>), and, as this\n   repository's own floor, never below medium. A premium fan-out outside the hard stages is a per-stage decision to justify\n   explicitly, never a default to inherit. Pointer for both cost judgments (delegating bulk work,\n   tuning effort per subtask):\n   \u003Chttps:\u002F\u002Fplatform.claude.com\u002Fdocs\u002Fen\u002Fabout-claude\u002Fmodels\u002Foptimizing-for-cost-and-intelligence#orchestrator-strategy-delegate-bulk-work>\n   and \u003Chttps:\u002F\u002Fplatform.claude.com\u002Fdocs\u002Fen\u002Fabout-claude\u002Fmodels\u002Foptimizing-for-cost-and-intelligence#tune-effort>.\n   Both judgments are ours.\n\nDiscipline: trigger-evaluation is mandatory; the ACTION stays calibrated (delegate on value +\nparallelism, not convenience). Treat every worker's return as unverified synthesis: check its\nevidence before accepting it, verify load-bearing claims against a primary source before acting,\nand merge a many-item fan-out into one table (item, verdict, evidence). An item whose return\nlacks a field its brief requires enters that table as a gap, never as a pass, unless a single\nconsolidated-scope respawn returns the field. Cite sources you actually fetched;\nnever label a claim \"known\" \u002F \"from memory\" \u002F \"obvious\".\n\n**Priming addendum (current session only).** As the main session, not a spawned non-fork worker,\nyou may also reach orchestration surfaces a non-fork worker cannot: agent teams (driven from the\nlead session; the docs do not state whether a fork of the lead can drive one; see\n[agent teams: architecture](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fagent-teams#architecture) and\n[limitations](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fagent-teams#limitations)) and dynamic workflows (withheld from non-fork workers). This session's reasoning effort is `${CLAUDE_EFFORT}`; if that value reads as a literal\nplaceholder, resolve the session's effort yourself before using it. Feed the value\ninto imperative 7's tier calibration: we treat it as the effort an unpinned spawn may run at when it\npasses no `effort`, and the pointer below says on which model that holds. Its gap from what a subtask needs IS\nthe over-provisioning imperative 7 exists to stop. Close that gap by passing the level on a\nnon-fork spawn of an unpinned agent, and omit it on a pinned named agent. Never read ultracode from this value. Pointer: for\nthe subagent `effort` field, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#supported-frontmatter-fields>;\nfor the per-spawn `effort` and its precedence, see\n\u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#choose-an-effort-level>; for which model the session's\nlevel reaches, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmodel-config#effort-level-of-subagents-and-teammates>;\nfor how ultracode relates to\neffort, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmodel-config#adjust-effort-level>. Where a\n`SendMessage` tool resolves in this session, run imperative 4's worker reuse and mid-flight\nintervention through it, addressed by the worker's agent ID. Never re-invoke the dispatch tool to\ncontinue a worker: that starts a second, independent worker. Read a refused message as a worker\nthe user stopped. Pointers and the empirical probe: `context\u002Fsources.md`, \"SendMessage\nworker continuation\".\n`worker_continuation` decides how a worker's next unit (a fix round, a follow-up, a retry, the\nnext queue item) runs. Resolve it once, at arm time, lowest layer first: the default `resume`; the\nuser's option `${user_config.worker_continuation}`, where a literal, unexpanded placeholder means\nunset; then the repository's `docs\u002Fconventions\u002Fsession-flow.yaml`, which wins when it sets the key\n(schema: [`${CLAUDE_PLUGIN_ROOT}\u002Fschemas\u002Fsession-flow.schema.json`](${CLAUDE_PLUGIN_ROOT}\u002Fschemas\u002Fsession-flow.schema.json)).\nRead the repository value with\n`bash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fretro\u002Fscripts\u002Fparse-concern-value.sh\" --strict \"\u003Cgit root>\u002Fdocs\u002Fconventions\u002Fsession-flow.yaml\" worker_continuation`,\nand only when the working directory's git root is neither `$HOME` nor an ancestor of it; otherwise\nskip that layer and say so. Empty output means unset only when the key is absent: if\n`grep -qE '^worker_continuation[[:space:]]*:' \"\u003Cfile>\"` matches, the key is present but empty or not\na scalar, which is invalid. Exit 3 means the parser rejected the file: name the file and drop the\nlayer. A value other than `resume` or `respawn` is named with its file or option, the key and the\nvalue, and that layer is dropped: the repository's valid value still wins over an invalid user\nvalue, and an invalid repository value or file resolves the default `resume`, never the user's\nvalue; the run never stops on it. Report one line, for\nexample `worker_continuation: respawn (docs\u002Fconventions\u002Fsession-flow.yaml)`. Under `resume`,\nimperative 4's reuse stands. Under `respawn`, each new unit goes to a fresh worker whose brief\nconsolidates the scope (the original brief, every later directive, the prior worker's report and\nits branch); resume instead only when the unit needs state that lives in that worker and is\ncostly to move: its checkout, uncommitted changes, or a running process. Settings page:\n[`${CLAUDE_PLUGIN_ROOT}\u002Freference\u002Fconfig.md`](${CLAUDE_PLUGIN_ROOT}\u002Freference\u002Fconfig.md).\nTo choose between a workflow and subagents, and the model and effort each spawn gets, run\n`\u002Fmulti-agent:assess` and `\u002Fmulti-agent:route` when they resolve in this session; otherwise read\n\u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#choose-a-model> (record: `context\u002Fsources.md`,\n\"Priming addendum: model and effort routing\").\nResolve the repository's verification level once with\n`bash \"${CLAUDE_PLUGIN_ROOT}\u002Flib\u002Fread-verification.sh\" --plugin session-flow --skill orchestrate --file docs\u002Fconventions\u002Fsession-flow.yaml`\nand report its line ([levels](https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fdocs\u002Fcontracts\u002Fverification-level\u002FREADME.md)).