[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6C4K9SNZVR3TKNWMVQF45ZJ":3,"skill-same:01M6C4K9SNZVR3TKNWMVQF45ZJ":81},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":22,"publishedAt":42,"linkOnly":43,"markdownContent":44,"markdownBody":45,"frontMatter":46,"refName":50,"commitSha":51,"skillDirectory":52,"forksCount":13,"repositoryHtmlUrl":53,"repositoryHomepageUrl":6,"ownerHtmlUrl":54,"ownerBlogUrl":55,"ownerDescription":6,"updatedAt":56,"install":57,"resources":70},"01M6C4K9SNZVR3TKNWMVQF45ZJ","resume",null,"Opens a work session — reads the continuity docs and recent state, synthesizes where things actually stand, proposes a re-prioritized next scope with any derived prerequisites or objections, and settles the decisions `\u002Fiyu:handoff` flagged the way a capable employee would — grounded options, trade-offs, and a recommendation that becomes the provisional decision, recorded at once for the owner to correct rather than answer. Only external blockers, owner-gated acts, and criterion gaps are left waiting on the owner. Then stops — execution is a separate, explicit step.","iyulab\u002Fclaude-plugins","iyulab","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F52388508?v=4","01M6C4JT2KEXFSG8YWSMFYT0NQ","01M6C4JT8BEJCZ0VPCC69N9JDN",0,"MIT","https:\u002F\u002Fgithub.com\u002Fiyulab\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fiyu\u002Fskills\u002Fresume\u002FSKILL.md",[17],{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000038","人力资源","PROFESSION","适用职业",[5,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41],"trade-offs","focus","self","fresh","websearch","when","invoke","continuity","cycle","re","scope","states","work","Bash","only","decision","short","model","never","2026-11-10 10:12:30",false,"---\nname: resume\ndescription: Opens a work session — reads the continuity docs and recent state, synthesizes where things actually stand, proposes a re-prioritized next scope with any derived prerequisites or objections, and settles the decisions `\u002Fiyu:handoff` flagged the way a capable employee would — grounded options, trade-offs, and a recommendation that becomes the provisional decision, recorded at once for the owner to correct rather than answer. Only external blockers, owner-gated acts, and criterion gaps are left waiting on the owner. Then stops — execution is a separate, explicit step.\nwhen_to_use: User-invoked with \"세션 시작\", \"다음 작업 확인\", \"핸드오프 확인하고 시작\", \"resume this session\". Model-invoked (announce before running, per rule 8) when a fresh session's first message states no explicit task and the continuity root already has a `HANDOFF.md` — the same self-invoke bar `\u002Fiyu:handoff` uses, since this skill proposes and decides but never implements. Not for mid-session re-checks (read `HANDOFF.md` directly for that), and not a substitute for `\u002Fiyu:run-cycle` — resume always stops short of execution; run-cycle is the separate, explicit step that starts it.\nargument-hint: \"[focus or scope note]\"\nallowed-tools: Read, Glob, Grep, Edit, TodoWrite, Bash, WebSearch, WebFetch\n---\n\n# Session Resume\n\nRead what `\u002Fiyu:handoff` left, synthesize where the work actually stands right now, and settle\nanything flagged as undecided — before any real work begins. Work this like a capable employee\npreparing to open the next work block, not like someone mechanically continuing where a diff left off:\npull the continuity docs and current repo state together, judge whether the recorded order still holds,\nsurface what it's missing, and question what looks wrong — then hand back a settled scope and **stop**.\nStarting the work is a separate act this skill never takes on its own.\n\nWhy it is shaped this way — split from `\u002Fiyu:handoff`, pre-deciding instead of asking, asking for\ncriteria instead of decisions — is in\n[references\u002Frationale.md](${CLAUDE_SKILL_DIR}\u002Freferences\u002Frationale.md).\n\n## Scope\n\n`$ARGUMENTS`, when given, narrows what to confirm (a submodule, \"just the decisions\"). Without it,\ncover the whole continuity root.\n\n## Process\n\n### 1. Resolve the continuity root\n\nSame rule as every other skill in this plugin —\n[continuity-docs.md §1](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md). In an umbrella repo,\nresolve the root covering what this session is actually about to touch; if that is not yet known,\nask.\n\n### 2. Read — ground the synthesis, don't redo the archaeology\n\nRead `\u003Croot>\u002FHANDOFF.md` and `\u003Croot>\u002FROADMAP.md` as they stand. Then ground judgment in current\nstate, not just the doc text: `git status` \u002F `git log -1` since the last handoff, the tail of\n`HISTORY.md` for what actually shipped, `\u003Croot>\u002FSTRANDS.md`'s `## 중단됨` section (if the file\nexists) for anything recently sidelined, and recent memory (`MEMORY.md` and relevant memory files)\ntouching this root. Run the session-boundary inbox check if the project defines one\n([continuity-docs.md §6](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md)) — what came in since the\nlast handoff is part of current state. This is not `\u002Fiyu:handoff`'s evidence reconstruction — no cycle-log reading, no\nhygiene pass, no migrating anything to `HISTORY.md`. Those stay `\u002Fiyu:handoff`'s job; this read grounds\nstep 3's judgment in what changed since, it does not redo the archaeology.\n\n**A stage contract changes what \"where things stand\" means.** If `\u003Croot>\u002Fstages\u002Fstages.json` has an\n`open` stage ([stages.md](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fstages.md)), read it and its newest\n`evidence\u002F` file, and lead the synthesis with the open stage: its id, scenarios passing x\u002Fy, the latest\nverdict, and every `owner_decisions` entry still `open` — each with the stage it opens, the one that\nunlocks the most first. The next step it proposes is `\u002Fiyu:drive` (or `\u002Fiyu:charter` when the stage is\n`passed` and the next is `upcoming`). A stage whose newest evidence is `fail`\u002F`unverified` with\n`status: open` is an interrupted drive; say so.\n\nNo `HANDOFF.md` → say so, point at `ROADMAP.md` or `\u002Fiyu:backlog-discover` if neither exists, and\nstop — there is nothing to resume.\n\n### 3. Synthesize and present In flight + Next\n\nDon't just replay `## In flight` \u002F `## Next` as written — apply judgment to what step 2 surfaced:\n\n- **Flag staleness.