본문으로 건너뛰기
cccskills
무료GitHub에 공개

create-issue-interaction-ui

Guide development of new interaction cards in Paperclip issue threads, such as approvals or structured questions. Covers shared data definitions, server behavior, interface states, CLI and MCP support, plugin helpers, documentation, and tests.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

원문원문 설명 보기

Add a new Paperclip issue-thread interaction kind end-to-end. Use when repo work introduces or extends interaction cards like request_confirmation, checkbox confirmations, ask_user_questions, or suggest_tasks.

설치 방법 보기

이럴 때 유용합니다

  • Adding an approval card to an issue thread
  • Creating an interaction with a new response format
  • Checking implementation coverage across layers
  • Adding CLI, MCP, and plugin support

포함된 파일(1)

  • SKILL.md13.4 KB

SKILL.md(원문)

설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.

Create a new issue-thread interaction UI (Developer/maintainer skill)

Developer/maintainer skill. Do NOT install this on production Paperclip agents.

This skill walks a Paperclip contributor through introducing a new issue-thread interaction kind from shared contract to issue-detail wiring, helpers, and docs. It is intentionally a developer/maintainer skill: the audience is a human or coding agent making code changes inside paperclipai/paperclip, not the operational agents that run inside a deployed Paperclip company.

Do NOT install this on production Paperclip agents. This guide is for repository contributors changing Paperclip itself.

When to use

  • A new interaction kind is being introduced (compact picker, structured rating, in-thread approval card, etc.).
  • An existing interaction needs a parallel variant with a distinct payload shape, validation, or resolution outcome (and ask_user_questions is the wrong fit because option count, target binding, or result shape differs).
  • A reviewer asks for "the same audit, staleness, supersede, and continuation semantics as the other interactions" on a new card.

When NOT to use

  • Adding fields to an existing interaction kind that does not need a new payload schema. Patch the existing validators/UI in place instead.
  • Changing how Paperclip agents call interactions. Update skills/paperclip or references/api-reference.md; that is agent guidance, not card work.
  • Building a non-thread UI (issue detail sidebar, project board widget, etc.). Those have their own component conventions.

Mental model

Every issue-thread interaction has four moving parts:

LayerOwns
SharedKind constant, payload/result interfaces, Zod validators, exported types, shared-test coverage.
ServerService create/accept/reject/respond, staleness, supersede, idempotency, activity log, wake send.
UICard pending/resolved/stale states, fixtures, Storybook, issue-thread/IssueDetail wiring.
Helpers/DocsCLI command, MCP tool, plugin SDK type+host+testing path, skills/paperclip guidance.

The four existing kinds are the canonical prior art. Pick the closest one and copy its plumbing rather than inventing parallel mechanics:

  • request_confirmation — single yes/no bound to a target with stale/supersede.
  • request_checkbox_confirmation — bounded multi-select against an immutable option set.
  • ask_user_questions — small typed form, no target binding.
  • suggest_tasks — proposes tasks the board can accept individually.

If your new card needs target binding and a yes/no-style resolution, model it after the two request_* kinds. If it is a structured form, model it after ask_user_questions. If it produces creatable child entities, model it after suggest_tasks.

The canonical worked example

The current best end-to-end reference is the checkbox confirmation rollout (merged in 4d5322c82, GitHub PR #7649). Read that diff before starting:

git show --stat 4d5322c82

The plan it implemented is preserved as an issue document on PAP-10415. Use it as the template for your own plan document if you are running this work through Paperclip itself.

Order of operations

Do the shared contract first. It is the smallest correct change you can land even before UI is final, and every later layer reads its types and validators.

1. Shared contract (smallest, lands first)

Touch:

  • packages/shared/src/constants.ts — add the kind string to ISSUE_THREAD_INTERACTION_KINDS and any size constant (mirror REQUEST_CHECKBOX_CONFIRMATION_OPTION_LIMIT = 200).
  • packages/shared/src/types/issue.ts — add Option, Payload, Result, and Interaction interfaces. Extend the IssueThreadInteraction and payload/result union types at the bottom of the file.
  • packages/shared/src/types/index.ts — re-export the new types.
  • packages/shared/src/validators/issue.ts — add Zod schemas for payload, result, and the create-input variant. Reuse the existing requestConfirmationTargetSchema when target binding applies.
  • packages/shared/src/validators/index.ts — re-export the new schemas.
  • packages/shared/src/index.ts — re-export at the package root.
  • packages/shared/src/issue-thread-interactions.test.ts — extend the table tests for the new payload variant.

Validation invariants that have already been litigated and must hold:

  • Option lists are bounded (the checkbox kind uses 200; pick a number the UX can render compactly).
  • Option ids are unique within a payload and any default selection must reference known ids.
  • Labels and descriptions are length-capped to match existing question options. Do not invent looser caps.
  • Target binding uses the shared RequestConfirmationTarget schema so stale expiration runs through one code path.