\nIt governs imperative 3 in this session: under `light`, and `off`, which acts as `light` here,\nhand nothing to a verifier and never self-audit in its place; write `verification: skipped\n(\u003Clevel>)` where the verdict would go. An exported brief carries no level: its target resolves\nits own.\nExport modes omit this addendum, a\npasted target reaches none of those surfaces, and the substitution would travel as dead text.\n\n## Tiered delegation, the shape of a deep tree\n\nGuidance for the main session once a task needs more than one layer; the export brief omits it,\nbecause a pasted worker sits inside a tree rather than authoring one.\n\n**A rough anchor for small\u002Fmedium\u002Flarge.** Not thresholds to enforce, the judgment still runs on context\nboundaries, not head-count, but our anchor is the platform's workflow size guideline values,\nread live. A run large enough to trip the platform's large-workflow warning is a size to justify\nout loud. An order-of-magnitude disagreement with this anchor is one to name, not skip. Before a\nworkflow run, read the size guideline in force for this session: the guideline the session sets\nis the one Claude receives, whatever this anchor says.\n\n- **Pointer**: for the size guideline values and their defaults, see\n  \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fworkflows#set-a-size-guideline>; for the large-workflow\n  warning, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fworkflows#cost>.\n\n**The top of the tree owns the loop, not the work.** Its context is the scarcest in the run,\neverything that enters it stays for the rest of the session. So it holds the objective, the\nstopping condition, and the decision about what to spawn next, and it delegates the rest. It keeps\nthe task list in a file (the plan or checklist the work already has, else one it creates), ticks\neach item as its return is verified, keeps an item it will not do listed as `skip: \u003Creason>`, and\nreads that file rather than the scrollback. It keeps going\nwhen a step needs no input, with status notes in the same message as the next action, and stops to\nask only when blocked on the user or before a destructive, hard-to-undo, or outward action. A top\ntier that reads findings, weighs them, and asks a follow-up question has converted a fan-out into a\nconversation, and the context it was protecting fills anyway.\n\n**Chatter belongs low.** Two workers resolving an ambiguity between themselves costs nothing at the\ntop. The same exchange routed through the parent costs the parent's window twice and permanently.\nPush coordination to the lowest tier that can resolve it, and let each tier return a compressed\nverdict rather than its reasoning.\n\n**Spec what crosses a boundary, not just what to do.** Every spawn already needs an objective and\nan output format (imperative 2). In a multi-tier tree the output format IS the context-economy\nlever: name the identifiers, the verdict, and where the bulky payload was parked, so the tier above\ncan act without re-reading the work. Our magnitude for \"compressed\": a worker that explored across tens\nof thousands of tokens still aims to return roughly 1,000 to 2,000. Treat it as the shape a return\nshould aim for, never a budget to spend up to.\n\n- **Pointer**: for a subagent returning a summary in place of its verbose output, see\n  \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#isolate-high-volume-operations>\n  (correlate with \u003Chttps:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Feffective-context-engineering-for-ai-agents>).\n  No docs page states a return size.\n\n**Workers are ephemeral, and the deeper the tier the shorter the life.** A worker that finishes and\nstays alive keeps costing the tier above, notifications, status, re-acknowledgement, for zero\nadditional output. Retire on completion. When the next grouping needs doing, spawn fresh rather than\nreusing a worker whose context now carries the last job. (The exception is imperative 4's long-lived\nworker across *related* subtasks, where cache reuse is the point; that is a deliberate trade, not\nthe default.)\n\n**Check one sampled brief per wave, beside the wave.** Pick one brief from each wave you dispatch\nand check it against imperative 2 while that wave runs. The check never gates the wave: what it\nfinds corrects the next wave's briefs, and a defect that would change a running worker's result\ngoes to that worker as a directive.\n\n**Treat a clean return as unverified, especially a suspiciously clean one.** An under-specified\nworker rarely stalls and asks; it substitutes the nearest plausible interpretation and reports\nsuccess. In a fan-out, most workers given a brief missing a resource will locate it and close the\ngap, and one will silently audit a different, similar artifact and return a confident,\nwell-formed, wrong-target result that nothing in its return distinguishes from the others. This\nis why imperative 3's fresh-context verify is not optional at depth, and why a return payload\nbenefits from naming its sources. Provenance is the field that makes a wrong-target answer\ndetectable from above.\n\n**Never author a tree that needs a specific depth.** The platform's nesting default is\nconfigurable and changes. Agent-tool subagents carry a depth cap and a concurrency cap (their\nsettings: the pointers below); workflow agents\nand agent-team teammates carry their own, and workflow concurrency has its own override, so \"read\nthe current values\" includes the workflows page whenever the run will use the Workflow tool. Read the current values rather than assuming\nthem, and design the tree so it degrades to a shallower one instead of failing. Never design a\ntree that needs a fork to spawn a fork: that is a shape constraint, not a tunable. Whether a\nbelow-limit fork can parent non-fork children is unconfirmed, so do not treat a fork as a\nforbidden intermediate tier either. The version history behind the caps lives in\n`context\u002Fsources.md`.