Skip to content

Add endgame skill and 0.18.0 test plans (duzitong tasks)#230

Open
duzitong wants to merge 3 commits into
mainfrom
endgame-skill
Open

Add endgame skill and 0.18.0 test plans (duzitong tasks)#230
duzitong wants to merge 3 commits into
mainfrom
endgame-skill

Conversation

@duzitong
Copy link
Copy Markdown
Contributor

Motivation

The endgame verification workflow was living as a standalone .github/prompts/endgame.prompt.md file that couldn't be invoked as a first-class agent skill. This PR converts it into a proper skill and also delivers the 0.18.0 test plans for the tasks assigned to @duzitong.

Changes

Endgame skill (endgame)

Converts endgame.prompt.md into .github/skills/endgame/SKILL.md with:

  • YAML frontmatter (name, description) matching the ui-action skill structure so it is discoverable and invocable
  • Simplified 4-step workflow: ask for inputs -> fetch issue (one gh issue view call) -> delegate each task to a subagent -> summarize
  • Subagent template writes test plans to test-plans/<slug>/<slug>.md, following the project format exemplified by test-plans/thinking-persistence/
  • manage_todo_list dependency removed (no longer needed)

SWTBot smoke-test probe

endgame-chat-open-001.json validates that the Copilot Chat view opens, the model picker resolves, and the chat input is ready. All 14 steps pass locally.

0.18.0 test plans (4 files)

File Covers
test-plans/skills-prompt-files/ TC-001-009: Enable Skills preference, workspace/global skill discovery, prompt files as slash commands, Ask-mode exclusion, auto-refresh, prefix filtering, persistence across restarts
test-plans/cls-session-persistence/ TC-002-009: Conversation history display, Eclipse restart persistence, new-conversation isolation, transcript directory, cancelled-turn restoration, 400 Bad Request regression (tool-call message ordering + duplicate content)
test-plans/subagent-event-leak/ TC-001-003, TC-005: Subagent progress events isolated by conversationId across session switches and new chats
test-plans/subagent-restoration/ TC-001-004: Subagent turns nested inside parent turn after restore, including cancelled turns

Test cases that were not observable from the UI (internal implementation details, LLM quality judgements, timing-dependent races) were excluded to keep the plans actionable.

duzitong and others added 2 commits May 19, 2026 11:31
Convert .github/prompts/endgame.prompt.md into a reusable skill at
.github/skills/endgame/SKILL.md:
- 4-step workflow: ask for inputs, fetch issue, delegate to subagents,
  summarise
- Subagent template writes test plans to test-plans/<slug>/<slug>.md
  following the project format (test-plans/thinking-persistence/ as
  reference)
- Removed manage_todo_list dependency

Add probe script endgame-chat-open-001.json: smoke-tests that the
Copilot Chat view opens, model picker resolves, and chat input is ready.
All 14 steps pass.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- skills-prompt-files: TC-001–009 covering Enable Skills preference,
  workspace/global skill discovery, prompt files, Ask-mode exclusion,
  auto-refresh, prefix filtering, and settings persistence
- cls-session-persistence: TC-002–009 covering conversation history
  display, Eclipse restart persistence, new-conversation isolation,
  transcript directory creation, cancelled-turn restoration, and 400
  Bad Request regression (tool-call history ordering + duplicate messages)
- subagent-event-leak: TC-001–003, TC-005 covering progress event
  isolation by conversationId across session switches and new chats
- subagent-restoration: TC-001–004 covering subagent turns nested
  correctly inside parent turn after restore, including cancelled turns

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot AI review requested due to automatic review settings May 19, 2026 06:06
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR formalizes the existing endgame verification workflow as a first-class agent skill (endgame), adds a SWTBot smoke-test probe to validate that Copilot Chat opens and the model picker resolves, and contributes four 0.18.0-focused manual test plans covering skills/prompt discovery and multiple chat persistence/subagent restoration scenarios.

Changes:

  • Added .github/skills/endgame/SKILL.md to orchestrate endgame verification via delegated subagents that generate test plans under test-plans/.
  • Added a SWTBot probe script (endgame-chat-open-001.json) to smoke-test that the Chat view opens and the model picker initializes.
  • Added four new markdown test plans under test-plans/ for skills/prompt files, CLS session persistence, subagent event isolation, and subagent restoration.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 4 comments.

Show a summary per file
File Description
.github/skills/endgame/SKILL.md Adds an invocable “endgame” skill that fetches a milestone issue once and delegates each assigned task to subagents to produce test plans.
com.microsoft.copilot.eclipse.swtbot.test/probe-scripts/endgame-chat-open-001.json Adds a SWTBot smoke probe to assert Chat view availability and model picker readiness.
test-plans/skills-prompt-files/skills-prompt-files.md Documents manual verification for skills + prompt file discovery and preferences behavior.
test-plans/cls-session-persistence/cls-session-persistence.md Documents manual verification for CLS-backed session persistence/restoration and transcript directory behavior.
test-plans/subagent-event-leak/subagent-event-leak.md Documents verification that subagent progress events are isolated to the active conversation when switching sessions.
test-plans/subagent-restoration/subagent-restoration.md Documents verification that subagent turns restore nested under the parent turn (including cancelled/partial scenarios).
Comments suppressed due to low confidence (1)

test-plans/cls-session-persistence/cls-session-persistence.md:12

  • The test case numbering starts at TC-002 and later skips TC-007. To keep the plan easy to reference (and consistent with other plans like test-plans/thinking-persistence), renumber these to be sequential starting from TC-001 (or add the missing TC sections).
## Test Cases

### TC-002: Conversation history displays correctly after session switch

**Type:** `Happy Path`

Comment thread test-plans/cls-session-persistence/cls-session-persistence.md
Comment thread test-plans/subagent-event-leak/subagent-event-leak.md Outdated
Comment thread .github/skills/endgame/SKILL.md
Comment thread .github/skills/endgame/SKILL.md
Comment thread test-plans/subagent-event-leak/subagent-event-leak.md Outdated
…-leak, update SKILL.md

- cls-session-persistence.md: renumber TCs to be sequential (TC-001..TC-007),
  change ~/ path notation to cross-platform <user home>/.copilot/eclipse
- Drop test-plans/subagent-event-leak/ — premise was wrong (session switch
  cancels subagent via conversation/destroy); scenarios are already covered
  correctly in swtbot.test/test-plans/subagent/subagent.md (TC-003, TC-004)
- SKILL.md: use gh issue view --json body --jq '.body' for reliable parsing
- Delete .github/prompts/endgame.prompt.md — skill is now the single source of truth

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants