[AI-7065] Refresh flow resume state in the Togo TUI without restarting the command

[luisorofino]

This is PR 1 of 2 in a stacked series:

1. Add composable deterministic task validation #24777 <- you are here
2. Add CHECKPOINTED status to phase nodes to separate it from PENDING #24702

What does this PR do?

Recomputes a flow's resume state whenever a TUI screen becomes active again, and collapses the resume-state API onto a single entry point.

Refresh on ScreenResume

• FlowScreen reads and applies resume state from on_screen_resume, the event Textual posts to a screen that becomes active. Textual posts it on push_screen as well as pop_screen, so one handler covers both the first open and every return — there is no on_mount counterpart to keep in sync, and no duplicate read per open.
• FlowCard holds the resume state in a single resume_state reactive rather than parallel display flags, with resumable derived from it.
• Both the tile label and the Resume button now follow the state in both directions. The old code could only ever set display = True, so a stale button survived a completed run.

One way to read resume state

• CheckpointManager.resume_state(resolved_flow) reads the checkpoint file once and returns a ResumeState of completed and frontier phases. is_resumable is simply a non-empty frontier: a flow that never ran has nothing to start from, so resuming it is just a launch.
• Replaces has_resumable_run, resume_completed_phases, and the module-level resolve_resume_state. runs.py keeps only its real job — locating run directories — behind one flow_resume_state(flow, runs_dir).
• A flow screen previously cost three checkpoint reads (emptiness check, closure, then the graph); it now costs one per open. The "is this resumable?" policy moves out of the TUI, which had been re-deriving orchestration semantics.
• ConfigError translation moves to PhaseOrchestrator, where the "delete it and restart" advice is actionable. The runtime module no longer imports the config error type.

A corrupt checkpoint is reported, not hidden

• ResumeState carries an error, so "nothing was recorded" and "the recording could not be read" stop collapsing onto the same value. resume_state() still raises CheckpointReadError and the orchestrator keeps its precise message; flow_resume_state converts it instead of discarding it, so one corrupt file still cannot crash a grid of flows.
• FlowCard renders a third footer state, ✕ checkpoint unreadable. Previously a corrupt file was indistinguishable from a flow that had never run, which left the remedy undiscoverable from the TUI.

Behaviour changes

• Resuming with nothing left to resume is now refused rather than silently doing something else. Before, no checkpoints ran every phase exactly as a plain launch would, and an already-finished run skipped every phase, registered no processors, and reported instant success without doing any work. PhaseOrchestrator validates is_resumable — the same property the TUI uses — and raises with a reason distinguishing the two cases.
• The TUI refuses first, so that case is an inline message rather than a failed run. Screen.dismiss invokes the result callback before popping, so the launch modal's dismiss commits a tick ahead of the ScreenResume refresh — the app would compute the correct answer immediately after acting on the stale one. _do_resume re-checks both before opening the modal and in the dismiss callback, resyncs the button, and reports why. The orchestrator check remains the CLI backstop.
• CheckpointManager.read() raised an unhandled AttributeError on a checkpoint file holding valid YAML that is not a mapping. The TUI's bare except Exception had been masking it; narrowing that handler to CheckpointReadError surfaced it. read() now raises CheckpointReadError like it does for every other unreadable file, so a corrupt file can no longer crash the flow grid on startup.

Motivation

Whether a flow can be resumed is derived from its checkpoint file, so the data on disk is always correct. The TUI only read it while first building a screen, which left the display stale for the rest of the session:

• A flow that failed mid-session kept showing no resumable label on its tile until the command was quit and rerun, because cards are built once at compose time and MainScreen is never popped.
• Returning to the flow screen after a failed run did not reveal the Resume button, because FlowScreen is resumed rather than re-mounted. The user had to navigate back to the grid and re-enter the flow so a fresh screen was built.

The result was confusing rather than broken — the resume genuinely worked, nothing in the UI said so.

Follow-up in AI-7066: interrupting a run before any phase completes writes no checkpoint at all, so an interrupted run cannot be told apart from one that never started. Telling those apart needs a run marker written at start; it is not recoverable from what is on disk today.

Review checklist (to be filled by reviewers)

• Feature or bugfix MUST have appropriate tests (unit, integration, e2e)
• Add qa/required if this PR needs QA validation, or qa/skip-qa if it does not. Exactly one of the two is required.
• If you need to backport this PR to another branch, you can add the backport/<branch-name> label to the PR and it will automatically open a backport PR once this one is merged

[luisorofino]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Another round soon, please!

Reviewed commit: 6dca3f56b6

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[datadog-prod-us1-5]

 

🎉 All green!

🧪 All tests passed
❄️ No new flaky tests detected

🎯 Code Coverage (details)
• Patch Coverage: 99.55%
• Overall Coverage: 89.10%

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 6c226de | Docs | Datadog PR Page | Give us feedback!}

[luisorofino]

• Add CHECKPOINTED status to phase nodes to separate it from PENDING #24702
• [AI-7065] Refresh flow resume state in the Togo TUI without restarting the command #24701 👈 (View in Graphite)
• loa/openmetrics-ai-gen

This stack of pull requests is managed by Graphite. Learn more about stacking.

[dd-octo-sts]

Validation Report

All 21 validations passed.

Show details

Validation	Description	Status
agent-reqs	Verify check versions match the Agent requirements file	✅
ci	Validate CI configuration and code coverage settings	✅
codeowners	Validate every integration has a CODEOWNERS entry	✅
config	Validate default configuration files against spec.yaml	✅
dep	Verify dependency pins are consistent and Agent-compatible	✅
http	Validate integrations use the HTTP wrapper correctly	✅
imports	Validate check imports do not use deprecated modules	✅
integration-style	Validate check code style conventions	✅
jmx-metrics	Validate JMX metrics definition files and config	✅
labeler	Validate PR labeler config matches integration directories	✅
legacy-signature	Validate no integration uses the legacy Agent check signature	✅
license-headers	Validate Python files have proper license headers	✅
licenses	Validate third-party license attribution list	✅
metadata	Validate metadata.csv metric definitions	✅
models	Validate configuration data models match spec.yaml	✅
openmetrics	Validate OpenMetrics integrations disable the metric limit	✅
package	Validate Python package metadata and naming	✅
qa-label	Validate the pull request declares whether it needs QA for the next Agent release	✅
readmes	Validate README files have required sections	✅
saved-views	Validate saved view JSON file structure and fields	✅
version	Validate version consistency between package and changelog	✅

View full run