\n\n- **Pointer**: for the depth cap, see\n  \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#let-subagents-spawn-their-own-subagents>; for the\n  concurrency cap, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#concurrent-subagent-limit>; for\n  workflow limits, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fworkflows#behavior-and-limits>; for forks,\n  see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#how-forks-differ-from-other-subagents>.\n\n**Confirm nesting from behavior, not from one page** (why: `context\u002Fgotchas.md`). Have a\nworker of the SAME definition you plan to use as the intermediate tier attempt a trivial nested\nspawn and report the outcome. The gate is definition-specific, so another agent type proves\nnothing, and holding `Agent` is necessary but not sufficient. Read a refusal: a depth rejection\nnames depth; a permission refusal (classified pre-launch) does not. Pointers: `context\u002Fsources.md`.\n\n## Export modes (handoff \u002F worker). Paste-ready brief\n\nOnly for a target that LEAVES the session. Emit the seven imperatives above between two full-width\n`─` (U+2500) dashed rails. Top rail, brief, bottom rail, nothing else between them; the\n`\u002Fclear`\u002Fpaste instruction or any commentary sits above the top rail or below the bottom rail,\nnever between (NOT a code fence, the user copies the text between the rails, not fence markers).\n\nLive shape: bare `─` rails, no fence. Shown inside a fence here for display only.\n\n```text\n──────────────────────────────────────────────────────────\nORCHESTRATION BRIEF — standing instructions for the whole task, regardless of which model or tool runs you.\n\nAt each decision boundary, evaluate these and ACT on a match without waiting to be told:\n[the seven numbered imperatives above, verbatim]\n\nDiscipline: [the Discipline line above, verbatim]\n──────────────────────────────────────────────────────────\n```\n\n- `handoff`, the opening line above already fits a fresh session; emit as-is.\n- `worker`, insert as the FIRST line between the rails: `You are a spawned worker and did NOT\n  inherit the parent session's context or the repo's conditional rules, these instructions are\n  your only copy.` Then insert this waiting line directly after it, before the `ORCHESTRATION BRIEF` line,\n  in full and `compact` exports alike: `Waiting: poll in the foreground with a bounded loop, or\n  return what you have; never end your turn to wait on your own background job.`\n- `compact`. Emit only the seven numbered HEADLINES (`1. DELEGATE \u002F FAN OUT`, `2. SPEC EVERY\n  SPAWN`, …) plus the closing Discipline line; drop every sub-clause. `worker compact` also keeps the\n  `worker` did-not-inherit and waiting lines; compact drops neither.\n\n## Next\n\n\u002Fplanning:draft-goal-condition turns the coordinated work into a completion condition.\n\n## What this skill does NOT do\n\n- **Default does not emit paste-text.** Priming the current session is a terse acknowledgment,\n  the work happens because the imperatives loaded into context, not because anything was printed.\n  Use `handoff` \u002F `worker` only when the target LEAVES the session.\n- **Not a surface-selection guide.** Which parallel-execution surface to pick (subagents vs nested\n  vs teams vs workflows) is a judgment the main session makes against current official docs; the\n  export brief deliberately omits agent teams + dynamic workflows because a pasted target cannot\n  reach either.\n- **Does not delegate, verify, nest, or spawn anything itself**. It arms the session or emits\n  instruction text.\n","## Purpose\n\nInvoking this skill **arms the current session** for an orchestration-heavy task: it loads the\nexpanded seven-imperative operational form into active working context, and it declares deliberate\nintent to orchestrate the work about to start, so the triggers get evaluated actively rather than\nsitting passively in background rules. That is the default: no paste, no rails, just preloaded\ncontext.\n\nThe same imperatives also **export** as a self-contained, paste-ready brief for targets that LEAVE\nthe session and therefore inherit none of its context: a spawned subagent\u002Fteammate, a fresh session\nyou will `\u002Fclear` into, or a non-Claude-Code tool. Export is model- and tool-agnostic by\nconstruction. Nothing in the pasted text depends on a specific model, env var, or repo file.\n\nPointers behind each imperative: `context\u002Fsources.md`; observed failure modes:\n`context\u002Fgotchas.md`. Read gotchas before authoring a nested tree or trusting a worker's return.\n\n## Actions\n\n| Action | What it does |\n|---|---|\n| *(default. Optional `\u003Ctask>`)* | **Prime THIS session.** The standing instructions below are now active for the upcoming task; respond with a terse acknowledgment and, if `\u003Ctask>` is given, one line orienting to it. Do NOT re-emit the imperatives. Loading them IS the priming. |\n| `handoff [compact]` | **Export** the imperatives as a paste-ready dashed-rail brief framed for a fresh session. `compact` = headlines only. |\n| `worker [compact]` | **Export** framed for a spawned worker (prepends the did-not-inherit-context line). `compact` = headlines only. |\n| `coordinator [\u003Carea>]` | **Arm THIS session as the one coordinator for `\u003Carea>`** to delegate that area's work to workers. The coordinator keeps no store of its own: state lives in the tracker and the repository's durable files, so it creates no store file. It runs on whatever surface the user chooses, a plain CLI session included, so account rotation never strands it. Load `context\u002Fcoordinator.md` and follow its four rules for the rest of the session. Respond with a terse acknowledgment that names the area and where state lives. |\n\n## Orchestration imperatives. Standing instructions\n\nAt each decision boundary in this task, evaluate these and ACT on a match without waiting to be\ntold:\n\n1. DELEGATE \u002F FAN OUT. Start with one agent; delegate only when work would flood context, fans\n   across genuinely independent paths, or needs a tool-restricted specialist. Decompose by what\n   CONTEXT each piece needs, not by head-count or work-type. Sequential or shared-context steps stay\n   in one agent, and one feature is never split across agents. Our operating figure for a fan-out\n   is 3–10× one agent's tokens (returns cost context too), and it is a floor: budget a\n   research-shaped fan-out above it. Spend it on value + parallelism, not convenience. Pointer: for\n   multi-agent token cost, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fcosts#agent-team-token-costs>; no\n   docs page covers research fan-out sizing as of 2026-10-01 (correlate with\n   \u003Chttps:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fmulti-agent-research-system>). As of: 2026-10-01.