Roadmap

What's coming in the next release and what's in development.

Providers

What’s moving in zurdo, relative to the released v1.7.0 these docs describe. Items here are subject to change until they ship.

Just shipped: v1.7.0 (2026-07-24)

The surface upgrades milestone — how you drive and read zurdo day to day. Everything below is documented in the main docs; highlights:

  • First-class subcommands. zurdo analyze and zurdo heal, promoted out of run --analyze / run --heal. The old flag spellings keep working through 1.x (with a stderr deprecation notice); removal waits for a future 2.0. Details.
  • The review TUI. zurdo review <prd> walks a run’s evidence — task statuses, per-criterion provenance, the live baseline diff — and takes [manual] sign-offs in-band, recorded to a tamper-evident log; signing a task’s last [manual] criterion flips it passed-pending-reviewpassed. Details.
  • A real help system. Every subcommand’s --help now describes its modes, exit codes, and examples, and zurdo help <topic> prints compiled-in guide pages (hints, prd-grammar, exit-codes, config, workflow, state-dir) offline in the terminal.
  • Completions and man pages. zurdo completions <shell> for five shells, generated from the live command tree; Homebrew installs completions and man pages automatically, and release tarballs bundle both.

Recently shipped: v1.6.0 (2026-07-23)

The capstone of the self-healing loop arc (v1.3 → v1.6):

  • The reason subsystem, complete. Stall detection on failure fingerprints, one-call reasoner diagnosis with retry_with_guidance / halt_task / suggest_heal verdicts, and a repository-scoped lesson library: a task that stalls and then recovers leaves behind a rule future runs are told about before they trip over the same quirk. Opt-in, fail-open, and never able to mark work passed. Details.
  • Lessons reach every authoring surface. The zurdo reason match preview CLI, a lessons section in analyze, lesson injection into heal proposals, and read-only lesson consultation by the bundled zurdo-prd-author and zurdo-hint-debugger skills.
  • Structural hints, executable end-to-end (v1.3–1.4, experimental) against the Lumen index, and the readable live tee (v1.5). Details, details.

In development

The next milestone is being scoped. Candidates noted (not scheduled) during the surface-upgrades design: --format json on more subcommands (validate, verify, state list, check-models) and a zurdo doctor environment check.

Nothing here is committed until it ships — watch this page after the next release, or open an issue to influence what comes next.

Release history

Version Date Highlights
1.7.0 2026-07-24 Surface upgrades: zurdo analyze/heal first-class, zurdo review TUI with logged [manual] sign-off, zurdo help <topic>, completions + man pages.
1.6.0 2026-07-23 Lesson library + cross-surface lesson reads: zurdo reason CLI, lessons in --analyze/--heal and the authoring skills.
1.5.0 2026-07-23 Live agent tee renders step summaries on a TTY; --raw-agent opt-out; claude adapter streams JSONL.
1.4.1 2026-07-22 Structural hints verify the current working tree; per-occurrence binding ambiguity fix.
1.4.0 2026-07-22 [references:]/[callers:] executable via the deterministic binding engine; Vela watcher (zurdo vela).
1.3.0 2026-07-20 Lumen structural index (zurdo lumen), experimental [symbol:] hints, stall detection groundwork.
1.2.0 2026-07-12 Evidence-first verification: baseline capture, pre-flight provenance, evidence-modified warnings, frozen paths, retry-prompt feedback, line-anchored grep.
1.1.3 2026-07-02 --analyze prompts now reference all seven hint types (the analyzer previously never suggested absence hints).
1.1.2 2026-07-01 Parser accepts ### Requirements in its documented position (before ### Description).
1.1.1 2026-06-26 Bundled zurdo-prd-author skill’s grammar reference updated for Requirements/[proves:].

Found a bug, or want to influence what ships next? Open an issue.