2. Server service and routes

Touch:

  • server/src/services/issue-thread-interactions.ts — add the kind to:
    • the supported-kinds list (SUPPORTED_KINDS near the top),
    • mapInteractionRow (the switch (row.kind) over payload/result parsers),
    • create-input validation (switch (data.kind)),
    • the accept/reject/respond/stale-expiration branches,
    • the activity-log payload and the continuation wake payload.
  • server/src/routes/issues.ts — extend any kind-specific branches (notably the response-shape branch around line 6096 in the checkbox PR).
  • server/src/__tests__/issue-thread-interactions-service.test.ts — cover create, accept-with-result, reject-with-reason, stale-target expiration, supersede-on-user-comment, idempotency conflict, and wake payload shape.
  • server/src/__tests__/issue-thread-interaction-routes.test.ts — cover create + respond/accept/reject HTTP behavior, company scoping, and authorization.

Server invariants:

  • Board-only resolution. Agent-authored accept/reject must be rejected with the existing 403 path; do not add a per-kind bypass.
  • Company scoping. Reads, writes, expiration, and supersede must all filter by companyId. Never trust an issueId alone.
  • Stale target. If target was specified at create time and a newer revision lands, the interaction expires with outcome: "stale_target". Do not write bespoke staleness — call the same helper the other request_* kinds use.
  • Supersede on user comment. Default supersedeOnUserComment: true unless the payload schema documents otherwise.
  • Idempotency. The deterministic idempotencyKey shape from the existing kinds (<kind>:<issueId>:<decisionKey>:<revisionId>) must be honored; duplicate POSTs must return the existing card, not stack.
  • Continuation policy. Support none, wake_assignee, and wake_assignee_on_accept. Pick a default that matches whether the asker is blocked waiting for the answer (wake_assignee) or only cares about acceptance (wake_assignee_on_accept).

3. UI card and issue-thread wiring

Touch:

  • ui/src/components/IssueThreadInteractionCard.tsx — add a card component (e.g. RequestCheckboxConfirmationCard) and a resolution component (e.g. RequestCheckboxConfirmationResolution). Branch the existing switch by interaction.kind. Reuse the card shell — do not introduce a parallel card frame.
  • ui/src/lib/issue-thread-interactions.ts — add typed helpers like getCheckboxConfirmationSelectedLabels so the card stays declarative.
  • ui/src/lib/issue-thread-interactions.test.ts — test the helpers.
  • ui/src/components/IssueThreadInteractionCard.test.tsx — pending, resolved, stale, disabled/submitting, and validation-error states.
  • ui/src/fixtures/issueThreadInteractionFixtures.ts — seed at least one pending and one resolved fixture for the new kind.
  • ui/storybook/stories/questions-and-approvals/issue-thread-interactions.stories.tsx — Storybook entries for the key states.
  • ui/src/pages/IssueDetail.tsx — extend the per-kind branches the card is rendered from (callback wiring, response submission).
  • ui/src/components/IssueChatThread.tsx — if the kind affects thread-level rendering (badge, summary, count), update the per-kind switches here.
  • ui/src/api/issues.ts — extend the typed accept/reject/respond bodies.

UI invariants:

  • Compact rendering. The card must render comfortably with ~100 options (bounded scroll area, count-first resolved-state summaries — do not chip every selected option inline).
  • Select all and clear selection live inside the card, not in a global menu.
  • The accept payload uses kind-specific field names (e.g. selectedOptionIds, not the suggest-tasks selectedClientKeys). Do not reuse another kind's field name.
  • Stale, superseded, and accepted states render distinct copy; reuse the existing resolution-component shell.

4. CLI, MCP, plugin SDK helpers

External callers must be able to create the new interaction without hand-writing JSON. Touch:

  • cli/src/commands/client/issue.ts — add a CLI sub-command or extend the generic interaction create path.
  • cli/src/__tests__/issue-subresources.test.ts — cover the new flag set.
  • packages/mcp-server/src/tools.ts — add an MCP tool that accepts the new payload shape; reuse the existing createIssueThreadInteraction codepath.
  • packages/mcp-server/src/tools.test.ts — cover the tool's payload shape.
  • packages/plugins/sdk/src/types.ts — add the typed CreateIssueThreadInteraction variant so plugin authors get autocomplete.
  • packages/plugins/sdk/src/worker-rpc-host.ts — extend the kind switch in the create call.
  • packages/plugins/sdk/src/testing.ts — extend the test harness so plugins can simulate the new kind end-to-end.
  • packages/plugins/sdk/tests/testing-actions.test.ts — round-trip test for the new kind through the test harness.

5. Agent guidance

Touch:

  • skills/paperclip/SKILL.md — add a row to the interaction-kinds table: when to use, when not to use, plus a copyable payload example.
  • skills/paperclip/references/api-reference.md — full payload and result schemas, validation limits, create/respond bodies, error codes.

The skills text is read by the runtime agents. Keep it concise — differentiate clearly from sibling kinds in one or two sentences each.

