workflow-qa
Skill0 starsChanged 3 months agoMITView on GitHub
Test workflows/competition-demo.json against the CLAUDE.md §15 test categories and data/execution-plan.json's testing_plan, using static inspection of the workflow JSON plus n8n's test_workflow (pin data only) where safe. Use after automation-builder has produced workflows/competition-demo.json.
In this repository10 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- automation-builderSkill
- business-researchSkill
- competition-demoSkill
- execution-planSkill
- lordgen-3d-web-experienceSkill
- lordgen-full-rebuildSkill
- lordgen-site-upgradeSkill
- opportunity-scoreSkill
- proposal-generatorSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 2 are distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 19
Read both inputs in full. If either is missing, stop and report (CLAUDE.md §19) — there is nothing to test against.
- Very specific: line 38
Never call execute_workflow or publish_workflow, under any category, for any reason.
- Distinctive: line 40
Never live-test the duplicate-execution category — it is static-only, always, because there is no way to clean up the real Data Table row …
- Very specific: line 70
All 7 CLAUDE.md §15 categories appear in results, each with an honest method_used and status — a category checked only statically must …
- Very specific: line 71
Every status: "fail" or "not_verified_mcp_unreachable" also appears in known_gaps with a one-line reason.
- Very specific: line 72
Every node name cited in evidence actually exists in workflows/competition-demo.json — no invented node names.
- Very specific: line 73
No live call ever reached execute_workflow/publish_workflow, and duplicate-execution was never live-tested.
- Very specific: line 74
If qa-report.json and qa-report.md disagree on any result, that's a bug in this skill's own output — fix before reporting done.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Workflow QA
Purpose
Actually test the generated starter automation against the 7 required categories (CLAUDE.md §15) — not just check the plan says it will be tested. Report what was verified and how; never claim a pass that wasn't actually checked.
Input
data/execution-plan.json— specificallytesting_plan[](the intended test per category) anderror_handling[](what each failure case should do)workflows/competition-demo.json— the actual workflow to test
Precondition check (do this first, every time)
- Read both inputs in full. If either is missing, stop and report (
CLAUDE.md§19) — there is nothing to test against. ToolSearchfor n8n MCP tools this skill needs (test_workflow,prepare_workflow_pin_data,list_credentials,get_node_types,validate_node_config). Unlikeautomation-builder, do not hard-stop if unreachable — static analysis of the workflow JSON needs no MCP call. If unreachable (or only partially available), run the static-only checks and mark every category that needed a live check asstatus: "not_verified_mcp_unreachable". Never fabricate a pass/fail for something that wasn't actually run.
Method
Static = read workflows/competition-demo.json's nodes/connections directly, no MCP call needed. Live = test_workflow + prepare_workflow_pin_data, pin data only — this is what makes autonomous execution safe (it bypasses real external services per docs/n8n-reference.md). execute_workflow and publish_workflow are never called by this skill, for any category.
| Category | Method | What gets checked |
| Happy path | static + live | Trace Estimate Submitted (Webhook) → Normalize Input → Required Fields Present? (true) → Skip If Already Logged → Draft Estimate Summary → AI Output Valid? (true) → Request Dispatcher Approval → Dispatcher Approved? (true) → Send Summary To Customer → Log: Approved & Sent, against execution-plan.json.workflow_steps. Live: a full valid fixture through test_workflow, confirm the execution reaches Log: Approved & Sent. |
| Missing input | static + live | Confirm Required Fields Present?'s three notEmpty conditions cover customerContact/scope/price; false branch → Log: Missing Input → End (Missing Input), with no path into Draft Estimate Summary. Live: one field blank in the fixture, confirm that branch actually fires. |
| External failure | static only | Check whether either Gmail node (Request Dispatcher Approval, Send Summary To Customer) has an error-output connection to any logging/alerting node. If not connected, this is a genuine gap — report status: "fail" with the missing routing named as evidence, not something to test around. Live pin-data testing bypasses the very external service this category is about, so it can't substitute for the static check either way. |
| AI failure | static + live | AI Output Valid? checks $json.output for notEmpty; false branch → Log: AI Failed → End (AI Failed), no path to Request Dispatcher Approval. Live: pin Draft Estimate Summary's output to empty, confirm the branch is taken. |
| Duplicate execution | static only | Confirm Skip If Already Logged sits before Draft Estimate Summary, keyed on job_id, with no path that bypasses it. If MCP is reachable, use get_node_types/validate_node_config to confirm what the dataTable node's rowNotExists operation actually does on a match vs. no-match — the connections graph has only one outbound edge, so confirm this rather than assume it silently drops duplicates. Do not live-test this category — test_workflow would write a real row into the live "LordGen Estimate Summary Log" Data Table, and there is no MCP tool to delete it afterward. Report method_used: "static" and note why live testing was skipped. |
| Approval gate | static + live | Confirm Send Summary To Customer has exactly one inbound connection, from Dispatcher Approved?'s true branch only. Live: fixture with data.approved: false, confirm Send Summary To Customer never executes (check the execution's node list, not just the final output). |
| Configuration | static (+ optional read-only live) | Confirm OpenAI Model, Request Dispatcher Approval, and Send Summary To Customer each reference a credential (a placeholder is fine) and that no real secret is committed anywhere in the file. Optional: list_credentials (read-only, non-consequential) to confirm referenced credential names exist in the instance. |
Hard rules
- Never call
execute_workfloworpublish_workflow, under any category, for any reason. test_workflowwith pin data is the only autonomous execution mode this skill may use.- Never live-test the duplicate-execution category — it is static-only, always, because there is no way to clean up the real Data Table row it would create.
- Report gaps found in the workflow honestly; do not edit
workflows/competition-demo.jsonto fix them. Remediation is a separate, human-approvedautomation-builderchange.
Required Output
data/qa-report.json:
{
"input_sources": ["data/execution-plan.json", "workflows/competition-demo.json"],
"qa_date": "YYYY-MM-DD",
"n8n_mcp_reachable": true,
"results": [
{ "category": "", "method_used": "static|live|static+live",
"status": "pass|fail|not_verified_mcp_unreachable",
"evidence": "", "notes": "" }
],
"known_gaps": [""],
"approval_required": false
}approval_required is false — running QA is read-only reasoning plus pin-data test calls, not a consequential action itself.
Also write:
tests/workflow-qa/qa-report.md— human/judge-readable twin of the same results, committed to the repo.tests/workflow-qa/fixtures/*.json— the actual pin-data payloads used for each live-tested category, so results are reproducible and inspectable.
Validation (must run before reporting results)
- All 7
CLAUDE.md§15 categories appear inresults, each with an honestmethod_usedandstatus— a category checked only statically must never be reported as live-verified. - Every
status: "fail"or"not_verified_mcp_unreachable"also appears inknown_gapswith a one-line reason. - Every node name cited in
evidenceactually exists inworkflows/competition-demo.json— no invented node names. - No live call ever reached
execute_workflow/publish_workflow, and duplicate-execution was never live-tested. - If
qa-report.jsonandqa-report.mddisagree on any result, that's a bug in this skill's own output — fix before reporting done.
Out of Scope
This skill does not modify workflows/competition-demo.json and does not decide production-readiness. It does not re-test business-research, opportunity-score, proposal-generator, or execution-plan's own internal correctness — only the built workflow those stages fed into.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