** If `## In flight` describes something half-done but the tree is clean, or the\n  last commit contradicts what `## Next` assumes, say so before presenting either — a stale doc handed\n  over silently defeats the point of resuming from it.\n- **Surface a recent interruption.** If `STRANDS.md`'s `## 중단됨` has an entry interrupted within\n  roughly the last session (freshest `마지막` entry relative to `HANDOFF.md`'s own last-updated\n  date), treat it as an objection: judge whether to resume `{strand}` now or keep the current order,\n  and carry that call into step 4 as a decision (\"`{strand}` 재개를 X 앞에 둠 — 이유 …, to correct:\n  …\"), never as a question back to the owner. This is not new discovery\n  (rule 9): the strand is already-known, already-scoped work being reordered, not scoped from\n  scratch. An entry that has stayed interrupted far longer than one session is not this skill's\n  concern — that is `\u002Fiyu:backlog-discover`'s `dormant-strand-review`, which judges reignite,\n  shrink-and-resume, or retire with evidence rather than a scheduling nudge.\n- **Reorder** when the recorded priority no longer fits current state (a dependency landed, a blocker\n  resolved, an external deadline shifted) — show the new order next to the recorded one, with the\n  reason (step 4 decides it).\n- **Derive prerequisites** the plan is missing — a step it silently assumes is already done, a\n  dependency it doesn't name. Add these as visible insertions, not silent edits.\n- **Object** to anything in `## Next` that no longer looks reasonable given current state — say why,\n  and what you'd do instead.\n\nThis is judgment over what the continuity docs and current state already contain — not discovery of new work.\nDiscovering work nobody has scoped yet stays `\u002Fiyu:backlog-discover`'s job (rule 9); if that's what's\nactually needed, recommend it, don't attempt it here.\n\n**Split each list into 코드 작업 \u002F 비코드 작업.** Code work is anything that changes what ships\n(source, tests, config, build). Non-code work is everything else the continuity docs track — docs, an\nissue draft, a roadmap edit, a release or external-communication step. The presentation split is what\ncarries the real payload to step 4 — that's where it's read to spot owner-gated acts, which live in\nthe non-code bucket — so a bucket with nothing in it just reads \"None\" rather than being dropped.\n\n**`## Next` empty, or every item in it blocked → scan `ROADMAP.md` first.** Each startable phase\nthere (classified per [continuity-docs.md §4](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md) —\norder is not a gate; a direction-only phase is startable as decomposition) is a staleness flag on\n`## Next` and goes to step 4 as a reorder or prerequisite insertion.\n\n**No startable phase in `ROADMAP.md` either → say so, put the goal-gap review\n([continuity-docs.md §4](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md)) first in `## Next`, and\nrecommend `\u002Fiyu:backlog-discover` for the wider lanes.** Same closing move `run-cycle`'s\nEnd-of-Run Report makes when a run ends on `FRONTIER-EXHAUSTED:` with budget left: an empty backlog\nisn't this skill's gap to fill — step 3's synthesis reorders and extends what's already scoped, it\ndoes not discover unscoped work from nothing (rule 9) — it's a pointer to the skill that does. A\npointer only: name the recommendation, do not invoke it. Only say this when no startable work is left;\nan empty `## Waiting on you` with items still in `## Next` is step 4's good outcome, not this one.\n\n### 4. Settle what needs deciding — decide it, don't hand it back\n\nTwo sources feed this step: handoff's `## Waiting on you` entries, and the reorder \u002F prerequisite \u002F\nobjection proposals step 3 raised. A proposal from step 3 is decision-class the moment it changes what\nships or in what order — where a decision came from doesn't change how it's handled.\n\n**The default is that you decide.** Handing a decision up as a question gives the owner both the\nwork and the wait. A recommendation, once formed, *is* the decision: recorded `provisional` in step\n5 and presented for correction, not for an answer. What reaches the owner as something to *supply*\nis only what you genuinely cannot — a resource from outside, the owner's hand on a gated act, or a\ncriterion.\n\nFor each entry, classify and handle:\n\n- **Resource-blocked** (blocker · tried · unblocks) — re-run the self-unblock check (the same\n  \"Self-unblock before parking\" logic `run-cycle` applies): is the resource *actually* still missing?\n  Time has passed since it was parked. Still blocked → it stays in Waiting on you. Resolved → say so\n  and drop it.\n- **Decision-class** — brief it per\n  [decision-briefing.md §2](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fdecision-briefing.md): the decision in one\n  line, **at least two grounded options** (feasibility and cost from what you read in step 2 or the\n  grounding pass below — never invented), the **cross-lens read**\n  ([decision-lenses.md](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fdecision-lenses.md), only the lenses that\n  separate the leading options), and a **named recommendation** with what it locks in. **Then decide\n  it**: the recommendation is the provisional decision. Present it as \"결정: X — 트레이드오프 …,\n  이견 있으면 알려주세요\", never as \"A와 B 중 어느 쪽으로 할까요?\".\n  - **Decide ≠ act.** If carrying the decision out needs an act the central policy gates on a human —\n    `git push`, publish\u002Frelease, registering a GitHub issue, a major-version bump — or an act that is\n    irreversible outside this repo, the *decision* is still made and recorded; only the *act* goes to\n    Waiting on you, in the resource-blocked shape where the owner is the resource: **blocker** = the\n    gate · **tried** = decision settled (`→ Decided this session`) · **unblocks** = approval to run\n    `\u003Cthe act>`. Confidence never lifts the gate; the gate never un-makes the decision.\n  - The step-3 split (코드\u002F비코드) is how you spot such acts: a code decision reverses through git and\n    has none; a non-code decision is where push\u002Fpublish\u002Fregistration live.\n- **Criterion request** — reachable only *after* the grounding pass, when the decision still will not\n  settle because (a) the lenses conflict irreducibly, or (d) a criterion does exist but is ambiguous\n  here or, followed faithfully, lands somewhere common sense says is wrong (decision-lenses.md's\n  cause table). **Do not ask the owner to pick an option — ask for the criterion.