[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79DZ77YVJT2W4MS9855XGXD":3,"skill-same:01M79DZ77YVJT2W4MS9855XGXD":45},{"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":24,"commitSha":25,"skillDirectory":26,"forksCount":14,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":29,"updatedAt":30,"install":31,"resources":44},"01M79DZ77YVJT2W4MS9855XGXD","ship",null,"Take a feature or fix all the way from idea to released-and-verified, adapting to whatever project you're in. A SINGLE command that auto-detects the current stage (new idea \u002F scoping \u002F implementing \u002F ready-to-PR \u002F merged \u002F releasing) and continues, stopping only at five consequential gates. Bakes in the ship-it guardrails: test before merge, plain-English PRs\u002Fissues\u002Frelease-notes, schema changes via the project's tool (never hand-DDL), version+lockfile mirroring, deploy handled by CI (then verified), and prod touched only through process. Use when the user says \"ship\", \"let's ship this\", \"ship a feature\u002Ffix\", or wants the end-to-end lifecycle. NOT for a one-off edit with no lifecycle. Portable: it reads THIS project's conventions rather than assuming any one stack.","sharepointoscar\u002Fagent-workflow-skills","sharepointoscar","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F1544014?v=4","01M79DYNNSWKW6C81A3C9QF0FQ","Oscar Medina","01M79DYNPDXTKQ5RFX4071RBM8",0,"MIT","https:\u002F\u002Fgithub.com\u002Fsharepointoscar\u002Fagent-workflow-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fdevopsoscar\u002Fskills\u002Fship\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: ship\ndescription: >-\n  Take a feature or fix all the way from idea to released-and-verified, adapting to\n  whatever project you're in. A SINGLE command that auto-detects the current stage\n  (new idea \u002F scoping \u002F implementing \u002F ready-to-PR \u002F merged \u002F releasing) and continues,\n  stopping only at five consequential gates. Bakes in the ship-it guardrails: test\n  before merge, plain-English PRs\u002Fissues\u002Frelease-notes, schema changes via the project's\n  tool (never hand-DDL), version+lockfile mirroring, deploy handled by CI (then verified),\n  and prod touched only through process. Use when the user says \"ship\", \"let's ship this\",\n  \"ship a feature\u002Ffix\", or wants the end-to-end lifecycle. NOT for a one-off edit with no\n  lifecycle. Portable: it reads THIS project's conventions rather than assuming any one stack.\n---\n\n# Ship — end-to-end feature\u002Ffix lifecycle\n\nA single `\u002Fship` that auto-detects where the work is and continues. It is **project-agnostic**:\nit adapts to the repo you're in and never hardcodes another project's specifics.\n\n## STEP 0 — Adapt to THIS project (always do this first)\n\nRead the project's own conventions before acting; treat these as the source of truth:\n- **`CLAUDE.md`** (root + nested) — test commands, which env file is for tests vs real data,\n  deploy target\u002Fplaybook, version & release rules, git workflow, and any hard guardrails.\n- **The ops\u002Fdeploy runbook** (`docs\u002FDEPLOYMENT.md`, `docs\u002Fdeployment.md`, a `runbooks\u002F` dir…) — how to\n  reach production: SSH\u002Fexec into a container, read logs, redeploy, and **run a migration**. Read this\n  BEFORE improvising any prod procedure, not after the third failed attempt. If it is missing or wrong,\n  fixing it is part of the work.\n- **`package.json` scripts** — how this project tests \u002F type-checks \u002F builds \u002F migrates \u002F releases.\n- **CI config** (`.github\u002Fworkflows\u002F*` or equivalent) — what runs on push and on tag. **This is\n  the deployment mechanism.** Your job is to reach a clean merge and then *verify* what CI does —\n  not to hand-run deploy steps CI already owns.\n- **Existing issue\u002Fbranch\u002Ftag conventions** — mirror them.\n\nDeployment must be **(a) best-practice, (b) a known pattern, (c) appropriate to the project's type\nand structure.** When something isn't documented, ask rather than invent.\n\n### STEP 0b — Inventory the SHIPPABLE ARTIFACTS (a repo often ships more than one thing)\n\nBefore you touch anything, list what this repo actually ships and **how each one reaches users**.\nThey rarely reach users the same way:\n\n| Artifact kind | Typically reaches users by | Merging to the default branch… |\n|---|---|---|\n| Continuously-deployed service (server, web app, API) | CI deploys on push to the default branch | **ships it** |\n| Versioned distributable (desktop app, mobile app, CLI binary, library\u002Fpackage, container image) | a **version tag** → CI builds\u002Fsigns → release → **distribution channel** | **does NOT ship it** |\n\nDerive this from the project's own CI, not from assumptions: a pipeline triggered by a push to the\ndefault branch is the deploy path; a pipeline triggered by a **tag** is the release path. Note each\ndistributable's **distribution channel** — whatever a user's copy actually consults to get a new\nversion (an auto-update manifest, an app store, a package registry, a downloads page).\n\nIf a repo has only continuously-deployed artifacts, this step costs one glance and Stage 10 is a\nno-op. Record the answer — Stage 10 depends on it.\n\n## Auto-detect the stage, then continue\n\n- No definition yet → **Discover**.\n- A scoping doc exists but scope isn't locked → keep **iterating the doc**.\n- Scope locked, no issue\u002Fbranch → **Issue → branch**.\n- Branch with work in progress → **Implement (TDD)**.\n- Tests green + UAT done, no PR → **PR**.\n- PR open\u002Fapproved → **Merge → verify deploy**.\n- Merged, and the diff touched a **packaged client app** → **Release is REQUIRED** (its users have\n  nothing yet) → **Release → verify feed + downloads**.