tangem-app-android-audited/.claude/skills/create-jira-task/SKILL.md
2026-07-06 13:08:06 +03:00

16 KiB
Raw Permalink Blame History


name: create-jira-task description: Create a Task in the Tangem Android Jira project (AND) via the Atlassian MCP. Pre-fills assignee (self), current active sprint, parent (Story or Epic), QA Notes, and other fields, asks the user for anything missing, shows a full preview, and creates the issue ONLY after explicit confirmation. Use when the user asks to "create a task", "создай задачу / таску в Jira", "заведи задачу", "open a Jira task". allowed-tools: Read, Bash, mcp__claude_ai_Atlassian_Rovo__atlassianUserInfo, mcp__claude_ai_Atlassian_Rovo__getAccessibleAtlassianResources, mcp__claude_ai_Atlassian_Rovo__searchJiraIssuesUsingJql, mcp__claude_ai_Atlassian_Rovo__getJiraIssue, mcp__claude_ai_Atlassian_Rovo__createJiraIssue, mcp__claude_ai_Atlassian_Rovo__createIssueLink, AskUserQuestion argument-hint: [summary text] [parent AND-xxxxx] [--dry-run]

Create a Task in the Tangem Android Jira project.

This skill is interactive — it runs locally for a developer, not on CI. Ask the user for any missing data. Never create the issue without an explicit confirmation step (Phase 4).

Dry-run mode: if $ARGUMENTS contains --dry-run, do all the work (preflight, gather inputs, resolve sprint, validate parent, build the preview and final payload) but make no changes in Jira — skip the confirmation gate and the createJiraIssue call. See Phase 4D.

Constants

Key Value
cloudId tangem.atlassian.net
Project key AND (quote as "AND" inside JQL — it collides with the AND keyword)
Issue type Task (localized name Задача, id 10002)
Issue browse URL https://tangem.atlassian.net/browse/{KEY}

Field map (use these exact field IDs)

Field How to set Notes
Summary summary (top-level param) required
Description description (top-level param), contentFormat: "markdown" optional but recommended
Issue type issueTypeName: "Task" fallback localized "Задача" if rejected
Assignee assignee_account_id defaults to the current user (self)
Sprint additional_fields: { "customfield_10021": <sprintId> } numeric id of the active sprint
Stream additional_fields: { "customfield_11931": { "id": "<optionId>" } } optional single-select (see option ids below)
Parent hierarchy parent accepts an Epic only Story/Task/Bug are all the same hierarchy level, so a Story can NOT be the parent of a Task (Jira rejects it: "parent does not belong to the hierarchy"). Epic parent → set parent: "<epic>". Story parent → set the hierarchy parent to the Story's own parent Epic (inherit it, so the Task lands in the same Epic) and create the Phase 5b "implements" link to the Story. If the Story has no parent Epic, omit parent and rely on the link only.
QA Notes additional_fields: { "customfield_11232": <ADF doc> } ADF only — a plain string is rejected ("must be an Atlassian document"). Wrap the text: {"type":"doc","version":1,"content":[{"type":"paragraph","content":[{"type":"text","text":"<text>"}]}]}
Story Points additional_fields: { "customfield_10025": <number> } optional
Developer additional_fields: { "customfield_11898": { "accountId": "<id>" } } optional
Labels additional_fields: { "labels": ["..."] } optional
Components additional_fields: { "components": [{ "name": "..." }] } optional

Stream option ids (single-select, optional)

If the user picks a Stream, map the chosen value to its option id (preferred); on a { "id": ... } rejection retry that field with { "value": "<value>" }.

Value id
Core 15117
Grow 15116
Visa 15981

Tool names: the phases below reference MCP tools by short name (e.g. createJiraIssue, getAccessibleAtlassianResources) for readability. These map to the fully-qualified Atlassian Rovo tools declared in allowed-tools (mcp__claude_ai_Atlassian_Rovo__*) — the connected server for this skill. Invoke them by their fully-qualified names.

Phase 0 — Preflight

  1. Verify the Atlassian MCP is reachable: call getAccessibleAtlassianResources (no params). If it fails or tangem.atlassian.net is absent, STOP with: FATAL: Atlassian MCP is not connected. Run 'claude mcp list' to check server status.
  2. Determine the current user (the default assignee / self):
    • If the current user's Jira accountId is already known from memory/prior context, reuse it and skip the API call — it is stable and does not change.
    • Otherwise call atlassianUserInfo, save account_id and name, and remember it for next time (so future runs skip this call).

