Tycho v0.14.0

Tycho v0.14.0

Inspect local harness catalogs

Use tycho harness list to list exact model IDs and reasoning suggestions in
text or JSON. The offline default reads shared fallback catalogs and configured
harness_catalogs for built-in and custom harnesses without running tools,
checking authentication, or contacting providers.

tycho harness list --harness claude --model claude-opus-5-5 --json

--harness and --model are exact, case-sensitive filters. A missing catalog
entry does not establish provider availability or support. The command is local
only; it has no --server option or remote forwarding.

Optional --discover runs the local catalog commands used by Settings. These
can depend on installed tools, credentials, and provider/network access. The
reported catalog_source identifies the source or fallback used. This command
adds no provider API query or authentication check of its own.

Unknown defaults remain JSON null or text unknown; harness decides.
Harness-wide effort suggestions do not prove model-specific support. For
claude-opus-5-5, low, medium, high, xhigh, and max are harness
suggestions; model effort support and defaults remain unknown. Model-specific
reasoning lists appear only when the source provides them. Catalogs do not prove
live entitlement or validate a model/reasoning pair. Project/template overrides
are not listed.

Keep agent work continuous

Context replacement preserves recurring schedules. Start New saves a fresh
clone before it archives the source. Start with Handoff keeps a concise
continuity record and archives the source after the replacement starts. Queue,
archive, and scheduler race fixes protect ordinary work and delegation history.
Context-pressure warnings can be dismissed for the current evidence.

Conversation records agent sends and retains expandable Read queue evidence.
Inquiry values save while you type and survive refresh and reload. Internal
runners avoid incompatible user Ruby gems, and valid configured custom project
harness profiles pass validation.

Review messages and files with less effort

• Preview selected Summary attachments in a third column on wide screens, with
  independent scrolling, safe renderers, keyboard focus, and corrected sizing.
• Read safe Markdown frontmatter as metadata and keep nested list structure.
  Rejected source remains visible; HTML stays sanitized.
• Inspect known errors in Summary details while the recorded result stays intact.
• Open Conversation at its latest message. Selected spacing includes a total
  64 px gap before qualifying top-level user blocks, without an extra first gap.
• See an animated marker indicate working and successful completion. Polling preserves motion
  state, completion plays once per loaded UI session, and reduced motion uses
  static states.
• Read the self-hosted Plus Jakarta Sans variable font, with a system fallback.

Clearer Settings and notification evidence

Mobile Settings scrolls above its navigation. Consistent headings, compact
healthy checks, visible warnings, correct tool counts, and aligned actions make
setup information easier to inspect.

Notification routes retain their targets while Conversation loads and correctly
resolve encoded retained IDs. Settings includes redacted click tracing, export,
and an explicit test sender for the current browser. The sender shares the normal
retained-record and send path, resolves Tycho and native session IDs, and reports
missing or ambiguous targets without changing the selected agent.

These changes improve routing and diagnosis. They do not establish that every
native Edge notification click, installed-PWA activation, or cold-window opening
works. Provider acceptance is not proof of native display or activation.

Ruby test commands use isolated storage and a Web Push guard so normal regression
runs cannot send real browser notifications.

Usage estimates

The embedded GPT-6 rate card includes cache-write accounting. Cache writes replace
the corresponding ordinary input cost. Unknown rates or incomplete telemetry
remain explicitly unpriced. Estimates represent API list-price equivalents,
not subscription charges or invoices.

Compatibility

• Ruby 3.2 or later remains required. No new Ruby gem dependency is added.
• No manual migration is required for projects, agents, schedules, delegation,
  queues, memory, credentials, notifications, or attachments.
• Context replacement archives the source at its safe lifecycle point and
  transfers schedule ownership. It blocks running agents, unresolved inquiries,
  and ordinary or mixed queued work.
• Manual archive stops active work, cancels inquiries, and records queue outcomes
  and delivery evidence without running the queued work. Protected delegation
  callbacks remain in read-only history; delivered or in-flight work is marked
  aborted with uncertainty.
• Existing browser token reads, removal after successful host persistence, and
  peer forwarding remain supported. No credential reset is required.
• Fonts and their SIL Open Font License ship in the package. No external font
  service is required. The packaged Tycho skill is unchanged from v0.13.0.
• Bottles target Apple Silicon macOS and Linux. v0.11.0 remains the final Intel
  macOS bottle; Intel users can continue from source.

See the v0.14.0 upgrade guide
for lifecycle details and verification commands.

Validation checklist

Release validation covers the full bin/test suite, package regression, version,
schema and skill checksums, bundled font files and license, and focused Chrome
checks with synthetic data. The release PR records exact results and its required
signoff/tycho-bin-test status for the final published commit.

Rollback

Before publication, revert the version-and-documentation change if required.
After publication, keep the public tag fixed and issue a corrected patch release.
For a severe fault, follow the correction process in the
release runbook.