\n- Merged, release pending → **Release → verify**.\n\n## The FIVE hard stops (get explicit approval; run autonomously between them)\n\n1. **Scope-lock** — before creating the issue + branch.\n2. **Pre-PR** — before opening the PR (all gates green + user's manual UAT passed).\n3. **Pre-merge** — before merging.\n4. **Pre-release** — before tagging\u002Freleasing.\n5. **Prod-migration** — before ANY production schema\u002Fdata change.\n\n## Stages\n\n### 1 · Discover\nIterate conversationally to define the feature\u002Ffix. For every architecture\u002FUX decision, present\noptions and let the user choose — never decide alone.\n\n### 2 · Scoping doc\nWrite a temporary planning doc where the project keeps them (e.g. `docs\u002Fplans\u002F\u003CFEATURE>.md`).\nCapture: capabilities in **plain English**, constraints, files likely touched, test strategy, and\n**schema \u002F env-var impact**. Keep it in sync with reality as the approach changes during work.\n\n### 3 · Structured planning (features) vs light path (fixes)\n- **Substantial feature** → if the project or environment provides a structured planning workflow\n  (a phased-planning system, an RFC or design-doc process, a requirements template), use it, so the\n  work is documented as a unit of planned work rather than only as a diff.\n- **Small fix** → skip it. The scoping doc + issue + PR + commits ARE the documentation; the\n  release-notes mining in stage 11 recovers feature history from PRs\u002Fcommits either way.\n\n### 4 · Lock scope → issue  ⟨STOP 1: scope-lock⟩\nCreate the issue: **plain-English description + capabilities first**, THEN the specific\nchanges\u002Ffiles, acceptance criteria, and testing. Out-of-scope discoveries → **spin-off issues**,\nnever scope creep.\n\n### 5 · Feature branch\nBranch from the default branch (`feature\u002F\u003Cname>` or `fix\u002F\u003Cname>`). **NEVER work on the default branch.**\n\n### 6 · Implement (TDD)\nRed → green → refactor.\n- Tests run against the project's **TEST env, never the real-data\u002Fdev env.**\n- Exercise **real code paths \u002F the real API** — no mocks of the thing under test, no ad-hoc scripts\n  to fabricate data, no storage shortcuts. Create test subjects through the real API.\n- Schema changes go through the project's schema tool (e.g. `db:push`) — **NEVER hand-written DDL.**\n- Update docs during implementation, not after.\n\n### 7 · Green gate  ⟨STOP 2: pre-PR⟩\nEvery project gate green: typecheck + unit + e2e + build (+ version-mirror \u002F lockfile parity if the\nproject has them). THEN the user's **manual UAT passes** — code shipped ≠ validated. Only then proceed.\n\n### 8 · Push + 9 · PR  ⟨STOP 3: pre-merge⟩\n\n**Before you merge, answer two questions out loud:**\n1. **Does this diff change the schema definition?** If yes → a production migration is part of shipping\n   it (10a). Plan it now; do not discover it after the deploy 500s.\n2. **Are there un-run migrations already pending from an EARLIER merge?** If deploys have been failing,\n   un-run migrations pile up invisibly — and *your* merge, by fixing the pipeline, is what ships that\n   code onto the old schema and takes prod down. Check before you unblock a broken deploy.\n\n**Commit in logical, atomic commits grouped by concern** — e.g. core logic\u002FAPI, UI, tests,\ndocs+version — never one squashed blob. Each commit builds and tells one part of the story; the\ngrouping follows the change, not a fixed count. Use semantic prefixes (feat\u002Ffix\u002Ftest\u002Fdocs\u002F…).\nPush the branch. Open the PR in **plain English**:\n- **Fix** → briefly the previous problem, then the solution.\n- **Feature** → an overview of what it is + its capabilities.\nNo function\u002Fendpoint\u002Fcolumn names in the bullet headlines (technical detail goes in a footnote).\nThen a short changes list, runnable testing commands, and deploy\u002FDB\u002Fenv notes.\n\n**Carry the issue's plain-English SCOPE into the PR.** The PR is not just a changelog — a reviewer\nshould be able to read it alone and understand *what problem this solves and why this solution*,\nwithout opening the issue. So **in addition to** the usual \"New feature\" \u002F \"Fix\" sections, restate\n(don't just link) the issue's plain-English framing:\n- **the problem** — what's wrong today, in user terms;\n- **the solution** — what we're doing about it, and the reasoning behind the approach;\n- **any obstacle or constraint that shaped the design** — the non-obvious thing that made the\n  simple version not work;\n- **what is explicitly NOT changing** — often the fastest way to make a reviewer comfortable.\n\nReuse the issue's own wording where it's already good. A linked issue is a convenience, not a\nsubstitute — PRs get reviewed in isolation and read later as history.\n\n### 10 · Merge → CI deploys → verify  ⟨STOP: prod-migration if schema\u002Fdata changes⟩\nMerge (per project convention) → git cleanup (delete remote+local branch, sync + prune the default\nbranch). **CI takes over deployment — do not hand-run what CI owns.** Then **verify** the deploy\nlanded (new version live \u002F health OK).\n\n#### 10a · The migration gate — check it BEFORE you merge, not after\n\nMost pipelines deploy code but **do NOT run migrations**. So shipping code that reads a new column\nonto a database that doesn't have it takes production down the instant the deploy lands — the new\ncode is live, the schema is old, every endpoint touching it 500s.\n\n**Before merging, ask: does this diff change the schema definition?** (Whatever the project's source\nof truth is — an ORM schema file, a migrations dir, a model.) If yes, then the migration is **part of\nshipping it**, and you plan it now:\n\n- Confirm whether the deploy pipeline runs migrations. Usually it does not. Read the CI config.