\n   Recheck trigger: that section moves or starts stating its own multiplier, or a docs page starts\n   covering fan-out sizing.\n   \"Would flood context\" is a measurement, not a hunch, when the instrument exists: with the\n   `context-usage` plugin enabled, resolve this session's zone word per its reader contract\n   before a fan-out decision\n   (the contract owns the snapshot path, staleness rule, and bands. Read them there; this\n   imperative consumes only the word, no band values). Never estimate your own remaining window,\n   that guess is the failure the seam replaces. A degraded or `unknown` zone shifts the balance\n   toward delegating context-heavy legs and shrinking what returns; a healthy zone is license to\n   keep sequential, shared-context work inline.\n2. SPEC EVERY SPAWN. Every brief states why the work is wanted, what done looks like, when to\n   stop and ask, and a deliberately chosen model tier, on top of the brief elements the\n   multi-agent post lists. Pointer: [Give the reason, not only the request](https:\u002F\u002Fplatform.claude.com\u002Fdocs\u002Fen\u002Fbuild-with-claude\u002Fprompt-engineering\u002Fprompting-claude-fable-5#give-the-reason-not-only-the-request);\n   correlate with \u003Chttps:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fmulti-agent-research-system> for the\n   brief elements, which no docs page covers. Name the model tier deliberately rather than leaving\n   it unspecified: what an omitted `model` resolves to, with or without a consumer-level\n   subagent-model override, is upstream behavior\n   (pointer: \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#choose-a-model>).\n   Holding only an objective, a worker resolves each ambiguity toward the sentence you wrote rather\n   than the outcome you wanted, and returns something well-formed and wrong.\n   A worker that measures is told which exact commits to compare and how to sample (how many\n   runs, in what order), and its return repeats both beside the numbers; a measurement whose\n   return leaves either out is not evidence. A brief for a worker that writes may state how long\n   the work should take, which imperative 4 uses.\n3. FRESH-CONTEXT VERIFY, after an edit batch or a finding set, hand it to a SEPARATE verifier;\n   never self-audit in the context that produced it. Give the verifier concrete pass\u002Ffail criteria\n   (\"run the full suite, report all failures\"), scope it to correctness\u002Frequirements (not style),\n   and judge the final STATE, not the process, an uncriteriaed verifier just rubber-stamps. When\n   the verdict is high-stakes, prefer a different-vendor advisor when one is set up and able to\n   judge this artifact, its blind spots are uncorrelated with yours, with the fresh-context\n   same-vendor verifier as the fallback. A `light` or `off` verification level, where the session resolved\n   one, skips this verifier. Scope it to what ships: a process record about the work\n   (ledger, checklist, status log) is not the work and gets no verifier, however many of them a\n   batch touched, and a record OF a verification is never itself verified, that loop feeds itself.\n4. RUN WORKERS WELL. Dispatch without blocking, and keep a worker across subtasks where the\n   runtime allows; watch each worker and step in when it drifts. Pointer:\n   [Parallel subagents](https:\u002F\u002Fplatform.claude.com\u002Fdocs\u002Fen\u002Fbuild-with-claude\u002Fprompt-engineering\u002Fprompting-claude-fable-5#parallel-subagents).\n   A worker that must wait on an external result polls it in the foreground with a bounded loop, or returns what\n   it has and lets the parent re-dispatch. A background command or watch the worker started is not\n   a wait: the runtime may stop it when the worker returns, a watch expires at its deadline, and no\n   sentinel file appears unless something the worker launched writes it. The runtime may also leave\n   it running: a hung background shell has been observed to keep a worker that already reported\n   listed as active (`context\u002Fsources.md`, \"A worker's own background work is not a wait\"). So\n   retiring a finished worker includes checking for its still-running background tasks and\n   surfacing each one; stopping one is gated like any kill (`\u002Fsession-flow:reconcile` step 3).\n   Liveness covers workers that write (edit files, commit, or run a process for the run), never\n   read-only ones such as a verifier. Read a writing worker's progress from its side effects\n   (new commits, written files, its task status), never by sending it a message to ask. Past the\n   runtime its brief states with no side effect, it is stuck: stop it (gated like any kill) and\n   respawn it with consolidated scope. With no stated runtime, confirm at each decision boundary\n   that its task is still running and still producing output. Every resume message restates the\n   brief's scope fence and standing constraints; after an interrupt, respawn the worker instead\n   of resuming it. Landing: \"land\" means a local commit on the run's own branch, never a push or\n   a merge. Land each unit as soon as it passes verification instead of holding a batch, and\n   apply no cutoff based on remaining budget, since imperative 1 forbids estimating the window.