** Name the criterion\n  and where it breaks, propose the revised criterion, and **decide under it** (provisional, as\n  above): \"판단기준 X가 이 지점에서 모호\u002F상식에 반함 → X′로 보완 제안, X′ 기준으로는 Y로 결정. X′를\n  확정해 주세요.\" The decision does not wait for the answer; only its gated act, if any, does.\n\n**Grounding pass — settle the unknown instead of reporting it.** When a recommendation does not form\non the first pass, close the gap before writing anything:\n\n- **Name the one unknown that separates the leading options.** Not everything unknown about the\n  decision — the single thing whose answer would move the pick. If nothing would, there is no gap and\n  the recommendation was already available.\n- **Settle it, bounded (~5 min per entry)**, per decision-lenses.md's cause table — (b) read the\n  file\u002Ftest\u002Fdependent or `WebSearch`\u002F`WebFetch` the convention; (c) derive the missing criterion\n  from CLAUDE.md\u002FREADME and\n  [philosophy-alignment-guide.md](${CLAUDE_SKILL_DIR}\u002F..\u002Fmindset\u002Freferences\u002Fphilosophy-alignment-guide.md) and recommend it with the option it\n  implies. Run the cheap command rather than naming it. Only (a) and (d) become criterion requests,\n  and only once (b) and (c) are ruled out.\n- **If the bounded search came back empty, say what you searched**, not only what is still unknown.\n\nThis does not widen the skill (rule 9). Grounding a decision that is *already on the table* is\njudgment over known work. Scoping work nobody has identified yet is still `\u002Fiyu:backlog-discover`'s\njob.\n\n**After this step, Waiting on you holds only resource-blocked entries (external, or an owner-gated\nact) and criterion requests.** A decision-class entry never remains there as an open question.\nNothing to decide → \"None\", and move straight to confirming In flight \u002F Next. Do not manufacture a\ndecision to fill the section.\n\n### 5. Record immediately, then take feedback\n\n**Record before presenting, and do not wait to record.** Every decision step 4 made goes into\n`\u003Croot>\u002FHANDOFF.md`'s `## Decided this session` (create the section if absent) in the shape\n`\u002Fiyu:handoff` uses — decision · trade-off · **to correct: \u003Cthe reverse action>** — marked\n`provisional`, with the lenses that carried it. In the same edit, rewrite `## Waiting on you` to what\nstep 4 left there: a flag that got decided moves out (its gated act, if any, stays as a\nresource-blocked entry pointing at the decision); criterion requests go in. Do this now, not at the\nnext `\u002Fiyu:handoff` — a session that ends first must not lose it.\n\nEvery entry left standing keeps its **Registered** date and **Blocks** line from `\u002Fiyu:handoff`\n(a new one gets today's date) and gains the two lines only this step can write:\n\n- **Default** — what stands while the owner has not answered: the provisional decision the entry\n  waits on, or, for a criterion request, the decision taken under the proposed criterion.\n- **Meanwhile** — what work proceeds without it, and what does not. **Never the gated act itself**:\n  silence does not run a push, a release, or an issue registration (rule 3), so \"on no reply: publish\"\n  is not a Meanwhile line.\n\nWhen the project keeps a decision ledger\n([continuity-docs.md §2](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md)), each decision goes\ninto it as well, in the ledger's own form, and `## Decided this session` carries its ID and the\nto-correct line rather than a second copy of the entry.\n\n**A reorder is applied to `## Next` itself, in the same edit** — self-decided or owner-confirmed.\nRewrite that section into the decided order (keeping the 코드\u002F비코드 split), and mark the\n`## Decided this session` entry as applied.\n[continuity-docs.md §2](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md) makes `## Next` the\nauthoritative statement of what comes next — what `\u002Fiyu:run-cycle` Preparation reads as its opening\nscope — so a decision recorded only in prose is one a later session can start without ever seeing.\n\n**`ROADMAP.md` is still not this skill's to write (rule 1).** `\u002Fiyu:handoff` reads\n`## Decided this session` as an inherited fact and carries a reorder into `ROADMAP.md`; an entry\nmarked applied means applied *to `## Next`* only.\n\n**Feedback is how the criteria get corrected — handle it, don't just transcribe it.** This applies\nwhen an answer arrives later in the same conversation, after step 6 has presented and stopped —\nresume does not wait for it.\n\n- **Owner confirms or says nothing** → the provisional decision stands. Silence does not block a\n  decision (it was already made); it never authorizes a gated act (rule 3).\n- **Owner picks differently from the recommendation** → apply the new pick (update the entry and\n  `## Next`), then ask one alignment question instead of silently moving on: **which criterion\n  weighed differently — should it be revised, or is this a one-off exception?** Name the lens or\n  rule that drove the recommendation so the owner is correcting something specific (\"근본을 세련보다\n  무겁게 봤는데, 이 프로젝트에선 반대로 둘까요?\").\n- **A criterion gets confirmed** (from that question, or an answered criterion request) → record it\n  in `## Decided this session` as a criterion entry, re-derive any pending decision under it (update\n  the entry if the pick changes), and insert a 비코드 `## Next` item to persist it where the\n  criterion lives — the project's `CLAUDE.md`\u002F`AGENTS.md`, or, for a plugin-level lens, a\n  note in this plugin's own backlog. This skill does not write those files itself (rule 1).\n- **One-off exception** → record it as such in the entry, so a later session does not read it as\n  precedent.\n\n### 6. Stop — hand off, don't execute\n\nOnce In flight \u002F Next is settled (including step 3's proposals), every decision is recorded, and\nWaiting on you holds only blockers and criterion requests, the session-opening work is done. **This\nskill does not proceed into implementation from here.** Present the settled scope as one **briefing\ncard** — always this order, an empty block is one line (\"None\"), the labels in the session's language:\n\n1. **State** (≤ 5 lines) — where things stand, staleness flags, what changed since the handoff.\n2. **Q** to confirm — criterion requests, plus any decision the project's own policy reserves for the\n   owner (step 4 resolves nothing else this way): each with the recommendation, the alternative and when\n   it wins, the one fact that would flip it, and what stands if there is no reply.