\n- Decide the order (migrate-then-deploy is the safe default for additive changes).\n- Treat it as ⟨STOP 5⟩ — the user approves before anything touches prod.\n\nThe subtle killer: **a broken or stalled deploy hides a missing migration.** The code never reached\nprod, so nothing 500s, and the pending migration looks fine. Then someone fixes the deploy — and *that*\nmerge is what takes prod down, blaming the wrong change. If deploys have been failing, **check for\nun-run migrations before you fix the pipeline.**\n\n#### 10b · Running the migration when the database is not reachable from a laptop\n\nProduction datastores are usually internal-only. Do **not** improvise a route to them. In order:\n\n1. **Read the project's ops runbook first** (STEP 0). If it documents the procedure, follow it. If it\n   documents a procedure that turns out to be *wrong*, fix the doc in the same pass — a stale runbook\n   is worse than none.\n2. **Use the project's sanctioned migration tool** (e.g. `db:push`, `migrate up`). **NEVER hand-author\n   DDL** and pipe it into a database client — that is not a migration, even when the statements happen\n   to be correct. It bypasses the source of truth and drifts every environment.\n3. **Get the tool to where the database is**, rather than the database to where you are. Options, in\n   order of preference:\n   - **Run the tool inside a deployed container\u002Fhost that is already on the private network.** This is\n     usually the right answer and is often already provisioned: the app container typically has the\n     schema, the migration tool in its dependencies, and the DB credentials in its environment.\n   - A tunnel\u002Fport-forward (`ssh -L`, `kubectl port-forward`, a bastion) — **only if the platform\n     actually supports it.** Verify, don't assume: some managed hosts refuse forwarding outright\n     (e.g. `open failed: unknown channel type`), and no amount of retrying fixes that.\n   - A platform-native one-off task\u002Fconsole, if the host offers one.\n4. **Preview the diff before applying it.** Tools that reconcile the whole schema can propose\n   *destructive* changes if prod has drifted. Run the dry\u002Fstrict mode, read the statements, and\n   **abort on any drop or type change** — that is drift, and it needs a decision, not a push.\n5. **Verify against the database afterwards**, then verify the endpoint that was broken actually\n   recovers.\n\nFinally: note that migrations belong **in the deploy pipeline**. Every time you run one by hand, say so\nand recommend folding it in — hand-run migrations are exactly how this failure recurs.\n\n**Merging ships only the artifacts CI deploys from the default branch (STEP 0b).** Now ask the\nquestion that decides whether you're actually done:\n\n> **Did this change touch a versioned distributable?**\n\nCheck the merged diff against each distributable's source tree — including **shared code it consumes**,\nwhich is the easy one to miss. If **yes**, the work is **NOT shipped**: its users are still running the\npreviously released version and will stay there until a tagged release goes out. **Stage 11 becomes\nmandatory, not optional.** Say that plainly instead of reporting \"merged ✅\" as though the feature had\nreached anyone.\n\nIf the change only touched continuously-deployed artifacts, verify the deploy and you're done.\n\n### 11 · Release  ⟨STOP 4: pre-release — approval to CUT the release, not to publish it⟩\nReleases are typically **batched, not per-merge** (signed builds \u002F notarization are expensive) — cut\none when the user is ready and has validated.\n- Bump version per project rules — **mirror ALL package files, including the lockfile,** in one commit;\n  tag == versions. Verify the CI install (`npm ci`, etc.) is satisfied first.\n- **Release notes in plain English** — what's new \u002F what's fixed, product-level, so a user (or the user\n  themself) understands it. **Mine merged PRs + commits since the last release** to assemble them; this\n  is also how feature history is recovered when no structured planning doc exists. Maintain the changelog.\n\n#### The tag is the trigger — CI owns the release artifact\nPush the **annotated** tag (pipelines often use its message as the release body). CI then builds,\nsigns, and **creates the release entry itself**, choosing the channel from the tag shape (stable vs\nprerelease).\n\n> ⛔ **NEVER hand-create a release entry that CI owns** — not via the forge's CLI, not by clicking\n> \"New Release\". Two things break at once:\n> 1. It **bypasses the project's publish gate** — pipelines commonly create a *draft* for a stable tag\n>    so a human can check the artifacts before any user gets them.\n> 2. A release that exists with **zero assets** can still become the public *\"latest\"* — so already-\n>    installed clients polling the distribution channel resolve a version with **no manifest\u002Fartifact**,\n>    and their update check breaks until the build finishes uploading.\n>\n> If you already created one: **revert it to a draft\u002Funpublished state immediately**, then confirm the\n> channel a real client reads still resolves the previous good version, intact.\n\n#### Then, in order\n1. **Watch the release pipeline to green** — build, sign\u002Fnotarize, and per-platform smoke. Signing and\n   notarization steps can legitimately take a long time; don't mistake slow for hung.\n2. **Inspect the produced artifacts** — for EVERY target platform: the installable artifact AND\n   whatever the distribution channel needs to serve it (manifest \u002F index \u002F checksums). A missing\n   manifest is a channel that cannot serve, even though the build was green.\n3. **Publish it.