Repository navigation
feat(desktop): Check it asks three questions; the recording fills in the rest (hold: founder review) - #1353
Merged
Conversation
A deployment wires the backend's own marker observer into every qualification case. A web app with no OpenAdapt markers (MockMed, most real apps) can't satisfy it, so a project could never be observed and every Desktop practice item stopped with "qualification environment observer does not match the project". The signed-in session branch was unreachable for the same reason. A project whose boundary names Flow's generic web observer (the one the qualify campaign uses) or the signed-in session observer now gets that observer at run time instead of the default adapter. The observer id, its contract, the origin, the declared version and the environment digest are still compared with the project, and the input-edge guard still re-observes before every input, so a page reload mid-run stops the run. observe_web_environment() reads the environment once, for binding a project before its first case. PlaywrightBackend.launch records whether the browser is headless, which that observer's digest includes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… recording Check it used to ask for every part by hand: what the save changes, how to recognize the record, a record-check form, a settings file from IT, a test values attestation, then practice, drills and approval. On MockMed that was 10 inputs, and the practice item was then refused by Flow anyway. Now the engine proposes what the recording shows (get_check_setup): - the record check, from the read the app made after the save (Flow's record-check synthesis, #1284), as the same effect binding a person would type, plus the app's own anti-double-save key; its fault test is run on exactly that binding and shows what it catches and misses; - which steps only read: the network capture shows the app sent data only when the person saved; - the right-record check from the recording, and the app and environment; - which values change per item, from one recording or several. The person answers three things: "Is this the right record check?", "Which values change per item?" (then one practice item whose save they approve, and the drills on their own) and "Approve?". A yes sends the same qualification commands the long form sends; "Change" opens the parts. Values marked as the same on every item are filled in by run_item and not asked in Run it. Without a learned check, the parts ask as before. The person's agent can answer the first two over the local socket and MCP (python -m openadapt_desktop.agent_mcp): read the setup, answer, run the checks. Socket answers are always recorded as the agent's, the window says so, and approving stays in the window. Fixes found on the way, each of which stopped a Check it run on a web app: - start the check bound to an environment observer a web app can satisfy (Flow's web observer, or the signed-in session for an attached browser); - the project gets its own identity policy for the save even when the compiled step already carries the recorded gate; - a case result is bound by the retained report.json's digest; the engine stopped writing run-report-receipt.json in July but still required it, so no practice or drill could be recorded; - a case proves Flow's required actions, as Flow's own qualify campaign does, not steps a person confirmed only read; - the frames and proofs a run report names are retained with the result, so Flow can credit a drill. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Flow binds every case result to the exact executable workflow, and sealing sets the workflow's encrypted flag, so results recorded on the plaintext version can never certify the sealed copy. Approving certified the plaintext version and then sealed a copy that was never certified and couldn't run; loading the copy also failed, because the seal kept the authorization template derived from the certification it cleared. Check it (and the agent's run_checks) now seals before the practice item, runs the practice item and drills on the sealed version, and approves that version. The seal drops the derived template along with the certification. When the agent's checks finish, the window reloads. Also: write the learned record check as a plain effects section (the deployment's own profile still applies), and line-length fixes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
docs/SETUP_INPUTS.md counts what a person gives from a finished recording to the first real item, with a screenshot of each input: - before (#1310): 10 inputs (2 typed), then Flow refused the practice item; - after: 3 decisions and 1 save approval, certified, sealed and approved; the first item ran, gated by the person, and Flow confirmed its save; - with the person's agent over MCP: 2 inputs, both approvals. scripts/setup_walk reproduces it: the real Desktop engine behind a loopback HTTP bridge, real Flow, MockMed's transactional server, and the UI from npm run dev in a headless browser. Synthetic data only. OPERATOR_SCREENS.md describes the new Check it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # packages/flow/openadapt_flow/backends/playwright_backend.py
0.21.1 never reached PyPI (the release check refused the Desktop asset set, fixed in #1403), and its tag cannot be rebuilt. Ship 0.21.0, 0.21.1 and this PR's changes as 0.22.0. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Hold: founder's visual review before merge. Builds on the Calm UI pass (#1309, #1310), now on
main.What this changes
From "I recorded something" to "it runs", Check it now asks three questions. Everything else the recording shows is filled in and shown once, folded:
The person's agent can do steps 1, 2 and the checks over MCP (
python -m openadapt_desktop.agent_mcp). Its answers are recorded as the agent's, the window says so, and only the person approves.Counts, on MockMed, real engine and real Flow runs
Every input, with a screenshot, is in packages/desktop/docs/SETUP_INPUTS.md.
scripts/setup_walk/walk.py --variant before|after|agentreproduces each walk (about 10 minutes).Before: 10 inputs, then a refusal Check it couldn't explain
After: three questions
After, with the agent answering
Bugs found on the way (each stopped a Check it run on a web app; fixed here)
fix(flow)commit); Desktop binds Flow's generic web observer, or the signed-in session for an attached browser.report.jsonbut still required the oldrun-report-receipt.json. Results are now bound by the retained report's digest.encrypted), and the copy failed to load (stale authorization template). Check it now seals before the practice item and approves that sealed version.Needs a decision
network.jsonlthe way Flow's own record-check tests do (stated in the doc).Not changed
No check is weaker: a yes sends the same qualification commands the form sends, with more fields matched; every environment, identity and evidence digest is still compared. Nothing approves on its own, and the agent socket has no approve, certify, seal or save-approval command.
Tests
openadapt-privacynot installed). New:test_check_setup.py,test_agent_setup.py, receipt binding.tscclean.tests/test_web_environment_binding.pyand the web-session tests pass.🤖 Generated with Claude Code