\n   Where a configured batched drain decides when returns are read, a unit lands once its return\n   has been read and verified. A worker runs where its parent session runs; cloud placement is\n   chosen per work item by the execution-target contract\n   (`docs\u002Fcontracts\u002Fexecution-target\u002FREADME.md` in the marketplace repository).\n5. NESTED SUBAGENTS, a worker may spawn its own workers when a delegated task itself subdivides\n   AND the depth is non-load-bearing. Reliability\n   degrades with depth and platforms cap it (for the cap, see\n   \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#let-subagents-spawn-their-own-subagents>), so never author a tree that needs a specific or deep\n   nesting level.\n6. SURFACE DRIFT, the moment you notice a stale reference, broken citation, or convention\n   conflict adjacent to your task, flag it in one line; don't fix it silently, don't deep-dive.\n7. CALIBRATE TO CONDITIONS. Size the whole orchestration (whether to delegate at all, fan-out\n   width, nesting depth) to the conditions in play, never a fixed recipe: the active model's\n   capability (a stronger model reaches further single-agent; a weaker one needs more decomposition\n   and tighter specs), whether a capable advisor\u002Fverifier is on hand, current context pressure\n   (delegate to protect a filling window; stay inline when it is roomy). Usage or rate-limit headroom\n   never lowers model tier, effort, review depth or worker count; pacing is the user's call. Back\n   off only on this session's own rate-limit errors or on live sibling-automation 429s already\n   visible to the session (for example review-lane infra comments classifying\n   `api_error_status: 429`); scale back up only after those reactive signals stop, never on a\n   guessed recovery. Sizing is\n   small\u002Fmedium\u002Flarge, a small ask stays single-agent, a medium one fans out a few, only a large\n   genuinely-independent surface earns a wide or nested tree. Single-agent is the floor, not the\n   fallback. Per-worker tier is part of sizing and follows the work: mechanical work runs on the\n   cheaper tier, which a wide mechanical fleet inherits as its DEFAULT because volume multiplies\n   every notch of over-provisioning, and the standing exception is an explicitly hard stage\n   (verify, judge\u002Fadjudicate, judgment-heavy synthesis), which keeps the parent tier. Tier is not only the\n   model: match the reasoning depth (effort) to the subtask too, not the parent session,\n   high-volume mechanical work (search, extraction, per-item transforms, formatting) runs cheaper\n   on both. Work that changes code, verifies a change, or is likely to hit edge cases is excluded\n   from the lower effort: pick its level from model-config's effort table\n   (\u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmodel-config#choose-an-effort-level>), and, as this\n   repository's own floor, never below medium. A premium fan-out outside the hard stages is a per-stage decision to justify\n   explicitly, never a default to inherit. Pointer for both cost judgments (delegating bulk work,\n   tuning effort per subtask):\n   \u003Chttps:\u002F\u002Fplatform.claude.com\u002Fdocs\u002Fen\u002Fabout-claude\u002Fmodels\u002Foptimizing-for-cost-and-intelligence#orchestrator-strategy-delegate-bulk-work>\n   and \u003Chttps:\u002F\u002Fplatform.claude.com\u002Fdocs\u002Fen\u002Fabout-claude\u002Fmodels\u002Foptimizing-for-cost-and-intelligence#tune-effort>.\n   Both judgments are ours.\n\nDiscipline: trigger-evaluation is mandatory; the ACTION stays calibrated (delegate on value +\nparallelism, not convenience). Treat every worker's return as unverified synthesis: check its\nevidence before accepting it, verify load-bearing claims against a primary source before acting,\nand merge a many-item fan-out into one table (item, verdict, evidence). An item whose return\nlacks a field its brief requires enters that table as a gap, never as a pass, unless a single\nconsolidated-scope respawn returns the field. Cite sources you actually fetched;\nnever label a claim \"known\" \u002F \"from memory\" \u002F \"obvious\".\n\n**Priming addendum (current session only).** As the main session, not a spawned non-fork worker,\nyou may also reach orchestration surfaces a non-fork worker cannot: agent teams (driven from the\nlead session; the docs do not state whether a fork of the lead can drive one; see\n[agent teams: architecture](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fagent-teams#architecture) and\n[limitations](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fagent-teams#limitations)) and dynamic workflows (withheld from non-fork workers). This session's reasoning effort is `${CLAUDE_EFFORT}`; if that value reads as a literal\nplaceholder, resolve the session's effort yourself before using it. Feed the value\ninto imperative 7's tier calibration: we treat it as the effort an unpinned spawn may run at when it\npasses no `effort`, and the pointer below says on which model that holds. Its gap from what a subtask needs IS\nthe over-provisioning imperative 7 exists to stop. Close that gap by passing the level on a\nnon-fork spawn of an unpinned agent, and omit it on a pinned named agent. Never read ultracode from this value. Pointer: for\nthe subagent `effort` field, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#supported-frontmatter-fields>;\nfor the per-spawn `effort` and its precedence, see\n\u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#choose-an-effort-level>; for which model the session's\nlevel reaches, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmodel-config#effort-level-of-subagents-and-teammates>;\nfor how ultracode relates to\neffort, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmodel-config#adjust-effort-level>. Where a\n`SendMessage` tool resolves in this session, run imperative 4's worker reuse and mid-flight\nintervention through it, addressed by the worker's agent ID. Never re-invoke the dispatch tool to\ncontinue a worker: that starts a second, independent worker. Read a refused message as a worker\nthe user stopped. Pointers and the empirical probe: `context\u002Fsources.md`, \"SendMessage\nworker continuation\".