** Publishing, not building, is what releases it to users — so a green build that\n   sits as an unpublished draft has shipped NOTHING. Publish the release yourself as part of the ship\n   (e.g. `gh release edit vX.Y.Z --draft=false --latest`), then verify the channel below. Do **not**\n   hand the draft back and call the work done; do **not** list \"publish the release\" as a remaining\n   step for the user. The one thing you still never do is hand-CREATE a release entry that CI owns\n   (see the prohibition above) — publishing a draft CI already built is not that.\n   *(If a project genuinely requires a human publish gate, it will say so in its own docs — honour it\n   there, and say plainly that the release is built but unpublished. Absent that, publish.)*\n4. **Verify the channel, not just the build.** Resolve the new version the way a *client* does — hit\n   the public endpoint an installed copy would poll — and confirm it returns the new version with a\n   usable artifact per platform. Green CI proves it built; only the channel proves users can get it.\n5. **Verify every download path a user could take** — e.g. a downloads\u002Finstall page must offer the new\n   version for every OS\u002Farch it advertises. Check each one, not just the platform you happen to be on.\n\n## What \"shipped\" means (state this before claiming done)\n\nA change is shipped when **the people it affects can actually get it** — which differs per artifact:\n\n- **Continuously-deployed artifact** → merged + the deploy verified live.\n- **Versioned distributable** → merged **AND** released **AND** the distribution channel resolves the\n  new version **AND** every advertised download path offers it.\n\n\"Merged\" is a *milestone*, not the finish line. Reporting a distributable's feature as done at merge\ntime tells the user their users have it when they demonstrably don't.\n\n## Universal guardrails (enforced every run)\n\n- Never work on the default branch; branch first.\n- **Merged ≠ shipped for a versioned distributable.** If the diff touched one (or shared code it\n  consumes), a release is REQUIRED — never close out at merge. The merge and the release are ONE\n  task: the moment such a PR merges, the NEXT action is the version bump + tag, not a status report.\n  Catching yourself writing \"a release is still needed\" means you have already broken this.\n- **A built draft is not a shipped release.** Publish it and verify the distribution channel serves\n  the new version. Reporting \"the draft is ready, publishing is yours\" is stopping one step short.\n- **Never hand-create a release entry that CI owns**, and never leave a published release with no\n  artifacts — it can hijack \"latest\" and break existing clients' update checks.\n- Tests in the project's TEST env only; real API\u002Fcode paths, no mocks\u002Fshortcuts\u002Fad-hoc data scripts.\n- Schema via the project's tool; **NEVER hand-author DDL — least of all on prod.** Preview the diff and\n  **abort on any drop or type change** (that's drift — it needs a decision, not a push).\n- **A schema change is not shipped until the migration has run on prod.** The pipeline almost certainly\n  won't run it for you. And a failing deploy HIDES a pending migration — so before you fix a broken\n  pipeline, check what un-run migrations that fix is about to unleash.\n- **Read the ops runbook before improvising any prod procedure.** Don't assume a tunnel\u002Fport-forward is\n  possible — verify. Prefer running the tool inside a container already on the private network.\n- Version bump = mirror ALL package files (incl. lockfile) in the same commit; confirm CI's install step\n  is satisfied before merge.\n- PRs \u002F issues \u002F release notes in **plain English**; no code identifiers in the headlines.\n- Nothing merges or releases until tests are green AND the user's manual validation passes.\n- New env vars → enumerate them for the user to add via their platform's UI (envs are often\n  full-replacement).\n- **Never touch production outside this process.**\n- Present options for architecture\u002FUX; the user decides.\n\n## Deploy\u002Fverify happens through CI, not by hand\nThe skill's responsibility ends at \"merge cleanly + verify.\" It does not manually run deploy commands\nthat CI owns. Where the project's CI does not yet cover a needed step (e.g. prod migrations), handle it\nat the matching hard-stop with the sanctioned tool, and recommend folding it into CI as the durable pattern.\n","# Ship — end-to-end feature\u002Ffix lifecycle\n\nA single `\u002Fship` that auto-detects where the work is and continues. It is **project-agnostic**:\nit adapts to the repo you're in and never hardcodes another project's specifics.\n\n## STEP 0 — Adapt to THIS project (always do this first)\n\nRead the project's own conventions before acting; treat these as the source of truth:\n- **`CLAUDE.md`** (root + nested) — test commands, which env file is for tests vs real data,\n  deploy target\u002Fplaybook, version & release rules, git workflow, and any hard guardrails.\n- **The ops\u002Fdeploy runbook** (`docs\u002FDEPLOYMENT.md`, `docs\u002Fdeployment.md`, a `runbooks\u002F` dir…) — how to\n  reach production: SSH\u002Fexec into a container, read logs, redeploy, and **run a migration**. Read this\n  BEFORE improvising any prod procedure, not after the third failed attempt. If it is missing or wrong,\n  fixing it is part of the work.\n- **`package.json` scripts** — how this project tests \u002F type-checks \u002F builds \u002F migrates \u002F releases.\n- **CI config** (`.github\u002Fworkflows\u002F*` or equivalent) — what runs on push and on tag. **This is\n  the deployment mechanism.** Your job is to reach a clean merge and then *verify* what CI does —\n  not to hand-run deploy steps CI already owns.\n- **Existing issue\u002Fbranch\u002Ftag conventions** — mirror them.