\n3. **G** to approve — gated acts whose decision is settled: each names the act and **what a yes opens\n   — that act only**. **H** to provide — what only the owner can supply (the step, never the secret).\n4. **D** decided by me — this session's provisional decisions, each with its reverse action.\n5. **C** code \u002F **N** non-code — the settled `## Next` in order; each row `[auto]` (startable with no\n   answer) or `[after Gn]` \u002F `[after Hn]` (the one row it waits on).\n6. **Reply** — the grammar and one example built from this card's codes: \"adopt all\" confirms every Q\n   at its recommendation and every D and opens **no** G or H; \"approve G1 G3\" opens exactly those\n   acts; \"Q2 → alternative\" \u002F \"revert D3\" overrides (step 5 then asks which criterion differed); \", start\"\n   is the explicit instruction below; no reply leaves Q and D standing and runs no G.\n\nNumber rows per briefing (`Q1`, `G2`), with the durable id beside it when one exists (`HD-12`). Row\nshapes, guards and a worked example: [references\u002Fbriefing-format.md](${CLAUDE_SKILL_DIR}\u002Freferences\u002Fbriefing-format.md).\nThen stop. Starting the work is a separate, explicit act: the human instructs it directly (a reply ending\n\"start\" counts), or `\u002Fiyu:run-cycle` is invoked. A settled scope is not that instruction. If the owner\nanswers with feedback instead, handle it per step 5's feedback rules, then stop again.\n\n## Rules\n\n1. **`HANDOFF.md` is the only file this skill writes** — plus the project's decision ledger, when\n   its instructions define one, for the decisions step 4 made. It never migrates `ROADMAP.md` →\n   `HISTORY.md`, never touches `ROADMAP.md`, `CLAUDE.md`, or `AGENTS.md`, and never runs\n   `\u002Fiyu:handoff`'s hygiene pass. Inside `HANDOFF.md` it writes exactly three things (step 5):\n   `## Decided this session` entries, `## Next` (a decided reorder, and any criterion-persistence\n   item), and `## Waiting on you` (what step 4 left there). Phase-level order in `ROADMAP.md` stays\n   `\u002Fiyu:handoff`'s to apply.\n2. **Ground options and proposals in what was actually read or run.** A step-3 reorder or\n   prerequisite needs a stated reason grounded in step 2's read, not a hunch. An unknown that\n   *separates the leading options* is **settled by step 4's grounding pass**, not reported as\n   unknown. Only an unknown that survives that bounded pass stays \"unknown\", reported together with\n   what was searched.\n3. **Decisions don't wait; gated acts do.** A recommendation is decided and recorded `provisional`\n   without waiting for a reply. An act the central policy gates on a human (push, publish\u002Frelease,\n   GitHub issue registration, major-version bump) or that is irreversible outside the repo never runs\n   on silence — silence is not consent for an *act*.\n4. **The owner is asked for resources and criteria, never for a pick.** Waiting on you holds\n   resource-blocked entries (including owner-gated acts) and criterion requests only. \"A or B?\" is\n   the one shape a decision may not reach the owner in.\n5. **An override is a signal about the criteria.** When the owner picks against the\n   recommendation, ask whether the criterion behind it should be revised — do not just record the\n   pick.\n6. **\"None\" is a complete outcome.** Nothing flagged and no step-3 proposal → say so; it is the good\n   result, not a gap to fill.\n7. **Never starts the work itself.** Step 6 ends the skill at a settled scope — implementation begins\n   only on a separate explicit instruction or `\u002Fiyu:run-cycle`. Reading, running a cheap command, or\n   searching to ground a decision is not starting the work.\n8. **Model-invoked runs announce first.** Same as `\u002Fiyu:handoff` rule 7 — say in one line that a\n   resume is running and why, before touching any file.\n9. **Synthesis is judgment over known work, not discovery of new work.** Step 3 and step 4's\n   grounding pass operate on what is already on the table — the boundary is what the work is aimed\n   at, not whether reading or searching happens. Scoping work nobody has identified yet stays\n   `\u002Fiyu:backlog-discover`'s job; a roadmap with no startable phase gets that skill *named*, never invoked.\n","# Session Resume\n\nRead what `\u002Fiyu:handoff` left, synthesize where the work actually stands right now, and settle\nanything flagged as undecided — before any real work begins. Work this like a capable employee\npreparing to open the next work block, not like someone mechanically continuing where a diff left off:\npull the continuity docs and current repo state together, judge whether the recorded order still holds,\nsurface what it's missing, and question what looks wrong — then hand back a settled scope and **stop**.\nStarting the work is a separate act this skill never takes on its own.\n\nWhy it is shaped this way — split from `\u002Fiyu:handoff`, pre-deciding instead of asking, asking for\ncriteria instead of decisions — is in\n[references\u002Frationale.md](${CLAUDE_SKILL_DIR}\u002Freferences\u002Frationale.md).\n\n## Scope\n\n`$ARGUMENTS`, when given, narrows what to confirm (a submodule, \"just the decisions\"). Without it,\ncover the whole continuity root.\n\n## Process\n\n### 1. Resolve the continuity root\n\nSame rule as every other skill in this plugin —\n[continuity-docs.md §1](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md). In an umbrella repo,\nresolve the root covering what this session is actually about to touch; if that is not yet known,\nask.\n\n### 2. Read — ground the synthesis, don't redo the archaeology\n\nRead `\u003Croot>\u002FHANDOFF.md` and `\u003Croot>\u002FROADMAP.md` as they stand. Then ground judgment in current\nstate, not just the doc text: `git status` \u002F `git log -1` since the last handoff, the tail of\n`HISTORY.md` for what actually shipped, `\u003Croot>\u002FSTRANDS.md`'s `## 중단됨` section (if the file\nexists) for anything recently sidelined, and recent memory (`MEMORY.md` and relevant memory files)\ntouching this root. Run the session-boundary inbox check if the project defines one\n([continuity-docs.md §6](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md)) — what came in since the\nlast handoff is part of current state. This is not `\u002Fiyu:handoff`'s evidence reconstruction — no cycle-log reading, no\nhygiene pass, no migrating anything to `HISTORY.md`. Those stay `\u002Fiyu:handoff`'s job; this read grounds\nstep 3's judgment in what changed since, it does not redo the archaeology.