Phase 1 — Gather inputs

Parse $ARGUMENTS for an obvious summary and/or a parent key (AND-\d+). Then collect the rest. Ask the user only for what is still missing, grouped into as few questions as possible (use AskUserQuestion where the choice is constrained, plain text otherwise):

  • Summary (required) — short task title. Must be in English (mandatory). If it was not provided in $ARGUMENTS, ask the user (via AskUserQuestion): "Generate the summary from your local changes?" with options:
    • Generate from local changes — inspect the working tree and derive a concise English title from it: run git status --porcelain, git diff --stat HEAD, git branch --show-current (and git log --oneline -5 for context). Propose a one-line summary and show it to the user for approval/editing before using it. If there are no local changes, say so and fall back to asking for the title manually.
    • Enter manually — ask the user to type the title.
  • Descriptiondo NOT ask the developer for it; generate it yourself. Write a short 23 sentence summary of what the change is and why (English or Russian are both acceptable; default to English), derived from the local changes (the same git inspection used for the summary) and the chosen title. It must be a concise digest, not a full listing of every local change, file, or diff. Show the generated description in the Phase 3 preview so the developer can approve or edit it before creation.
  • Parent (optional) — a Story or Epic. Ask via AskUserQuestion with these options:
    • No parent — proceed without one (omit the parent field).
    • Provide a link/key — the user knows the parent. When this option is chosen, ask in a separate follow-up message for the link or AND-xxxxx key. Then extract the key (AND-\d+) from whatever the user pastes (a plain key or a full browse/AND-... URL).
    • Search by keyword — the user doesn't know the key; ask for a keyword and run searchJiraIssuesUsingJql with jql: 'project = "AND" AND issuetype IN (Story, Epic) AND statusCategory != Done AND summary ~ "<kw>" ORDER BY updated DESC', then let them pick. Sanitize <kw> before interpolating — escape \ and " (and drop other JQL metacharacters) so the keyword can't break or alter the query; if it can't be safely escaped, ask the user to rephrase.
  • QA Notes — testing notes for QA. Must be strictly in Russian. Optional. Ask via AskUserQuestion. The option labels are in English, but the value written to the QA Notes field stays in Russian:
    • Nothing to test → write the exact Russian text Ничего тестировать не нужно.
    • Test within the story → testing happens within the parent Story; write the Russian text Тестируйте в рамках стори. If the change is gated behind a feature toggle (detect a toggle name from the local changes / context, e.g. a new entry in feature_toggles_config.json or an XxxFeatureToggles usage), append it: Тестируйте в рамках стори, закрыто тогглом "<название>". If no toggle is found, use the plain text without the toggle clause.
    • Generate from changes → generate QA Notes (in Russian) from the local changes, written for testers: describe the user-facing behaviour to verify in plain language, without any code-level names (no class / component / function / file names). You may include concrete test cases (step → expected result). If you know the feature toggle that gates this functionality (detect it from the local changes / context — e.g. a new entry in feature_toggles_config.json or an XxxFeatureToggles usage), add Закрыто тогглом "<название>". Show the generated text in the Phase 3 preview for approval/editing.
    • Enter manually → let the user type custom QA Notes (in Russian).
    • No QA Notes → leave the field empty.
  • Stream (optional) — ask via AskUserQuestion offering the values from the Stream option ids table plus a Skip (no Stream) option (last). Core is the common default for most work; pick a specific stream only when it clearly applies. If the user skips, omit the field. Map the chosen value to its option id.
  • Story Points — only ask if the user mentions it or hints at it; otherwise skip silently.
  • labels / components / Developernever ask for these. Only set them if the user provided them explicitly in $ARGUMENTS; otherwise omit them entirely.

Validate any provided parent key with getJiraIssue (fields ["summary","issuetype","parent"]) so the preview can show the parent's title, confirm it exists, and read its own parent. Use the parent's issuetype to decide what goes where:

  • Epic → the Epic is the hierarchy parent (Phase 5) and the target of the Phase 5b link.
  • Story → the Task inherits the Story's parent Epic as its hierarchy parent (so it lands in the same Epic), and the Phase 5b "implements" link points to the Story. Read the Story's Epic from its parent field (the getJiraIssue above; if absent there, fetch the Story with fields:["parent"]). If the Story has no parent Epic, omit the parent field and rely on the link only.
  • Any other type (Task, Bug, Sub-task, …) → a non-Epic, non-Story key is not a valid parent here: it can't be a hierarchy parent (same hierarchy level) and isn't an "implements" target for a Task. Warn the user that the key is a <issuetype> and ask (via AskUserQuestion) whether to provide an Epic/Story key instead or proceed with no parent (omit parent and skip the Phase 5b link). Never silently treat it as an Epic or Story.

Track two distinct values: linkTarget (the chosen parent — Story or Epic, used for the Phase 5b link) and hierarchyParent (the Epic that goes into the parent field — the Epic itself, or the Story's inherited Epic, or none). Reflect both in the preview, e.g. Parent: [REDACTED_TASK_KEY] (Story, link) — hierarchy Epic [REDACTED_TASK_KEY] (inherited) or Parent: [REDACTED_TASK_KEY] (Epic, hierarchy + link).

Phase 2 — Resolve the current sprint

Run:

searchJiraIssuesUsingJql
  jql: 'project = "AND" AND sprint IN openSprints() ORDER BY updated DESC'
  fields: ["customfield_10021"]
  maxResults: 50

Inspect the customfield_10021 arrays across the returned issues (each is an array of sprint objects — a single maxResults: 1 issue could belong only to a future open sprint). Collect the distinct sprint with state == "active"; use its id (number) for the Sprint field and its name for the preview.

If no active sprint is found, tell the user and ask whether to create the Task without a sprint or to supply a sprint id manually. Never guess a sprint id.

Phase 3 — Build the preview

Render a compact table of every field that WILL be sent, e.g.:

About to create a Jira TASK in project AND:

  Summary      : <summary>
  Type         : Task
  Assignee     : <self name> (you)
  Sprint       : Mobile Sprint 208 (id 4181)
  Stream       : Core (or "—" if skipped)
  Parent       : [REDACTED_TASK_KEY] — Wallet registration (Story, link) — hierarchy Epic [REDACTED_TASK_KEY] (inherited)
  QA Notes     : <text or "—">
  Story Points : <n or "—">
  Description  :
    <first ~5 lines, or "—">

Show empty fields as so the developer sees exactly what is and isn't set.

Phase 4D — Dry-run exit (when --dry-run is set)

If $ARGUMENTS contains --dry-run, do not ask for confirmation and do not call createJiraIssue. Instead, after the Phase 3 preview, print the exact createJiraIssue payload that would be sent (all params and additional_fields), prefixed with a clear banner:

DRY RUN — no Jira issue was created. Payload that would be sent:

If a parent was set, also note the issue link that would be created after the issue: createIssueLink type="Polaris work item link", inwardIssue=<new Task>, outwardIssue=<linkTarget> (reads " is implemented by "), and — for a Story parent — that the hierarchy parent would be the Story's inherited Epic. Then stop — this is the full extent of the run in dry-run mode.

Phase 4 — Confirm (mandatory gate)

(Skipped entirely in dry-run mode — see Phase 4D.)

Call AskUserQuestion with the question "Create this Jira Task?" and options:

  • Create — proceed to Phase 5.
  • Edit fields — go back to Phase 1 and adjust the field(s) the user names, then re-preview.
  • Cancel — stop; create nothing.

Do not call createJiraIssue until the user selects Create.

Phase 5 — Create

Call createJiraIssue:

cloudId: "tangem.atlassian.net"
projectKey: "AND"
issueTypeName: "Task"
summary: "<summary>"
description: "<description>"            # omit if empty
contentFormat: "markdown"
assignee_account_id: "<self account_id>"
parent: "<hierarchyParent>"            # the Epic (chosen Epic, or the Story's inherited Epic); omit if none
additional_fields: {
  "customfield_10021": <sprintId>,      # omit if no sprint
  "customfield_11931": { "id": "<streamOptionId>" },   # Stream — omit if skipped
  # QA Notes — ADF document, NOT a plain string (omit if empty):
  "customfield_11232": {"type":"doc","version":1,"content":[{"type":"paragraph","content":[{"type":"text","text":"<QA Notes>"}]}]},
  "customfield_10025": <points>         # omit if not set
}

For multi-line QA Notes, use one paragraph per line inside the ADF content array.

Fallbacks (retry once, only on a field-specific error):

  • Issue type rejected → retry with issueTypeName: "Задача".
  • Stream (customfield_11931) rejected for { "id": ... } → retry that field with { "value": "<value>" }.
  • parent rejected with "does not belong to the hierarchy" → hierarchyParent should already be an Epic, so this is unexpected (e.g. a non-Epic slipped through). Drop the parent field and create the Task without it; the relationship is still carried by the Phase 5b "implements" link.
  • parent (an Epic) rejected for another reason → drop parent and set additional_fields.customfield_10014: "<epic>" (legacy Epic Link).
  • A single custom field rejected → report which field failed and ask whether to retry without it.

After the Task is created, if a parent was provided, create an issue link to linkTarget (the chosen parent — the Story, or the Epic) so it reads "is implemented by " (the new Task implements its parent). Note: the hierarchy parent set in Phase 5 is the Epic (hierarchyParent), while this link points to the chosen parent — for a Story parent those are two different issues (the Task sits under the Story's Epic and implements the Story).

createIssueLink
  cloudId: "tangem.atlassian.net"
  type: "Polaris work item link"     # the implements / is implemented by link type
  inwardIssue: "<new issue key>"     # new Task → "implements" the parent
  outwardIssue: "<linkTarget>"       # chosen parent (Story or Epic) → "is implemented by" the new Task

If the link call fails, do not treat the whole run as failed (the Task is already created) — report the error and the link parameters so it can be added manually.

Phase 6 — Report

On success, output the new issue key, summary, and URL https://tangem.atlassian.net/browse/{KEY}, plus a one-line recap of the set fields (sprint, parent, assignee) and — if a parent was set — confirm the " is implemented by " link was created. On failure, surface the API error message verbatim and the field values you attempted.