\n\nDeployment must be **(a) best-practice, (b) a known pattern, (c) appropriate to the project's type\nand structure.** When something isn't documented, ask rather than invent.\n\n### STEP 0b — Inventory the SHIPPABLE ARTIFACTS (a repo often ships more than one thing)\n\nBefore you touch anything, list what this repo actually ships and **how each one reaches users**.\nThey rarely reach users the same way:\n\n| Artifact kind | Typically reaches users by | Merging to the default branch… |\n|---|---|---|\n| Continuously-deployed service (server, web app, API) | CI deploys on push to the default branch | **ships it** |\n| Versioned distributable (desktop app, mobile app, CLI binary, library\u002Fpackage, container image) | a **version tag** → CI builds\u002Fsigns → release → **distribution channel** | **does NOT ship it** |\n\nDerive this from the project's own CI, not from assumptions: a pipeline triggered by a push to the\ndefault branch is the deploy path; a pipeline triggered by a **tag** is the release path. Note each\ndistributable's **distribution channel** — whatever a user's copy actually consults to get a new\nversion (an auto-update manifest, an app store, a package registry, a downloads page).\n\nIf a repo has only continuously-deployed artifacts, this step costs one glance and Stage 10 is a\nno-op. Record the answer — Stage 10 depends on it.\n\n## Auto-detect the stage, then continue\n\n- No definition yet → **Discover**.\n- A scoping doc exists but scope isn't locked → keep **iterating the doc**.\n- Scope locked, no issue\u002Fbranch → **Issue → branch**.\n- Branch with work in progress → **Implement (TDD)**.\n- Tests green + UAT done, no PR → **PR**.\n- PR open\u002Fapproved → **Merge → verify deploy**.\n- Merged, and the diff touched a **packaged client app** → **Release is REQUIRED** (its users have\n  nothing yet) → **Release → verify feed + downloads**.\n- Merged, release pending → **Release → verify**.\n\n## The FIVE hard stops (get explicit approval; run autonomously between them)\n\n1. **Scope-lock** — before creating the issue + branch.\n2. **Pre-PR** — before opening the PR (all gates green + user's manual UAT passed).\n3. **Pre-merge** — before merging.\n4. **Pre-release** — before tagging\u002Freleasing.\n5. **Prod-migration** — before ANY production schema\u002Fdata change.\n\n## Stages\n\n### 1 · Discover\nIterate conversationally to define the feature\u002Ffix. For every architecture\u002FUX decision, present\noptions and let the user choose — never decide alone.\n\n### 2 · Scoping doc\nWrite a temporary planning doc where the project keeps them (e.g. `docs\u002Fplans\u002F\u003CFEATURE>.md`).\nCapture: capabilities in **plain English**, constraints, files likely touched, test strategy, and\n**schema \u002F env-var impact**. Keep it in sync with reality as the approach changes during work.\n\n### 3 · Structured planning (features) vs light path (fixes)\n- **Substantial feature** → if the project or environment provides a structured planning workflow\n  (a phased-planning system, an RFC or design-doc process, a requirements template), use it, so the\n  work is documented as a unit of planned work rather than only as a diff.\n- **Small fix** → skip it. The scoping doc + issue + PR + commits ARE the documentation; the\n  release-notes mining in stage 11 recovers feature history from PRs\u002Fcommits either way.\n\n### 4 · Lock scope → issue  ⟨STOP 1: scope-lock⟩\nCreate the issue: **plain-English description + capabilities first**, THEN the specific\nchanges\u002Ffiles, acceptance criteria, and testing. Out-of-scope discoveries → **spin-off issues**,\nnever scope creep.\n\n### 5 · Feature branch\nBranch from the default branch (`feature\u002F\u003Cname>` or `fix\u002F\u003Cname>`). **NEVER work on the default branch.**\n\n### 6 · Implement (TDD)\nRed → green → refactor.\n- Tests run against the project's **TEST env, never the real-data\u002Fdev env.**\n- Exercise **real code paths \u002F the real API** — no mocks of the thing under test, no ad-hoc scripts\n  to fabricate data, no storage shortcuts. Create test subjects through the real API.\n- Schema changes go through the project's schema tool (e.g. `db:push`) — **NEVER hand-written DDL.**\n- Update docs during implementation, not after.\n\n### 7 · Green gate  ⟨STOP 2: pre-PR⟩\nEvery project gate green: typecheck + unit + e2e + build (+ version-mirror \u002F lockfile parity if the\nproject has them). THEN the user's **manual UAT passes** — code shipped ≠ validated. Only then proceed.\n\n### 8 · Push + 9 · PR  ⟨STOP 3: pre-merge⟩\n\n**Before you merge, answer two questions out loud:**\n1. **Does this diff change the schema definition?** If yes → a production migration is part of shipping\n   it (10a). Plan it now; do not discover it after the deploy 500s.\n2. **Are there un-run migrations already pending from an EARLIER merge?** If deploys have been failing,\n   un-run migrations pile up invisibly — and *your* merge, by fixing the pipeline, is what ships that\n   code onto the old schema and takes prod down. Check before you unblock a broken deploy.\n\n**Commit in logical, atomic commits grouped by concern** — e.g. core logic\u002FAPI, UI, tests,\ndocs+version — never one squashed blob. Each commit builds and tells one part of the story; the\ngrouping follows the change, not a fixed count. Use semantic prefixes (feat\u002Ffix\u002Ftest\u002Fdocs\u002F…).\nPush the branch. Open the PR in **plain English**:\n- **Fix** → briefly the previous problem, then the solution.\n- **Feature** → an overview of what it is + its capabilities.\nNo function\u002Fendpoint\u002Fcolumn names in the bullet headlines (technical detail goes in a footnote).\nThen a short changes list, runnable testing commands, and deploy\u002FDB\u002Fenv notes.