\n\n**A stage contract changes what \"where things stand\" means.** If `\u003Croot>\u002Fstages\u002Fstages.json` has an\n`open` stage ([stages.md](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fstages.md)), read it and its newest\n`evidence\u002F` file, and lead the synthesis with the open stage: its id, scenarios passing x\u002Fy, the latest\nverdict, and every `owner_decisions` entry still `open` — each with the stage it opens, the one that\nunlocks the most first. The next step it proposes is `\u002Fiyu:drive` (or `\u002Fiyu:charter` when the stage is\n`passed` and the next is `upcoming`). A stage whose newest evidence is `fail`\u002F`unverified` with\n`status: open` is an interrupted drive; say so.\n\nNo `HANDOFF.md` → say so, point at `ROADMAP.md` or `\u002Fiyu:backlog-discover` if neither exists, and\nstop — there is nothing to resume.\n\n### 3. Synthesize and present In flight + Next\n\nDon't just replay `## In flight` \u002F `## Next` as written — apply judgment to what step 2 surfaced:\n\n- **Flag staleness.** If `## In flight` describes something half-done but the tree is clean, or the\n  last commit contradicts what `## Next` assumes, say so before presenting either — a stale doc handed\n  over silently defeats the point of resuming from it.\n- **Surface a recent interruption.** If `STRANDS.md`'s `## 중단됨` has an entry interrupted within\n  roughly the last session (freshest `마지막` entry relative to `HANDOFF.md`'s own last-updated\n  date), treat it as an objection: judge whether to resume `{strand}` now or keep the current order,\n  and carry that call into step 4 as a decision (\"`{strand}` 재개를 X 앞에 둠 — 이유 …, to correct:\n  …\"), never as a question back to the owner. This is not new discovery\n  (rule 9): the strand is already-known, already-scoped work being reordered, not scoped from\n  scratch. An entry that has stayed interrupted far longer than one session is not this skill's\n  concern — that is `\u002Fiyu:backlog-discover`'s `dormant-strand-review`, which judges reignite,\n  shrink-and-resume, or retire with evidence rather than a scheduling nudge.\n- **Reorder** when the recorded priority no longer fits current state (a dependency landed, a blocker\n  resolved, an external deadline shifted) — show the new order next to the recorded one, with the\n  reason (step 4 decides it).\n- **Derive prerequisites** the plan is missing — a step it silently assumes is already done, a\n  dependency it doesn't name. Add these as visible insertions, not silent edits.\n- **Object** to anything in `## Next` that no longer looks reasonable given current state — say why,\n  and what you'd do instead.\n\nThis is judgment over what the continuity docs and current state already contain — not discovery of new work.\nDiscovering work nobody has scoped yet stays `\u002Fiyu:backlog-discover`'s job (rule 9); if that's what's\nactually needed, recommend it, don't attempt it here.\n\n**Split each list into 코드 작업 \u002F 비코드 작업.** Code work is anything that changes what ships\n(source, tests, config, build). Non-code work is everything else the continuity docs track — docs, an\nissue draft, a roadmap edit, a release or external-communication step. The presentation split is what\ncarries the real payload to step 4 — that's where it's read to spot owner-gated acts, which live in\nthe non-code bucket — so a bucket with nothing in it just reads \"None\" rather than being dropped.\n\n**`## Next` empty, or every item in it blocked → scan `ROADMAP.md` first.** Each startable phase\nthere (classified per [continuity-docs.md §4](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md) —\norder is not a gate; a direction-only phase is startable as decomposition) is a staleness flag on\n`## Next` and goes to step 4 as a reorder or prerequisite insertion.\n\n**No startable phase in `ROADMAP.md` either → say so, put the goal-gap review\n([continuity-docs.md §4](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md)) first in `## Next`, and\nrecommend `\u002Fiyu:backlog-discover` for the wider lanes.** Same closing move `run-cycle`'s\nEnd-of-Run Report makes when a run ends on `FRONTIER-EXHAUSTED:` with budget left: an empty backlog\nisn't this skill's gap to fill — step 3's synthesis reorders and extends what's already scoped, it\ndoes not discover unscoped work from nothing (rule 9) — it's a pointer to the skill that does. A\npointer only: name the recommendation, do not invoke it. Only say this when no startable work is left;\nan empty `## Waiting on you` with items still in `## Next` is step 4's good outcome, not this one.\n\n### 4. Settle what needs deciding — decide it, don't hand it back\n\nTwo sources feed this step: handoff's `## Waiting on you` entries, and the reorder \u002F prerequisite \u002F\nobjection proposals step 3 raised. A proposal from step 3 is decision-class the moment it changes what\nships or in what order — where a decision came from doesn't change how it's handled.\n\n**The default is that you decide.** Handing a decision up as a question gives the owner both the\nwork and the wait. A recommendation, once formed, *is* the decision: recorded `provisional` in step\n5 and presented for correction, not for an answer. What reaches the owner as something to *supply*\nis only what you genuinely cannot — a resource from outside, the owner's hand on a gated act, or a\ncriterion.\n\nFor each entry, classify and handle:\n\n- **Resource-blocked** (blocker · tried · unblocks) — re-run the self-unblock check (the same\n  \"Self-unblock before parking\" logic `run-cycle` applies): is the resource *actually* still missing?\n  Time has passed since it was parked. Still blocked → it stays in Waiting on you. Resolved → say so\n  and drop it.\n- **Decision-class** — brief it per\n  [decision-briefing.md §2](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fdecision-briefing.md): the decision in one\n  line, **at least two grounded options** (feasibility and cost from what you read in step 2 or the\n  grounding pass below — never invented), the **cross-lens read**\n  ([decision-lenses.md](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fdecision-lenses.md), only the lenses that\n  separate the leading options), and a **named recommendation** with what it locks in. **Then decide\n  it**: the recommendation is the provisional decision. Present it as \"결정: X — 트레이드오프 …,\n  이견 있으면 알려주세요\", never as \"A와 B 중 어느 쪽으로 할까요?\".\n  - **Decide ≠ act.