\n`worker_continuation` decides how a worker's next unit (a fix round, a follow-up, a retry, the\nnext queue item) runs. Resolve it once, at arm time, lowest layer first: the default `resume`; the\nuser's option `${user_config.worker_continuation}`, where a literal, unexpanded placeholder means\nunset; then the repository's `docs\u002Fconventions\u002Fsession-flow.yaml`, which wins when it sets the key\n(schema: [`${CLAUDE_PLUGIN_ROOT}\u002Fschemas\u002Fsession-flow.schema.json`](${CLAUDE_PLUGIN_ROOT}\u002Fschemas\u002Fsession-flow.schema.json)).\nRead the repository value with\n`bash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fretro\u002Fscripts\u002Fparse-concern-value.sh\" --strict \"\u003Cgit root>\u002Fdocs\u002Fconventions\u002Fsession-flow.yaml\" worker_continuation`,\nand only when the working directory's git root is neither `$HOME` nor an ancestor of it; otherwise\nskip that layer and say so. Empty output means unset only when the key is absent: if\n`grep -qE '^worker_continuation[[:space:]]*:' \"\u003Cfile>\"` matches, the key is present but empty or not\na scalar, which is invalid. Exit 3 means the parser rejected the file: name the file and drop the\nlayer. A value other than `resume` or `respawn` is named with its file or option, the key and the\nvalue, and that layer is dropped: the repository's valid value still wins over an invalid user\nvalue, and an invalid repository value or file resolves the default `resume`, never the user's\nvalue; the run never stops on it. Report one line, for\nexample `worker_continuation: respawn (docs\u002Fconventions\u002Fsession-flow.yaml)`. Under `resume`,\nimperative 4's reuse stands. Under `respawn`, each new unit goes to a fresh worker whose brief\nconsolidates the scope (the original brief, every later directive, the prior worker's report and\nits branch); resume instead only when the unit needs state that lives in that worker and is\ncostly to move: its checkout, uncommitted changes, or a running process. Settings page:\n[`${CLAUDE_PLUGIN_ROOT}\u002Freference\u002Fconfig.md`](${CLAUDE_PLUGIN_ROOT}\u002Freference\u002Fconfig.md).\nTo choose between a workflow and subagents, and the model and effort each spawn gets, run\n`\u002Fmulti-agent:assess` and `\u002Fmulti-agent:route` when they resolve in this session; otherwise read\n\u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#choose-a-model> (record: `context\u002Fsources.md`,\n\"Priming addendum: model and effort routing\").\nResolve the repository's verification level once with\n`bash \"${CLAUDE_PLUGIN_ROOT}\u002Flib\u002Fread-verification.sh\" --plugin session-flow --skill orchestrate --file docs\u002Fconventions\u002Fsession-flow.yaml`\nand report its line ([levels](https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fdocs\u002Fcontracts\u002Fverification-level\u002FREADME.md)).\nIt governs imperative 3 in this session: under `light`, and `off`, which acts as `light` here,\nhand nothing to a verifier and never self-audit in its place; write `verification: skipped\n(\u003Clevel>)` where the verdict would go. An exported brief carries no level: its target resolves\nits own.\nExport modes omit this addendum, a\npasted target reaches none of those surfaces, and the substitution would travel as dead text.\n\n## Tiered delegation, the shape of a deep tree\n\nGuidance for the main session once a task needs more than one layer; the export brief omits it,\nbecause a pasted worker sits inside a tree rather than authoring one.\n\n**A rough anchor for small\u002Fmedium\u002Flarge.** Not thresholds to enforce, the judgment still runs on context\nboundaries, not head-count, but our anchor is the platform's workflow size guideline values,\nread live. A run large enough to trip the platform's large-workflow warning is a size to justify\nout loud. An order-of-magnitude disagreement with this anchor is one to name, not skip. Before a\nworkflow run, read the size guideline in force for this session: the guideline the session sets\nis the one Claude receives, whatever this anchor says.\n\n- **Pointer**: for the size guideline values and their defaults, see\n  \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fworkflows#set-a-size-guideline>; for the large-workflow\n  warning, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fworkflows#cost>.\n\n**The top of the tree owns the loop, not the work.** Its context is the scarcest in the run,\neverything that enters it stays for the rest of the session. So it holds the objective, the\nstopping condition, and the decision about what to spawn next, and it delegates the rest. It keeps\nthe task list in a file (the plan or checklist the work already has, else one it creates), ticks\neach item as its return is verified, keeps an item it will not do listed as `skip: \u003Creason>`, and\nreads that file rather than the scrollback. It keeps going\nwhen a step needs no input, with status notes in the same message as the next action, and stops to\nask only when blocked on the user or before a destructive, hard-to-undo, or outward action. A top\ntier that reads findings, weighs them, and asks a follow-up question has converted a fan-out into a\nconversation, and the context it was protecting fills anyway.\n\n**Chatter belongs low.** Two workers resolving an ambiguity between themselves costs nothing at the\ntop. The same exchange routed through the parent costs the parent's window twice and permanently.\nPush coordination to the lowest tier that can resolve it, and let each tier return a compressed\nverdict rather than its reasoning.\n\n**Spec what crosses a boundary, not just what to do.** Every spawn already needs an objective and\nan output format (imperative 2). In a multi-tier tree the output format IS the context-economy\nlever: name the identifiers, the verdict, and where the bulky payload was parked, so the tier above\ncan act without re-reading the work. Our magnitude for \"compressed\": a worker that explored across tens\nof thousands of tokens still aims to return roughly 1,000 to 2,000. Treat it as the shape a return\nshould aim for, never a budget to spend up to.