pyto-agent v1.0.25

pyto-agent v1.0.25

Shortcut fixture instructions printed in Pyto

The validation script now supports --setup, which prints the setup steps for all A1–A20
Shortcut fixtures. Before running a case, it also prints the recipe for each fixture that
case will invoke. This keeps the required fixture names, actions, safety notes, and A1
iCloud install link available on-device without returning to the documentation.

Pyto cannot author or install Shortcuts from Python. A1 remains available through the
shared iCloud link; other fixtures are created once in the Shortcuts app. The script still
invokes only explicitly named pyto-harness-test- fixtures and requires typing RUN.

Verification status

• No automated test suite or new Pyto device run was performed for this release.
• Device behavior remains unverified until the user runs the kit in Pyto and reviews the
  timestamped JSONL output.

pyto-agent v1.0.24

pyto-agent v1.0.24

A1 Shortcut fixture install link

The validation guide and A1 prompt now include a shared iCloud link to the harmless
pyto-harness-test-return fixture. This avoids manually building the A1 fixture. Apple
still requires a person to tap Get Shortcut to add it. The validation script itself
does not make network calls; the one-time import is handled by Shortcuts and iCloud.

The A1 fixture is the only test shortcut included through this link. Other A1–A20 fixtures
retain their existing instructions.

Verification status

• The shared iCloud URL was supplied by the user; this environment could not inspect its
  contents. The fixture’s expected behavior is based on the user’s successful A1 run.
• No automated tests or new Pyto device tests were performed for this release.

pyto-agent v1.0.23

pyto-agent v1.0.23

Shortcut validation diagnostics

The A1 result now records the installed harness version and includes a short exception
message when the direct xcallback.open_url call fails. URL strings are redacted and the
message is capped at 240 characters. This gives the on-device check enough detail to
distinguish common callback and handoff errors without logging the callback URL.

The updated script remains bundled with the installer beside run.py; no separate script
download is needed.

Verification status

• No automated tests or new Pyto device runs were performed for this release.
• Existing device observations remain specific to the kit versions recorded in their logs.

pyto-agent v1.0.22

pyto-agent v1.0.22

Released 7 October 2026.

Shortcut validation URL correction

The A1 validation case now follows Pyto's documented open_shortcut example: it passes
the shortcuts://x-callback-url/run-shortcut action URL to xcallback.open_url without
adding an x-success parameter itself. Pyto's xcallback module manages the return path.
The registered shortcut_run_wait wrapper still constructs its own callback URL; A7
checks that adapter separately.

This corrects the validation kit after an A1 run using v1.0.21 returned RuntimeError.
That observation used the former test URL and does not establish whether the direct Pyto
callback path works. Re-run A1 with this release before drawing a device conclusion.

Verification status

• No new automated or Pyto device tests were run for this release.
• Device behavior remains unverified until a person runs the corrected A1 case and keeps
  the dated JSONL result.

pyto-agent v1.0.21

pyto-agent v1.0.21

Released 7 October 2026.

Evidence-based capability foundation

The documentation now inventories all 46 fixed model-facing capabilities with their
inputs, outputs, effects, prerequisites, retry metadata and evidence sources. Effects and
data-egress entries are descriptive metadata; they do not enforce policy or prevent
generated Python from using Pyto APIs directly.

The manual A1–A20 Shortcut validation kit and its focused guide describe harmless test
fixtures, structured observations and optional stress/recovery checks. It makes no LLM or
API calls. Shortcut callback behavior remains unverified until someone runs the script on
a real Pyto installation and records the results.

The installer extracts shortcut_validation.py into the installed pyto-agent folder
beside run.py. Open that script in Pyto and tap Run; a separate script download or copy
is not required. For the quickest check, create pyto-harness-test-return, choose A1,
and confirm the named fixture when prompted.

Local handle pipeline prototype

An example demonstrates a workspace-file flow through local transformation and output.
Handles expose metadata while retaining text or typed binary artifacts outside model
context. This remains a local prototype; it does not add a registered model-facing
capability or enforce data-egress policy.

Verification status

• No automated test suite or Pyto device run was performed for this release.
• Device-dependent Shortcut behavior remains unknown pending a dated run log from Pyto.
• No new iOS capabilities were registered.

pyto-agent v1.0.20

pyto-agent v1.0.20

Browser chat is the only graphical harness interface

This release removes the harness's native Pyto chat window. The --ui option is no
longer accepted by run.py or install.py. The installer now offers --web to start
browser chat after setup; --chat still starts terminal chat. Running PytoUI from
generated programs and interactive previews remains supported.