** If carrying the decision out needs an act the central policy gates on a human —\n    `git push`, publish\u002Frelease, registering a GitHub issue, a major-version bump — or an act that is\n    irreversible outside this repo, the *decision* is still made and recorded; only the *act* goes to\n    Waiting on you, in the resource-blocked shape where the owner is the resource: **blocker** = the\n    gate · **tried** = decision settled (`→ Decided this session`) · **unblocks** = approval to run\n    `\u003Cthe act>`. Confidence never lifts the gate; the gate never un-makes the decision.\n  - The step-3 split (코드\u002F비코드) is how you spot such acts: a code decision reverses through git and\n    has none; a non-code decision is where push\u002Fpublish\u002Fregistration live.\n- **Criterion request** — reachable only *after* the grounding pass, when the decision still will not\n  settle because (a) the lenses conflict irreducibly, or (d) a criterion does exist but is ambiguous\n  here or, followed faithfully, lands somewhere common sense says is wrong (decision-lenses.md's\n  cause table). **Do not ask the owner to pick an option — ask for the criterion.** Name the criterion\n  and where it breaks, propose the revised criterion, and **decide under it** (provisional, as\n  above): \"판단기준 X가 이 지점에서 모호\u002F상식에 반함 → X′로 보완 제안, X′ 기준으로는 Y로 결정. X′를\n  확정해 주세요.\" The decision does not wait for the answer; only its gated act, if any, does.\n\n**Grounding pass — settle the unknown instead of reporting it.** When a recommendation does not form\non the first pass, close the gap before writing anything:\n\n- **Name the one unknown that separates the leading options.** Not everything unknown about the\n  decision — the single thing whose answer would move the pick. If nothing would, there is no gap and\n  the recommendation was already available.\n- **Settle it, bounded (~5 min per entry)**, per decision-lenses.md's cause table — (b) read the\n  file\u002Ftest\u002Fdependent or `WebSearch`\u002F`WebFetch` the convention; (c) derive the missing criterion\n  from CLAUDE.md\u002FREADME and\n  [philosophy-alignment-guide.md](${CLAUDE_SKILL_DIR}\u002F..\u002Fmindset\u002Freferences\u002Fphilosophy-alignment-guide.md) and recommend it with the option it\n  implies. Run the cheap command rather than naming it. Only (a) and (d) become criterion requests,\n  and only once (b) and (c) are ruled out.\n- **If the bounded search came back empty, say what you searched**, not only what is still unknown.\n\nThis does not widen the skill (rule 9). Grounding a decision that is *already on the table* is\njudgment over known work. Scoping work nobody has identified yet is still `\u002Fiyu:backlog-discover`'s\njob.\n\n**After this step, Waiting on you holds only resource-blocked entries (external, or an owner-gated\nact) and criterion requests.** A decision-class entry never remains there as an open question.\nNothing to decide → \"None\", and move straight to confirming In flight \u002F Next. Do not manufacture a\ndecision to fill the section.\n\n### 5. Record immediately, then take feedback\n\n**Record before presenting, and do not wait to record.** Every decision step 4 made goes into\n`\u003Croot>\u002FHANDOFF.md`'s `## Decided this session` (create the section if absent) in the shape\n`\u002Fiyu:handoff` uses — decision · trade-off · **to correct: \u003Cthe reverse action>** — marked\n`provisional`, with the lenses that carried it. In the same edit, rewrite `## Waiting on you` to what\nstep 4 left there: a flag that got decided moves out (its gated act, if any, stays as a\nresource-blocked entry pointing at the decision); criterion requests go in. Do this now, not at the\nnext `\u002Fiyu:handoff` — a session that ends first must not lose it.\n\nEvery entry left standing keeps its **Registered** date and **Blocks** line from `\u002Fiyu:handoff`\n(a new one gets today's date) and gains the two lines only this step can write:\n\n- **Default** — what stands while the owner has not answered: the provisional decision the entry\n  waits on, or, for a criterion request, the decision taken under the proposed criterion.\n- **Meanwhile** — what work proceeds without it, and what does not. **Never the gated act itself**:\n  silence does not run a push, a release, or an issue registration (rule 3), so \"on no reply: publish\"\n  is not a Meanwhile line.\n\nWhen the project keeps a decision ledger\n([continuity-docs.md §2](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md)), each decision goes\ninto it as well, in the ledger's own form, and `## Decided this session` carries its ID and the\nto-correct line rather than a second copy of the entry.\n\n**A reorder is applied to `## Next` itself, in the same edit** — self-decided or owner-confirmed.\nRewrite that section into the decided order (keeping the 코드\u002F비코드 split), and mark the\n`## Decided this session` entry as applied.\n[continuity-docs.md §2](${CLAUDE_SKILL_DIR}\u002F..\u002F_shared\u002Fcontinuity-docs.md) makes `## Next` the\nauthoritative statement of what comes next — what `\u002Fiyu:run-cycle` Preparation reads as its opening\nscope — so a decision recorded only in prose is one a later session can start without ever seeing.\n\n**`ROADMAP.md` is still not this skill's to write (rule 1).** `\u002Fiyu:handoff` reads\n`## Decided this session` as an inherited fact and carries a reorder into `ROADMAP.md`; an entry\nmarked applied means applied *to `## Next`* only.\n\n**Feedback is how the criteria get corrected — handle it, don't just transcribe it.** This applies\nwhen an answer arrives later in the same conversation, after step 6 has presented and stopped —\nresume does not wait for it.\n\n- **Owner confirms or says nothing** → the provisional decision stands. Silence does not block a\n  decision (it was already made); it never authorizes a gated act (rule 3).\n- **Owner picks differently from the recommendation** → apply the new pick (update the entry and\n  `## Next`), then ask one alignment question instead of silently moving on: **which criterion\n  weighed differently — should it be revised, or is this a one-off exception?** Name the lens or\n  rule that drove the recommendation so the owner is correcting something specific (\"근본을 세련보다\n  무겁게 봤는데, 이 프로젝트에선 반대로 둘까요?\").\n- **A criterion gets confirmed** (from that question, or an answered criterion request) → record it\n  in `## Decided this session` as a criterion entry, re-derive any pending decision under it (update\n  the entry if the pick changes), and insert a 비코드 `## Next` item to persist it where the\n  criterion lives — the project's `CLAUDE.md`\u002F`AGENTS.md`, or, for a plugin-level lens, a\n  note in this plugin's own backlog. This skill does not write those files itself (rule 1).