nspec.dev launches in public preview
The marketing site, waitlist API, and legal/company pages went live. Dogfooded a full NSPEC standard-tier run against the site itself; four verified bugs were caught on the first pass (mobile menu close button, a dead pricing CTA, an enterprise-tier link, and placeholder footer stubs), all fixed within the day. MX, SPF, DKIM, and DMARC records configured end-to-end so the waitlist confirmation email lands in the inbox, not spam.
Server-side verifier dispatcher and hang watchdog
Promoted the bug-verifier from subagent-only to a first-class server-side dispatcher, so every verdict is recorded with its reproduction attempts, confidence, and source artefacts in the manifest regardless of how an agent was invoked. Shipped a watchdog that kills hung agent processes instead of stalling the run. Fixed the pages/ walker regression that was inflating evidence counts from real steps to thousands of stale shots.
Pre-warm MCP handshake, kill the startup race
The run orchestrator was occasionally losing the MCP handshake on cold starts. Moved handshake verification into a pre-warm subprocess and made the runner refuse to schedule work until init returns clean. Pure stdio, process isolation preserved.
Agent loader + --agents CLI flag
Project-scoped agents under .claude/agents/ weren't being discovered by the runner in headless mode. Built an explicit agent_loader plus a --agents flag that passes a JSON manifest, removing the auto-discovery assumption. Verified green on a full standard-tier run against a reference app.
Streaming parallel verification
Bug verification now streams in parallel instead of sequencing through a single verifier. Standard-tier wall time for comprehensive runs dropped under four minutes on typical apps. Per-tier timeout bumps prevent false P1s on slow environments.
Project memory, false-positive prevention, Docker infra
Introduced cross-run project memory: flaky-selector detection, known-false-positive suppression, and a cross-run diff so every pass focuses on what changed. Containerised the runner end-to-end for repeatable deployment. Dashboard gained a runs-over-time view, faster filtering, and N+1 query fixes.
Report server for historical runs
Ship a lightweight report server so teams can browse any prior run without needing the CLI. URLs are stable and link directly into a specific bug bundle.
Self-contained HTML reports
HTML reports now embed screenshots as base64 so the artefact is a single drop-in file · easy to email, easy to attach, no broken image links.
Auth, RBAC, admin portal, prompt editor
Multi-user foundations: authentication, role-based access control, an admin portal for managing projects and members, and an inline prompt editor for tuning each specialist agent.
Initial NSPEC v2 runtime
Ground-up rewrite of the QA orchestrator, eight specialist agents (orchestrator, ui-explorer, component-auditor, responsive-tester, performance-profiler, bug-verifier, test-case-designer, accessibility-reviewer), six-viewport coverage, and the evidence-bundle output format that's still in use.
Get new entries in your inbox · join the waitlist on the home page.