\n\n- **Pointer**: for a subagent returning a summary in place of its verbose output, see\n  \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#isolate-high-volume-operations>\n  (correlate with \u003Chttps:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Feffective-context-engineering-for-ai-agents>).\n  No docs page states a return size.\n\n**Workers are ephemeral, and the deeper the tier the shorter the life.** A worker that finishes and\nstays alive keeps costing the tier above, notifications, status, re-acknowledgement, for zero\nadditional output. Retire on completion. When the next grouping needs doing, spawn fresh rather than\nreusing a worker whose context now carries the last job. (The exception is imperative 4's long-lived\nworker across *related* subtasks, where cache reuse is the point; that is a deliberate trade, not\nthe default.)\n\n**Check one sampled brief per wave, beside the wave.** Pick one brief from each wave you dispatch\nand check it against imperative 2 while that wave runs. The check never gates the wave: what it\nfinds corrects the next wave's briefs, and a defect that would change a running worker's result\ngoes to that worker as a directive.\n\n**Treat a clean return as unverified, especially a suspiciously clean one.** An under-specified\nworker rarely stalls and asks; it substitutes the nearest plausible interpretation and reports\nsuccess. In a fan-out, most workers given a brief missing a resource will locate it and close the\ngap, and one will silently audit a different, similar artifact and return a confident,\nwell-formed, wrong-target result that nothing in its return distinguishes from the others. This\nis why imperative 3's fresh-context verify is not optional at depth, and why a return payload\nbenefits from naming its sources. Provenance is the field that makes a wrong-target answer\ndetectable from above.\n\n**Never author a tree that needs a specific depth.** The platform's nesting default is\nconfigurable and changes. Agent-tool subagents carry a depth cap and a concurrency cap (their\nsettings: the pointers below); workflow agents\nand agent-team teammates carry their own, and workflow concurrency has its own override, so \"read\nthe current values\" includes the workflows page whenever the run will use the Workflow tool. Read the current values rather than assuming\nthem, and design the tree so it degrades to a shallower one instead of failing. Never design a\ntree that needs a fork to spawn a fork: that is a shape constraint, not a tunable. Whether a\nbelow-limit fork can parent non-fork children is unconfirmed, so do not treat a fork as a\nforbidden intermediate tier either. The version history behind the caps lives in\n`context\u002Fsources.md`.\n\n- **Pointer**: for the depth cap, see\n  \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#let-subagents-spawn-their-own-subagents>; for the\n  concurrency cap, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#concurrent-subagent-limit>; for\n  workflow limits, see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fworkflows#behavior-and-limits>; for forks,\n  see \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#how-forks-differ-from-other-subagents>.\n\n**Confirm nesting from behavior, not from one page** (why: `context\u002Fgotchas.md`). Have a\nworker of the SAME definition you plan to use as the intermediate tier attempt a trivial nested\nspawn and report the outcome. The gate is definition-specific, so another agent type proves\nnothing, and holding `Agent` is necessary but not sufficient. Read a refusal: a depth rejection\nnames depth; a permission refusal (classified pre-launch) does not. Pointers: `context\u002Fsources.md`.\n\n## Export modes (handoff \u002F worker). Paste-ready brief\n\nOnly for a target that LEAVES the session. Emit the seven imperatives above between two full-width\n`─` (U+2500) dashed rails. Top rail, brief, bottom rail, nothing else between them; the\n`\u002Fclear`\u002Fpaste instruction or any commentary sits above the top rail or below the bottom rail,\nnever between (NOT a code fence, the user copies the text between the rails, not fence markers).\n\nLive shape: bare `─` rails, no fence. Shown inside a fence here for display only.\n\n```text\n──────────────────────────────────────────────────────────\nORCHESTRATION BRIEF — standing instructions for the whole task, regardless of which model or tool runs you.\n\nAt each decision boundary, evaluate these and ACT on a match without waiting to be told:\n[the seven numbered imperatives above, verbatim]\n\nDiscipline: [the Discipline line above, verbatim]\n──────────────────────────────────────────────────────────\n```\n\n- `handoff`, the opening line above already fits a fresh session; emit as-is.\n- `worker`, insert as the FIRST line between the rails: `You are a spawned worker and did NOT\n  inherit the parent session's context or the repo's conditional rules, these instructions are\n  your only copy.` Then insert this waiting line directly after it, before the `ORCHESTRATION BRIEF` line,\n  in full and `compact` exports alike: `Waiting: poll in the foreground with a bounded loop, or\n  return what you have; never end your turn to wait on your own background job.`\n- `compact`. Emit only the seven numbered HEADLINES (`1. DELEGATE \u002F FAN OUT`, `2. SPEC EVERY\n  SPAWN`, …) plus the closing Discipline line; drop every sub-clause. `worker compact` also keeps the\n  `worker` did-not-inherit and waiting lines; compact drops neither.\n\n## Next\n\n\u002Fplanning:draft-goal-condition turns the coordinated work into a completion condition.\n\n## What this skill does NOT do\n\n- **Default does not emit paste-text.** Priming the current session is a terse acknowledgment,\n  the work happens because the imperatives loaded into context, not because anything was printed.\n  Use `handoff` \u002F `worker` only when the target LEAVES the session.