Release: Runtime Confidence Learning Loops

Forgeflow 4.3.24 closes the loop between first-run evidence, profile review, release verification, and next-work recommendations. The new helpers keep these signals local, advisory, and testable across installed runtimes.

First-Run And Next-Work Evidence

• New /forgeflow-first-run-rollup summarizes local first-run result files into aggregate health, smoke, profile, decision, friction, and runtime counts.
• New /forgeflow-next-work-outcome records whether a next-work recommendation was useful, ignored, incorrect, or blocked.
• Project intelligence now includes a Next-Work Confidence section and can raise an advisory calibration task when previous recommendations missed the mark.

Operational Review And Release Checks

• /forgeflow-profile-review now includes copy-ready advisory commands and a --commands-only view for applying safe profile cleanup steps deliberately.
• /forgeflow-release-verify --github optionally checks the GitHub release and remote tag state using read-only commands.
• Installed-runtime dogfood now verifies the new profile, first-run, next-work, and release verification helpers after installation.

Safety And Documentation

• Next-work outcome text uses conservative public-safe validation to avoid storing secrets, paths, snippets, private URLs, or likely customer names.
• Health, release-check, install-manifest, runtime-helper, README, and wiki coverage were updated for the new commands and helpers.
• The release keeps outcome history advisory only: agents still verify current scope, code, tests, and user priorities before acting.

Validation

• Full scripts/forgeflow/test-*.js suite
• node scripts/forgeflow/smoke-check.js --mode source --json
• git diff --check
• Forgeflow review loop reached approval from Warden and Atlas after rendering and installed-runtime coverage fixes.