griptape-nodes-e2e-pipeline

name	griptape-nodes-e2e-pipeline
description	End-to-end test generation pipeline for a Griptape Node. Orchestrates the survey, inspect, plan, and workflow skills in sequence. Confirms the output directory with the user, manages subagent execution, presents the test plan for approval, and gates workflow generation on that approval.
compatibility	Requires an MCP connection to a running griptape-nodes engine (extended tools mode).
metadata	{"author":"the-foundry-visionmongers","version":"0.1"}

E2E Test Generation Pipeline

Purpose

Orchestrate the full end-to-end test generation pipeline for a Griptape Node:

1. Survey — static analysis of the node's source code.
2. Inspect — live confirmation against the running engine.
3. Plan — propose a test matrix for user approval.
4. Workflow — generate and save test workflows.

This skill runs in the main conversation and handles all user interaction — confirming the output directory, presenting the test plan, collecting approval, and relaying results. Each step is delegated to a subagent that invokes the corresponding skill.

---

Inputs

Before starting, collect two pieces of information:

1. Node class name — the target node to test (e.g. Agent).
2. Output root directory — the absolute path where artifacts are written. Inspections go to {output_root}/inspections/ and test workflows go to {output_root}/tests/<NodeType>/.

If the user's message already contains both values, use them directly. Otherwise use AskUserQuestion to collect whatever is missing. Suggest the current working directory as the default output root.

---

Resuming the Pipeline

The user may want to resume from a specific step — for example, re-running only the plan step after manually editing the inspection report, or jumping straight to workflow generation after approving a plan in a previous session.

If the user's message specifies a starting step (e.g. "start from plan", "just run workflows", "re-run inspect"), skip earlier steps. Before skipping, verify the required input files exist:

Step	Required input files
Survey	(none)
Inspect	{output_root}/inspections/<NodeType>.survey.md
Plan	{output_root}/inspections/<NodeType>.survey.md, <NodeType>.inspect.md
Workflow	{output_root}/inspections/<NodeType>.plan.md

If a required file is missing, tell the user and ask whether to run the missing step first.

---

Execution

Step 1: Survey

Spawn an Agent subagent with this prompt (substitute {node_name} and {output_root}):

Invoke the /griptape-nodes-e2e-survey skill to survey the {node_name} node. The output root directory is {output_root}.

Wait for the subagent to complete. If it reports an error, tell the user and stop.

After success, confirm to the user: "Survey complete. Proceeding to inspect."

Step 2: Inspect

Spawn an Agent subagent with this prompt:

Invoke the /griptape-nodes-e2e-inspect skill to inspect the {node_name} node. The output root directory is {output_root}.

Wait for the subagent to complete. If the subagent reports a BLOCKED verdict, tell the user the specific blockers and stop — do not proceed to the plan step.

After success (verdict: PASS), confirm to the user: "Inspection complete (verdict: PASS). Proceeding to plan."

Step 3: Plan

Spawn an Agent subagent with this prompt:

Invoke the /griptape-nodes-e2e-plan skill to create a test plan for the {node_name} node. The output root directory is {output_root}.

Wait for the subagent to complete.

Step 4: Present the Plan

Read the plan file at {output_root}/inspections/{node_name}.plan.md. Present a focused summary to the user covering these sections:

1. Configuration Tests — show the full table from the ## Configuration Tests section.
2. Error Tests — summarise counts: how many pre-execution validation tests, how many runtime error tests, and the design-time input handling status.
3. Rejected Configurations — show the table from ## Rejected Configurations if it has any rows.
4. Caveats — mention any environment dependencies or limitations from the ## Caveats section.

Then use AskUserQuestion to ask the user what to do next:

• Approve — proceed to workflow generation.
• Edit — describe the changes (the user will provide change instructions in a free-text response or via the "Other" option).

Step 5: Plan Edit Loop

If the user requests edits:

1. Collect the user's change description. This may come from the AskUserQuestion response text or from a follow-up message.

2. Spawn an Agent subagent with this prompt:

   Invoke the /griptape-nodes-e2e-plan skill to edit the existing test plan for the {node_name} node. The output root directory is {output_root}.

   Apply these changes: {user_requested_changes}

3. Wait for the subagent to complete.

4. Go back to Step 4 — re-read the updated plan file and present it again.

Repeat until the user approves.

Step 6: Workflow Generation

Once the user approves the plan, spawn an Agent subagent with this prompt:

Invoke the /griptape-nodes-e2e-workflow skill to generate test workflows for the {node_name} node. The output root directory is {output_root}.

Wait for the subagent to complete.

Step 7: Present Results

Read the workflow summary at {output_root}/inspections/{node_name}.workflows.md. Present to the user:

• Total number of workflows generated.
• Pass / fail / skipped counts.
• Any failures with error details.
• Location of saved workflow files.

---

Error Handling

• If any subagent fails, report the error to the user and ask whether to retry the step or stop.
• If the inspection verdict is BLOCKED, stop after inspect and report the blockers — do not proceed to plan.
• If the user cancels during the plan approval loop, stop without running workflows.