Tests to run before requesting review

The checkbox PR ran exactly this focused set under NODE_ENV=test. Use the same shape for any new kind, swapping in your new test files:

NODE_ENV=test pnpm run preflight:workspace-links
NODE_ENV=test pnpm exec vitest run \
  packages/shared/src/issue-thread-interactions.test.ts \
  server/src/__tests__/issue-thread-interaction-routes.test.ts \
  server/src/__tests__/issue-thread-interactions-service.test.ts \
  ui/src/components/IssueThreadInteractionCard.test.tsx \
  ui/src/lib/issue-thread-interactions.test.ts \
  cli/src/__tests__/issue-subresources.test.ts \
  packages/mcp-server/src/tools.test.ts \
  packages/plugins/sdk/tests/testing-actions.test.ts

If UI vitest fails with act is not a function, the shell is running with NODE_ENV=production (it picks up React's prod build). Re-run with NODE_ENV=test explicitly.

Pre-merge checklist

  • New kind appears in ISSUE_THREAD_INTERACTION_KINDS and is exported.
  • Payload and result interfaces are versioned (start at version: 1).
  • Zod validators enforce option/label/description limits and id uniqueness.
  • Target binding (if any) uses the shared RequestConfirmationTarget path.
  • Service handles create, accept, reject/respond, stale-target, supersede-on-user-comment, idempotency, activity log, and continuation wake.
  • Routes honor board-only resolution and company scoping.
  • UI renders pending, resolved, stale, disabled/submitting, and validation-error states; resolved-state large selections summarize by count first.
  • Fixtures and Storybook entries exist for the new kind.
  • CLI, MCP, and plugin SDK helpers all accept the new payload shape and have test coverage.
  • skills/paperclip/SKILL.md and references/api-reference.md updated.
  • Focused test set above is green; CI gates pass.

Anti-patterns observed in review

These came out of the checkbox PR review thread and are worth avoiding next time:

  • Reusing another kind's accept-payload field name (e.g. piggybacking on selectedClientKeys instead of introducing selectedOptionIds). Each kind owns its own field names.
  • Writing parallel staleness or supersede logic instead of routing through the existing request_confirmation helpers. This silently drifts behavior.
  • Rendering hundreds of selected-option chips in the resolved state. Resolved large selections must summarize by count first.
  • Skipping plugin SDK / MCP / CLI coverage on the theory that "the API is generic enough." External callers do not pick up new kinds without typed helpers, and the absence shows up later as broken agent flows.
  • Adding the kind to skills guidance before the server route accepts it. Agents will try the new kind and 400 in production.

리뷰

아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.

댓글

로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.

아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.

보내면 바로 공개됩니다. 이메일 주소나 전화번호 같은 개인 정보는 쓰지 마세요. URL은 2개까지입니다.

0 / 1,000자

이 양식은 reCAPTCHA로 보호되며 Google 개인정보처리방침 및 서비스 약관이 적용됩니다.

같은 저장소의 스킬

Add or extend Paperclip end-to-end evaluations that exercise real browser workflows across the server, database, and runner. Define expected outcomes, check independent evidence, and connect results to the existing reporting pipeline for local or Daytona execution.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Adding an evaluation for a complete user workflow
  • Updating test data and outcome checks
  • Including screenshots and artifacts in evaluation reports
paperclipai/paperclip10.4만2026년 10월 18일 업데이트

Add or extend Paperclip Runner evaluations that test real sessions against a mock control service. Define observable pass conditions, validate cases and run selections, and retain source revisions, evidence, and costs for authorized live runs.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Adding a focused Runner protocol evaluation
  • Testing grading rules with valid and invalid evidence
  • Updating evaluation selections and campaign membership
paperclipai/paperclip10.4만2026년 10월 18일 업데이트

Use a real browser to verify web pages and apps through short, supervised interactions. Capture screenshots, inspect errors and network requests, and document reproducible steps for UI bugs or expected behavior checks.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Capture screenshots to verify UI changes
  • Read content rendered by JavaScript
  • Reproduce UI bugs and record errors
paperclipai/paperclip10.4만2026년 10월 18일 업데이트

agentmail

무료

Read email linked to assigned Paperclip tasks, send new messages or replies through an assigned AgentMail inbox, and check delivery outcomes. Follow task permissions and preserve retry identifiers to avoid duplicate sends.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Reading email context attached to an assigned task
  • Sending a new email from an assigned inbox
  • Replying to a specific message in a conversation
paperclipai/paperclip10.4만2026년 10월 18일 업데이트

check-pr

무료

Inspect proposed changes in GitHub, GitLab, or Perforce for review feedback, failed automated checks, and incomplete descriptions. It separates actionable findings from informational or resolved comments, then helps fix issues and close addressed review threads when requested.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Inspect a change before submission
  • Identify failed checks and required fixes
  • Organize human and bot review feedback
paperclipai/paperclip10.4만2026년 10월 18일 업데이트

Create an AI agent company or team package from an idea or an existing repository. Define roles, reporting lines, skills, and work handoffs, then generate structured files that follow the Agent Companies specification.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Build an agent company from an idea
  • Turn a repository into an agent team package
  • Define agent roles and work handoffs
paperclipai/paperclip10.4만2026년 10월 18일 업데이트

paperclipai의 스킬 모두 보기

이 스킬의 문제 신고