\n- **Not a surface-selection guide.** Which parallel-execution surface to pick (subagents vs nested\n  vs teams vs workflows) is a judgment the main session makes against current official docs; the\n  export brief deliberately omits agent teams + dynamic workflows because a pasted target cannot\n  reach either.\n- **Does not delegate, verify, nest, or spawn anything itself**. It arms the session or emits\n  instruction text.\n",{"metadata":24,"description":7,"argument-hint":27,"user-invocable":28,"disable-model-invocation":20},{"summary":25,"workflow-stage":26},"Arm the session with proactive-orchestration imperatives","session","[\u003Ctask>|handoff [compact]|worker [compact]|coordinator [\u003Carea>]]",true,"main","5aff8ba54896bc54fd0eb46e4a954099449cf67d","plugins\u002Fsession-flow\u002Fskills\u002Forchestrate",7,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins","https:\u002F\u002Fgithub.com\u002Fmelodic-software","https:\u002F\u002Fmelodicsoftware.com","Independent software studio building developer tools, automation, and application services, with roots in music.","2026-11-15 05:15:06",{"source":8,"skillName":5,"commands":39,"prompt":49,"zipAvailable":28},[40,43,46],{"tool":41,"command":42},"npx","npx skills add melodic-software\u002Fclaude-code-plugins --skill orchestrate",{"tool":44,"command":45},"bunx","bunx skills add melodic-software\u002Fclaude-code-plugins --skill orchestrate",{"tool":47,"command":48},"pnpm","pnpm dlx skills add melodic-software\u002Fclaude-code-plugins --skill orchestrate","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fsession-flow\u002Fskills\u002Forchestrate\u002FSKILL.md",[51,57,61,65],{"path":52,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":55,"sourceHtmlUrl":56},"context\u002Fcoordinator.md","REFERENCE","参考文档",2940,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fsession-flow\u002Fskills\u002Forchestrate\u002Fcontext\u002Fcoordinator.md",{"path":58,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":59,"sourceHtmlUrl":60},"context\u002Fgotchas.md",2893,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fsession-flow\u002Fskills\u002Forchestrate\u002Fcontext\u002Fgotchas.md",{"path":62,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":63,"sourceHtmlUrl":64},"context\u002Fsources.md",17128,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fsession-flow\u002Fskills\u002Forchestrate\u002Fcontext\u002Fsources.md",{"path":66,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":67,"sourceHtmlUrl":68},"evals\u002Fevals.json",22995,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fsession-flow\u002Fskills\u002Forchestrate\u002Fevals\u002Fevals.json",[70,77,84,91,96,103],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M7DYYDG31B9YX7N9MB9QW2GW","track","Backlog reads through the bound tracker: search open and closed items, view one item with its comments and state history. Use when another skill searches or reads tracker items.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fsession-flow\u002Fevals\u002Frecall-still-reports-reopened-bug\u002Fwork-items-stub\u002Fskills\u002Ftrack\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":19},"01M7DYXVBAVG3NV30PA7GKW9Q6","reviewer-media","Reviewer media for one pull request: decides whether it needs a narrated walkthrough, builds the storyboard from \u002Freview:explain-change's digest, has \u002Fexplainer-video:produce render it and hands the manifest to \u002Fsource-control:pull-request media. Use when asked for 'reviewer media', 'a walkthrough video for this PR', 'a produced video of this change', or when pull-request ready or monitor hands a PR over.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Freview\u002Fskills\u002Freviewer-media\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":19},"01M7DYRZ5YD44AMYH26YR25GZ8","route","Model and effort per agent role (orchestrator, worker, verifier, retrieval) for a multi-agent run, resolved through the bundled defaults and the user, team and personal override config layers into fenced JSON with single-agent and fan-out variants and the layer behind each value. Use when asked 'which model for the subagents', 'route the roles' or 'model routing', or before a skill passes a role's model on a spawn.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fdiscovery\u002Fevals\u002Fex-model-pin-without-override\u002Fmulti-agent-stub\u002Fskills\u002Froute\u002FSKILL.md",[],[],{"skillNo":92,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"publishedAt":19},"01M7DYRTVT4FWTGDJY0HCCFJMZ","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fdiscovery\u002Fevals\u002Fex-model-override-from-route\u002Fmulti-agent-stub\u002Fskills\u002Froute\u002FSKILL.md",[],[],{"skillNo":97,"name":98,"nameZh":6,"description":99,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":100,"categories":101,"keywords":102,"publishedAt":19},"01M7DYNZC8PNW3T27V2ADKECKJ","principles","Module-design vocabulary for architecture findings, and the checks built on it. Use when: 'what is a deep module', 'what is a seam', 'deep vs shallow module', 'what is the deletion test', 'leverage and locality', 'module design vocabulary', 'which word do we use for X in architecture findings'. Scanning a codebase for friction is \u002Farchitecture:improve.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Farchitecture\u002Fskills\u002Fprinciples\u002FSKILL.md",[],[],{"skillNo":104,"name":105,"nameZh":6,"description":106,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":107,"categories":108,"keywords":109,"publishedAt":19},"01M7DYNMCZNEWBFHV9KTSS6H60","write","Structured five-field bug report (title, steps to reproduce, expected vs actual, justified severity, suggested fix location) from an informal description, read-only. Use when the user names a defect they observed, or asks for one written up ('report a bug', 'file a bug', 'write this up as a bug').","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Farchitecture\u002Fevals\u002Fdeepening-defect-goes-to-bugs-write\u002Fbugs-stub\u002Fskills\u002Fwrite\u002FSKILL.md",[],[]]