Artifacts

Reference

Most of the product’s value shows up in the artifacts it writes. Treat these files as part of the operating model, not temporary scratch output.

Output root

Most useful artifacts live under one working directory

That makes it easier to archive them in CI, inspect them locally, and feed them into release reviews or downstream tooling.

Artifact root

Default working directory

<testsRoot>/.e2e-ai-agents/

Core Planning Artifacts

File	Produced By	What It Is For
plan.json	plan	Structured coverage plan with impacted families, run sets, confidence, and decisions
ci-summary.md	plan	Markdown summary for PR comments or release reports
metrics.jsonl	plan	Append-only event log for metrics and later cost reporting
metrics-summary.json	plan	Aggregated metrics for dashboards or CI summaries

Advisory Report

plan --advisory (also suggest or gate with --advisory) bypasses the ordinary plan, summary, metric and status writers. It emits one JSON object on stdout; the caller saves or uploads it. See the report contract.

The existing plan schema includes an advisory object with full commit/suite identity, complete changed files, exact static spec inventory and selected paths, declared mapping provenance, input hashes and per-file fallback reasons. confidence is null; coverage and execution are unavailable, and release safety is not assessed. Identical inputs produce identical substantive plans, while generatedAt may differ. An uploaded file can also contain an error, so validate its content and process exit status.

Knowledge And Training Artifacts

File	Produced By	What It Is For
route-families.json	train or bootstrap	The core manifest mapping code paths to product areas and flows
train-report.json	train	Training stats, coverage numbers, and enrichment timing
traceability.json	traceability-ingest	Rolling file-to-test mapping manifest
traceability-state.json	traceability-ingest	Hit counts and rolling ingest state
failure-history.json	failure correlation	Historical file-to-failure relationships used for confidence boosts

Review And Generation Artifacts

File or Directory	Produced By	What It Is For
review-generate-summary.json	review --generate	Review-driven generation results with scenario mapping
agentic-summary.json	generate	Standalone agentic generation results
generated-needs-review/	generation stage	Quarantine area for specs blocked by hallucination guardrails or verification failures
qa-screenshots/	impact-gate-qa	Browser screenshots captured during exploratory testing or fix verification
reports/	reporting flows	Structured outputs consumed by downstream tooling or humans

How Teams Usually Use Them

Operating model

How teams usually work with the artifacts

• plan.json feeds automation and deeper tooling
• ci-summary.md is posted into PRs or release discussions
• route-families.json is maintained as product knowledge
• traceability.json improves precision over time
• generated-needs-review/ becomes a manual review queue

Practical habit

Inspect the written artifacts, not just stdout

The artifacts are the easiest way to understand what the tool decided, why it made that decision, and what a release owner or QA lead should do next.