\n- **One-off exception** → record it as such in the entry, so a later session does not read it as\n  precedent.\n\n### 6. Stop — hand off, don't execute\n\nOnce In flight \u002F Next is settled (including step 3's proposals), every decision is recorded, and\nWaiting on you holds only blockers and criterion requests, the session-opening work is done. **This\nskill does not proceed into implementation from here.** Present the settled scope as one **briefing\ncard** — always this order, an empty block is one line (\"None\"), the labels in the session's language:\n\n1. **State** (≤ 5 lines) — where things stand, staleness flags, what changed since the handoff.\n2. **Q** to confirm — criterion requests, plus any decision the project's own policy reserves for the\n   owner (step 4 resolves nothing else this way): each with the recommendation, the alternative and when\n   it wins, the one fact that would flip it, and what stands if there is no reply.\n3. **G** to approve — gated acts whose decision is settled: each names the act and **what a yes opens\n   — that act only**. **H** to provide — what only the owner can supply (the step, never the secret).\n4. **D** decided by me — this session's provisional decisions, each with its reverse action.\n5. **C** code \u002F **N** non-code — the settled `## Next` in order; each row `[auto]` (startable with no\n   answer) or `[after Gn]` \u002F `[after Hn]` (the one row it waits on).\n6. **Reply** — the grammar and one example built from this card's codes: \"adopt all\" confirms every Q\n   at its recommendation and every D and opens **no** G or H; \"approve G1 G3\" opens exactly those\n   acts; \"Q2 → alternative\" \u002F \"revert D3\" overrides (step 5 then asks which criterion differed); \", start\"\n   is the explicit instruction below; no reply leaves Q and D standing and runs no G.\n\nNumber rows per briefing (`Q1`, `G2`), with the durable id beside it when one exists (`HD-12`). Row\nshapes, guards and a worked example: [references\u002Fbriefing-format.md](${CLAUDE_SKILL_DIR}\u002Freferences\u002Fbriefing-format.md).\nThen stop. Starting the work is a separate, explicit act: the human instructs it directly (a reply ending\n\"start\" counts), or `\u002Fiyu:run-cycle` is invoked. A settled scope is not that instruction. If the owner\nanswers with feedback instead, handle it per step 5's feedback rules, then stop again.\n\n## Rules\n\n1. **`HANDOFF.md` is the only file this skill writes** — plus the project's decision ledger, when\n   its instructions define one, for the decisions step 4 made. It never migrates `ROADMAP.md` →\n   `HISTORY.md`, never touches `ROADMAP.md`, `CLAUDE.md`, or `AGENTS.md`, and never runs\n   `\u002Fiyu:handoff`'s hygiene pass. Inside `HANDOFF.md` it writes exactly three things (step 5):\n   `## Decided this session` entries, `## Next` (a decided reorder, and any criterion-persistence\n   item), and `## Waiting on you` (what step 4 left there). Phase-level order in `ROADMAP.md` stays\n   `\u002Fiyu:handoff`'s to apply.\n2. **Ground options and proposals in what was actually read or run.** A step-3 reorder or\n   prerequisite needs a stated reason grounded in step 2's read, not a hunch. An unknown that\n   *separates the leading options* is **settled by step 4's grounding pass**, not reported as\n   unknown. Only an unknown that survives that bounded pass stays \"unknown\", reported together with\n   what was searched.\n3. **Decisions don't wait; gated acts do.** A recommendation is decided and recorded `provisional`\n   without waiting for a reply. An act the central policy gates on a human (push, publish\u002Frelease,\n   GitHub issue registration, major-version bump) or that is irreversible outside the repo never runs\n   on silence — silence is not consent for an *act*.\n4. **The owner is asked for resources and criteria, never for a pick.** Waiting on you holds\n   resource-blocked entries (including owner-gated acts) and criterion requests only. \"A or B?\" is\n   the one shape a decision may not reach the owner in.\n5. **An override is a signal about the criteria.** When the owner picks against the\n   recommendation, ask whether the criterion behind it should be revised — do not just record the\n   pick.\n6. **\"None\" is a complete outcome.** Nothing flagged and no step-3 proposal → say so; it is the good\n   result, not a gap to fill.\n7. **Never starts the work itself.** Step 6 ends the skill at a settled scope — implementation begins\n   only on a separate explicit instruction or `\u002Fiyu:run-cycle`. Reading, running a cheap command, or\n   searching to ground a decision is not starting the work.\n8. **Model-invoked runs announce first.** Same as `\u002Fiyu:handoff` rule 7 — say in one line that a\n   resume is running and why, before touching any file.\n9. **Synthesis is judgment over known work, not discovery of new work.** Step 3 and step 4's\n   grounding pass operate on what is already on the table — the boundary is what the work is aimed\n   at, not whether reading or searching happens. Scoping work nobody has identified yet stays\n   `\u002Fiyu:backlog-discover`'s job; a roadmap with no startable phase gets that skill *named*, never invoked.\n",{"name":5,"description":7,"when_to_use":47,"allowed-tools":48,"argument-hint":49},"User-invoked with \"세션 시작\", \"다음 작업 확인\", \"핸드오프 확인하고 시작\", \"resume this session\". Model-invoked (announce before running, per rule 8) when a fresh session's first message states no explicit task and the continuity root already has a `HANDOFF.md` — the same self-invoke bar `\u002Fiyu:handoff` uses, since this skill proposes and decides but never implements. Not for mid-session re-checks (read `HANDOFF.md` directly for that), and not a substitute for `\u002Fiyu:run-cycle` — resume always stops short of execution; run-cycle is the separate, explicit step that starts it.","Read, Glob, Grep, Edit, TodoWrite, Bash, WebSearch, WebFetch","[focus or scope note]","main","40338be269bc724d7c245017c0f3e7ea198f7c8e","plugins\u002Fiyu\u002Fskills\u002Fresume","https:\u002F\u002Fgithub.com\u002Fiyulab\u002Fclaude-plugins","https:\u002F\u002Fgithub.com\u002Fiyulab","www.iyulab.com","2026-11-13 10:31:03",{"source":8,"skillName":5,"commands":58,"prompt":68,"zipAvailable":69},[59,62,65],{"tool":60,"command":61},"npx","npx skills add iyulab\u002Fclaude-plugins --skill resume",{"tool":63,"command":64},"bunx","bunx skills add iyulab\u002Fclaude-plugins --skill resume",{"tool":66,"command":67},"pnpm","pnpm dlx skills add iyulab\u002Fclaude-plugins --skill resume","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fiyulab\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fiyu\u002Fskills\u002Fresume\u002FSKILL.md",true,[71,77],{"path":72,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":75,"sourceHtmlUrl":76},"references\u002Fbriefing-format.md","REFERENCE","参考文档",7233,"https:\u002F\u002Fgithub.com\u002Fiyulab\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fiyu\u002Fskills\u002Fresume\u002Freferences\u002Fbriefing-format.md",{"path":78,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":79,"sourceHtmlUrl":80},"references\u002Frationale.md",4240,"https:\u002F\u002Fgithub.com\u002Fiyulab\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fiyu\u002Fskills\u002Fresume\u002Freferences\u002Frationale.md",[82,105,119,138,169,190],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":42},"01M6WKK4AZHX2ERR9RW97WX0BN","drive","Pushes the open stage of a product to its agreed satisfaction bar — implement one failing scenario at a time, have an independent evaluator judge the running build, repeat until the stage passes or needs something only the owner can supply, then show the owner a demo. Use after `\u002Fiyu:charter` has a signed stage contract.","https:\u002F\u002Fgithub.com\u002Fiyulab\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fiyu\u002Fskills\u002Fdrive\u002FSKILL.md",[],[84,89,90,91,92,93,94,95,96,37,97,98,36,27,99,100,101,102,103,104],"supply","running","show","time","failing","true","open","contract","use","stage","implement","can","task","demo","id","product",{"skillNo":106,"name":107,"nameZh":6,"description":108,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":109,"categories":110,"keywords":111,"publishedAt":42},"01M6WKK02K2DQYHRR9KKY8C522","charter","Interviews the owner to set a product's goal and writes the stage contract — an ordered ladder of stages that each work on their own, with scenarios an observer can watch, a calibrated satisfaction bar, the promises each stage must keep, and every owner decision that could block it. Use before `\u002Fiyu:drive`, and again at each stage boundary.","https:\u002F\u002Fgithub.com\u002Fiyulab\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fiyu\u002Fskills\u002Fcharter\u002FSKILL.md",[],[107,112,113,33,32,114,38,94,95,96,115,97,98,116,36,27,100,117,35,118],"SET","promises","work on","observer","keep","with","boundary",{"skillNo":120,"name":121,"nameZh":6,"description":122,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":123,"categories":124,"keywords":125,"publishedAt":42},"01M6WKJKPMSG3MPR0CZQZ9MVG1","accept","Judges the open stage by having an independent evaluator subagent drive the real, running build like a user — scenario by scenario, against the rubric — and record an evidence report and the status fields. Use to get a verdict without building anything; `\u002Fiyu:drive` dispatches the same evaluator each round.","https:\u002F\u002Fgithub.com\u002Fiyulab\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fiyu\u002Fskills\u002Faccept\u002FSKILL.md",[],[121,90,126,127,128,94,95,97,98,36,129,103,130,131,132,133,134,135,136,137],"evidence","smoke","get","subagent","record","verdict","build","read","report","rubric","grep","status",{"skillNo":139,"name":140,"nameZh":6,"description":141,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":142,"categories":143,"keywords":152,"publishedAt":42},"01M6C4KKDYZE639N931JNRYZEY","telemetry-az","Analyzes Azure Application Insights telemetry since the last run to surface defects, performance regressions, and feature-drop signals, tracked run-over-run as a quality trend (error-rate\u002Fp95\u002Fissue-count direction, with closed-issue recurrence detection), plus a report-only user-analytics reading (active-user growth, feature\u002Fpage preference shifts, engagement). Triages findings against project philosophy, files issues only for threshold-crossing ones, and leads every report with a both-purpose Trend section backed by a 12-run history.","https:\u002F\u002Fgithub.com\u002Fiyulab\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fiyu\u002Fskills\u002Ftelemetry-az\u002FSKILL.md",[144,149],{"categoryNo":145,"categoryName":146,"taxonomyType":147,"taxonomyTypeDesc":148},"01JBSEEDCAT000000000000013","分析与可视化","SCENARIO","使用场景",{"categoryNo":150,"categoryName":151,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000033","数据分析师",[153,154,155,156,157,37,36,158,159,160,161,162,163,164,165,28,166,27,167,168],"telemetry","telemetry az","performance","up","issues","analyze","error","app","direction","project","trending","insights","files","drop","dry","philosophy",{"skillNo":170,"name":171,"nameZh":6,"description":172,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":173,"categories":174,"keywords":175,"publishedAt":42},"01M6C4KGM2AA9JXQNHB7MW8JJ1","ship","Takes finished work out — bumps the version, commits, pushes, and watches the CI run to completion, reporting what failed if it fails. Checks first whether now is the right moment — if remaining backlog work would touch the same consumer-facing surface, it says so and asks rather than publishing a version the next few items obsolete.","https:\u002F\u002Fgithub.com\u002Fiyulab\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fiyu\u002Fskills\u002Fship\u002FSKILL.md",[],[171,176,177,178,179,180,37,97,181,36,182,183,184,35,185,186,187,133,188,189],"ready","push","commits","ci","publishing","version","patch","commit","publish","first","cicd","run","reporting","edit",{"skillNo":191,"name":192,"nameZh":6,"description":193,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":194,"categories":195,"keywords":196,"publishedAt":42},"01M6C4KD17SZCFRPYD3Q45XKAG","run-cycle","Executes adaptive iterative development cycles — each cycle is a self-contained plan\u002Fexecute\u002Fverify\u002Freflect loop that reshapes the roadmap and derives emergent follow-on scope from its own output","https:\u002F\u002Fgithub.com\u002Fiyulab\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fiyu\u002Fskills\u002Frun-cycle\u002FSKILL.md",[],[197,187,31,198,199,200,201,36,202,203,204,133,205,189,33,27,25,206,207,208],"run cycle","development","iterative development","loop","roadmap","reflect","plan","command","write","scripts","stop","adaptive"]