\n\n**Carry the issue's plain-English SCOPE into the PR.** The PR is not just a changelog — a reviewer\nshould be able to read it alone and understand *what problem this solves and why this solution*,\nwithout opening the issue. So **in addition to** the usual \"New feature\" \u002F \"Fix\" sections, restate\n(don't just link) the issue's plain-English framing:\n- **the problem** — what's wrong today, in user terms;\n- **the solution** — what we're doing about it, and the reasoning behind the approach;\n- **any obstacle or constraint that shaped the design** — the non-obvious thing that made the\n  simple version not work;\n- **what is explicitly NOT changing** — often the fastest way to make a reviewer comfortable.\n\nReuse the issue's own wording where it's already good. A linked issue is a convenience, not a\nsubstitute — PRs get reviewed in isolation and read later as history.\n\n### 10 · Merge → CI deploys → verify  ⟨STOP: prod-migration if schema\u002Fdata changes⟩\nMerge (per project convention) → git cleanup (delete remote+local branch, sync + prune the default\nbranch). **CI takes over deployment — do not hand-run what CI owns.** Then **verify** the deploy\nlanded (new version live \u002F health OK).\n\n#### 10a · The migration gate — check it BEFORE you merge, not after\n\nMost pipelines deploy code but **do NOT run migrations**. So shipping code that reads a new column\nonto a database that doesn't have it takes production down the instant the deploy lands — the new\ncode is live, the schema is old, every endpoint touching it 500s.\n\n**Before merging, ask: does this diff change the schema definition?** (Whatever the project's source\nof truth is — an ORM schema file, a migrations dir, a model.) If yes, then the migration is **part of\nshipping it**, and you plan it now:\n\n- Confirm whether the deploy pipeline runs migrations. Usually it does not. Read the CI config.\n- Decide the order (migrate-then-deploy is the safe default for additive changes).\n- Treat it as ⟨STOP 5⟩ — the user approves before anything touches prod.\n\nThe subtle killer: **a broken or stalled deploy hides a missing migration.** The code never reached\nprod, so nothing 500s, and the pending migration looks fine. Then someone fixes the deploy — and *that*\nmerge is what takes prod down, blaming the wrong change. If deploys have been failing, **check for\nun-run migrations before you fix the pipeline.**\n\n#### 10b · Running the migration when the database is not reachable from a laptop\n\nProduction datastores are usually internal-only. Do **not** improvise a route to them. In order:\n\n1. **Read the project's ops runbook first** (STEP 0). If it documents the procedure, follow it. If it\n   documents a procedure that turns out to be *wrong*, fix the doc in the same pass — a stale runbook\n   is worse than none.\n2. **Use the project's sanctioned migration tool** (e.g. `db:push`, `migrate up`). **NEVER hand-author\n   DDL** and pipe it into a database client — that is not a migration, even when the statements happen\n   to be correct. It bypasses the source of truth and drifts every environment.\n3. **Get the tool to where the database is**, rather than the database to where you are. Options, in\n   order of preference:\n   - **Run the tool inside a deployed container\u002Fhost that is already on the private network.** This is\n     usually the right answer and is often already provisioned: the app container typically has the\n     schema, the migration tool in its dependencies, and the DB credentials in its environment.\n   - A tunnel\u002Fport-forward (`ssh -L`, `kubectl port-forward`, a bastion) — **only if the platform\n     actually supports it.** Verify, don't assume: some managed hosts refuse forwarding outright\n     (e.g. `open failed: unknown channel type`), and no amount of retrying fixes that.\n   - A platform-native one-off task\u002Fconsole, if the host offers one.\n4. **Preview the diff before applying it.** Tools that reconcile the whole schema can propose\n   *destructive* changes if prod has drifted. Run the dry\u002Fstrict mode, read the statements, and\n   **abort on any drop or type change** — that is drift, and it needs a decision, not a push.\n5. **Verify against the database afterwards**, then verify the endpoint that was broken actually\n   recovers.\n\nFinally: note that migrations belong **in the deploy pipeline**. Every time you run one by hand, say so\nand recommend folding it in — hand-run migrations are exactly how this failure recurs.\n\n**Merging ships only the artifacts CI deploys from the default branch (STEP 0b).** Now ask the\nquestion that decides whether you're actually done:\n\n> **Did this change touch a versioned distributable?**\n\nCheck the merged diff against each distributable's source tree — including **shared code it consumes**,\nwhich is the easy one to miss. If **yes**, the work is **NOT shipped**: its users are still running the\npreviously released version and will stay there until a tagged release goes out. **Stage 11 becomes\nmandatory, not optional.** Say that plainly instead of reporting \"merged ✅\" as though the feature had\nreached anyone.\n\nIf the change only touched continuously-deployed artifacts, verify the deploy and you're done.\n\n### 11 · Release  ⟨STOP 4: pre-release — approval to CUT the release, not to publish it⟩\nReleases are typically **batched, not per-merge** (signed builds \u002F notarization are expensive) — cut\none when the user is ready and has validated.\n- Bump version per project rules — **mirror ALL package files, including the lockfile,** in one commit;\n  tag == versions. Verify the CI install (`npm ci`, etc.) is satisfied first.