The existing browser interface retains chat, approvals, saved-program forms, session
history, cancellation, and same-device server protections. Terminal chat, saved-program
commands, Pyto device tools, Shortcuts, and CLI/Shortcut input handling remain available.

Verification

• Focused browser, installer, CLI, terminal, and loop regressions passed: 149 tests.
• The full offline suite passed: 880 tests in 98.363 seconds.
• python3 stdlib_audit.py passed with no third-party runtime modules and Python 3.10
  syntax compatibility.
• No new Pyto device test was performed for this release.

pyto-agent v1.0.19

pyto-agent v1.0.19

Formatted assistant replies in the browser interface

Assistant replies in the optional --web interface now render a bounded Markdown
subset, including headings, paragraphs, emphasis, inline and fenced code, blockquotes,
links, and ordered or unordered lists. Code formatting preserves whitespace and scrolls
horizontally on narrow screens. Tool output, approvals, saved-program output, errors, and
session history remain plain text.

Raw HTML and Markdown images are not interpreted. Links are active only for http,
https, and mailto destinations. Malformed or unsupported destinations remain plain
text, and oversized formatted events fall back to plain text. The implementation uses no
Markdown package or external browser resources. See the browser interface guide.

Verification

• The complete offline suite passed: 898 tests.
• The standard-library audit passed; no third-party runtime modules are required.
• Browser checks covered formatted output, unsafe markup, links, and narrow-screen code
  block scrolling.
• Safari behavior, Pyto background behavior, and iOS lifecycle handling still need
  acceptance on a real device.

pyto-agent v1.0.18

pyto-agent v1.0.18

Browser interface for the harness

This release adds an optional python run.py --web front end for using the existing
Pyto harness from Safari on the same device. It provides chat, approval controls, saved
program forms, and session history. The server listens only on loopback, uses a random
per-run token, and keeps the provider API key inside the Python process.

The interface includes controls to stop a turn or close the web session. When started in
Pyto it uses a background task to help the server stay active while Safari is foregrounded.
As with other iOS background work, suspension or termination under system pressure remains
possible. The browser interface guide lists the device checks.

Verification

• The complete offline suite passed: 889 tests.
• The focused web tests passed, covering the local HTTP interface, authentication, chat,
  approvals, saved-program execution, cancellation, and shutdown.
• A desktop browser smoke check covered the main page and a narrow mobile viewport.
• Safari behavior, Pyto background behavior, and iOS lifecycle handling still need
  acceptance on a real device.

pyto-agent v1.0.17

pyto-agent v1.0.17

Released 6 October 2026.

Shortcuts diagnostic capture fix

The v1.0.16 device report showed that the five Shortcuts saved-run cases reached their
expected return codes and preserved the no-execution checks, but their expected response
text was missing from captured stdout/stderr. This is consistent with Pyto's native console
not following contextlib.redirect_stdout and redirect_stderr in the test process.

The diagnostic tests now capture run.py's module-level print calls directly, without
replacing the interpreter's global console streams. They still assert the opt-in and refusal
messages, and the Unicode saved input. No Shortcuts command behavior changed in this release.

This scripted check does not launch the Shortcuts app or certify a real Pyto Run Script
handoff. Manual Goal 12 remains NOT RUN until that device interaction is performed.

Verification

• The focused Shortcuts saved-run tests pass on desktop, including in-process program execution.
• The complete offline suite and standard-library audit are run for this release.
• Desktop results do not count as native iPhone/Pyto acceptance.

The standalone diagnostic is attached as device_release_diagnostic_v1.0.17.py.

pyto-agent v1.0.16

pyto-agent v1.0.16

Released 6 October 2026.

Shortcuts diagnostic follow-up

The v1.0.15 iPhone report showed that five scripted Shortcuts saved-run cases failed,
but omitted the failure details. The diagnostic now includes a short exception summary for
each failed case, while redacting credential-shaped values, URLs and local paths. This makes
the next device report actionable without including test tracebacks or user files.

The saved-run behavior and its test cases are unchanged in this release. The next on-device
run will show whether those failures are caused by the saved-run implementation or by a
Pyto-specific test/runtime difference. The checks still do not invoke the Shortcuts app or
prove native handoff behavior.

Verification

• The full offline desktop test suite and standard-library audit are run for this release.
• Desktop execution does not count as iPhone/Pyto acceptance; report the updated Goal 12
  result from the device to identify the remaining cause.
• All manual UI acceptance items remain NOT RUN until performed and recorded on-device.

The standalone diagnostic is attached as device_release_diagnostic_v1.0.16.py.