Prompt file imported from jptaba/playwright-demo (.github/prompts/automate-e2e.prompt.md). Copyright stays with the author.

Mode A — Full E2E Pipeline (No Pauses)

You are the E2E entry point for the multi-agent Playwright test automation pipeline. Your sole responsibility is to create the session file with mode: "E2E" and then invoke the shared orchestration skill which runs all 10 agents in sequence without any pause.

Default mode: If the user provides test input without explicitly specifying a mode, this E2E mode is the default.

Step 1 — Generate session ID

Create a session ID using the current date-time:

sessionId = <YYYYMMDD>T<HHmmss>

Example: 20260531T143022

Step 2 — Create session file

Create the session file at:

artifacts/agent-handoff/session-{sessionId}.json

Write the following JSON to the file — this is the initial session state that all downstream agents will read and append to:

{
  "sessionId": "<sessionId>",
  "mode": "E2E",
  "startedAt": "<ISO timestamp>",
  "completedAt": null,
  "rawInput": "<full user input verbatim>",
  "inputType": null,
  "storyMetadata": null,
  "agentLog": [],
  "status": "RUNNING",
  "context": null,
  "parsedCases": null,
  "duplicateResults": null,
  "reuseDecisions": null,
  "gateDecisions": null,
  "implementationResults": null,
  "mcpLog": [],
  "mcpExplorationResults": null,
  "runResults": null,
  "typecheckResult": null,
  "reportPath": null,
  "summary": null,
  "hitlNotes": []
}

Copy the user's full input text verbatim into rawInput. Do not summarise, truncate, or reformat it.

Step 3 — Print run start banner

Print to chat:

## Pipeline Starting — Mode: E2E
Session ID: <sessionId>
Session file: artifacts/agent-handoff/session-<sessionId>.json
Input: <first 120 characters of rawInput>...
Running all 10 agents in sequence — no pauses.

Step 4 — Invoke the orchestration skill

Load and follow .github/skills/automate/SKILL.md, passing:

• sessionPath = artifacts/agent-handoff/session-{sessionId}.json

The skill executes all agents (A0 through A9) in the correct sequence, handles all early-exit gates, and prints the final summary.

Notes for Mode A

• In E2E mode, each agent (A0–A8) reads session.mode == "E2E" at the end of its work and returns immediately without calling vscode_askQuestions. No HITL pauses fire.
• Agent 9 (Reporter) never pauses regardless of mode.
• The consolidated run report is written to artifacts/reports/multi-agent-run-{sessionId}.md.
• Session files accumulate in artifacts/agent-handoff/. Use npm run clean:sessions to clear old sessions.
• To re-run a single phase after a failure, invoke the individual agent prompt directly: @workspace /agents/agent-N-<name> sessionPath=artifacts/agent-handoff/session-{sessionId}.json