\n- **Release notes in plain English** — what's new \u002F what's fixed, product-level, so a user (or the user\n  themself) understands it. **Mine merged PRs + commits since the last release** to assemble them; this\n  is also how feature history is recovered when no structured planning doc exists. Maintain the changelog.\n\n#### The tag is the trigger — CI owns the release artifact\nPush the **annotated** tag (pipelines often use its message as the release body). CI then builds,\nsigns, and **creates the release entry itself**, choosing the channel from the tag shape (stable vs\nprerelease).\n\n> ⛔ **NEVER hand-create a release entry that CI owns** — not via the forge's CLI, not by clicking\n> \"New Release\". Two things break at once:\n> 1. It **bypasses the project's publish gate** — pipelines commonly create a *draft* for a stable tag\n>    so a human can check the artifacts before any user gets them.\n> 2. A release that exists with **zero assets** can still become the public *\"latest\"* — so already-\n>    installed clients polling the distribution channel resolve a version with **no manifest\u002Fartifact**,\n>    and their update check breaks until the build finishes uploading.\n>\n> If you already created one: **revert it to a draft\u002Funpublished state immediately**, then confirm the\n> channel a real client reads still resolves the previous good version, intact.\n\n#### Then, in order\n1. **Watch the release pipeline to green** — build, sign\u002Fnotarize, and per-platform smoke. Signing and\n   notarization steps can legitimately take a long time; don't mistake slow for hung.\n2. **Inspect the produced artifacts** — for EVERY target platform: the installable artifact AND\n   whatever the distribution channel needs to serve it (manifest \u002F index \u002F checksums). A missing\n   manifest is a channel that cannot serve, even though the build was green.\n3. **Publish it.** Publishing, not building, is what releases it to users — so a green build that\n   sits as an unpublished draft has shipped NOTHING. Publish the release yourself as part of the ship\n   (e.g. `gh release edit vX.Y.Z --draft=false --latest`), then verify the channel below. Do **not**\n   hand the draft back and call the work done; do **not** list \"publish the release\" as a remaining\n   step for the user. The one thing you still never do is hand-CREATE a release entry that CI owns\n   (see the prohibition above) — publishing a draft CI already built is not that.\n   *(If a project genuinely requires a human publish gate, it will say so in its own docs — honour it\n   there, and say plainly that the release is built but unpublished. Absent that, publish.)*\n4. **Verify the channel, not just the build.** Resolve the new version the way a *client* does — hit\n   the public endpoint an installed copy would poll — and confirm it returns the new version with a\n   usable artifact per platform. Green CI proves it built; only the channel proves users can get it.\n5. **Verify every download path a user could take** — e.g. a downloads\u002Finstall page must offer the new\n   version for every OS\u002Farch it advertises. Check each one, not just the platform you happen to be on.\n\n## What \"shipped\" means (state this before claiming done)\n\nA change is shipped when **the people it affects can actually get it** — which differs per artifact:\n\n- **Continuously-deployed artifact** → merged + the deploy verified live.\n- **Versioned distributable** → merged **AND** released **AND** the distribution channel resolves the\n  new version **AND** every advertised download path offers it.\n\n\"Merged\" is a *milestone*, not the finish line. Reporting a distributable's feature as done at merge\ntime tells the user their users have it when they demonstrably don't.\n\n## Universal guardrails (enforced every run)\n\n- Never work on the default branch; branch first.\n- **Merged ≠ shipped for a versioned distributable.** If the diff touched one (or shared code it\n  consumes), a release is REQUIRED — never close out at merge. The merge and the release are ONE\n  task: the moment such a PR merges, the NEXT action is the version bump + tag, not a status report.\n  Catching yourself writing \"a release is still needed\" means you have already broken this.\n- **A built draft is not a shipped release.** Publish it and verify the distribution channel serves\n  the new version. Reporting \"the draft is ready, publishing is yours\" is stopping one step short.\n- **Never hand-create a release entry that CI owns**, and never leave a published release with no\n  artifacts — it can hijack \"latest\" and break existing clients' update checks.\n- Tests in the project's TEST env only; real API\u002Fcode paths, no mocks\u002Fshortcuts\u002Fad-hoc data scripts.\n- Schema via the project's tool; **NEVER hand-author DDL — least of all on prod.** Preview the diff and\n  **abort on any drop or type change** (that's drift — it needs a decision, not a push).\n- **A schema change is not shipped until the migration has run on prod.** The pipeline almost certainly\n  won't run it for you. And a failing deploy HIDES a pending migration — so before you fix a broken\n  pipeline, check what un-run migrations that fix is about to unleash.\n- **Read the ops runbook before improvising any prod procedure.** Don't assume a tunnel\u002Fport-forward is\n  possible — verify. Prefer running the tool inside a container already on the private network.\n- Version bump = mirror ALL package files (incl. lockfile) in the same commit; confirm CI's install step\n  is satisfied before merge.\n- PRs \u002F issues \u002F release notes in **plain English**; no code identifiers in the headlines.\n- Nothing merges or releases until tests are green AND the user's manual validation passes.\n- New env vars → enumerate them for the user to add via their platform's UI (envs are often\n  full-replacement).\n- **Never touch production outside this process.**\n- Present options for architecture\u002FUX; the user decides.\n\n## Deploy\u002Fverify happens through CI, not by hand\nThe skill's responsibility ends at \"merge cleanly + verify.\" It does not manually run deploy commands\nthat CI owns. Where the project's CI does not yet cover a needed step (e.g. prod migrations), handle it\nat the matching hard-stop with the sanctioned tool, and recommend folding it into CI as the durable pattern.\n",{"name":5,"description":7},"main","f27dd0e6c87efa56cfdea304aa3e89e63c02711b","plugins\u002Fdevopsoscar\u002Fskills\u002Fship","https:\u002F\u002Fgithub.com\u002Fsharepointoscar\u002Fagent-workflow-skills","https:\u002F\u002Fgithub.com\u002Fsharepointoscar","Author | DevOps Practitioner | Docker | Terraform | Kubernetes | Jenkins X  | AWS | GCP | Azure and SP Veteran (since code-name Tahoe)","2026-11-11 12:44:52",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add sharepointoscar\u002Fagent-workflow-skills --skill ship",{"tool":37,"command":38},"bunx","bunx skills add sharepointoscar\u002Fagent-workflow-skills --skill ship",{"tool":40,"command":41},"pnpm","pnpm dlx skills add sharepointoscar\u002Fagent-workflow-skills --skill ship","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fsharepointoscar\u002Fagent-workflow-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fdevopsoscar\u002Fskills\u002Fship\u002FSKILL.md",true,[],[46,53,60,67,74,81],{"skillNo":47,"name":48,"nameZh":6,"description":49,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":50,"categories":51,"keywords":52,"publishedAt":19},"01M79FQNYDNR7TYQAM5061GVKR","web-performance-optimization","Optimize website and web application performance including loading speed, Core Web Vitals, bundle size, caching strategies, and runtime performance","https:\u002F\u002Fgithub.com\u002Fsharepointoscar\u002Fagent-workflow-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fskills-web-frontend\u002Fskills\u002Fweb-performance-optimization\u002FSKILL.md",[],[],{"skillNo":54,"name":55,"nameZh":6,"description":56,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":57,"categories":58,"keywords":59,"publishedAt":19},"01M79FQK396EEGX40QA7B9S404","web-design-guidelines","Review UI code for Web Interface Guidelines compliance. Use when asked to \"review my UI\", \"check accessibility\", \"audit design\", \"review UX\", or \"check my site against best practices\".","https:\u002F\u002Fgithub.com\u002Fsharepointoscar\u002Fagent-workflow-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fskills-web-frontend\u002Fskills\u002Fweb-design-guidelines\u002FSKILL.md",[],[],{"skillNo":61,"name":62,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":19},"01M79FQFK4GJCG8HVRG8JQF4QH","wcag-audit-patterns","Conduct WCAG 2.2 accessibility audits with automated testing, manual verification, and remediation guidance. Use when auditing websites for accessibility, fixing WCAG violations, or implementing accessible design patterns.","https:\u002F\u002Fgithub.com\u002Fsharepointoscar\u002Fagent-workflow-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fskills-web-frontend\u002Fskills\u002Fwcag-audit-patterns\u002FSKILL.md",[],[],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":71,"categories":72,"keywords":73,"publishedAt":19},"01M79FQ7WRQXC4D20Q2HTBMDG7","ui-ux-pro-max","UI\u002FUX design intelligence. 50 styles, 21 palettes, 50 font pairings, 20 charts, 9 stacks (React, Next.js, Vue, Svelte, SwiftUI, React Native, Flutter, Tailwind, shadcn\u002Fui). Actions: plan, build, create, design, implement, review, fix, improve, optimize, enhance, refactor, check UI\u002FUX code. Projects: website, landing page, dashboard, admin panel, e-commerce, SaaS, portfolio, blog, mobile app, .html, .tsx, .vue, .svelte. Elements: button, modal, navbar, sidebar, card, table, form, chart. Styles: glassmorphism, claymorphism, minimalism, brutalism, neumorphism, bento grid, dark mode, responsive, skeuomorphism, flat design. Topics: color palette, accessibility, animation, layout, typography, font pairing, spacing, hover, shadow, gradient. Integrations: shadcn\u002Fui MCP for component search and examples.","https:\u002F\u002Fgithub.com\u002Fsharepointoscar\u002Fagent-workflow-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fskills-web-frontend\u002Fskills\u002Fui-ux-pro-max\u002FSKILL.md",[],[],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":19},"01M79FPZC11BPBWDCVK2TER0RC","stitch-ui-design","Expert guide for creating effective prompts for Google Stitch AI UI design tool. Use when user wants to design UI\u002FUX in Stitch, create app interfaces, generate mobile\u002Fweb designs, or needs help crafting Stitch prompts. Covers prompt structure, specificity techniques, iteration strategies, and design-to-code workflows for Stitch by Google.","https:\u002F\u002Fgithub.com\u002Fsharepointoscar\u002Fagent-workflow-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fskills-web-frontend\u002Fskills\u002Fstitch-ui-design\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M79FPTMMM3C57JN25RTRX2HR","scroll-experience","Expert in building immersive scroll-driven experiences - parallax storytelling, scroll animations, interactive narratives, and cinematic web experiences. Like NY Times interactives, Apple product pages, and award-winning web experiences. Makes websites feel like experiences, not just pages. Use when: scroll animation, parallax, scroll storytelling, interactive story, cinematic website.","https:\u002F\u002Fgithub.com\u002Fsharepointoscar\u002Fagent-workflow-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fskills-web-frontend\u002Fskills\u002Fscroll-experience\u002FSKILL.md",[],[]]