LoopTroop Files

Summary

• Added live-only workspace status progress labels for coding, coverage checks, and retried phases.
• Made merged PR completion independent of the user's local checkout cleanliness.
• Fixed long ticket descriptions being unscrollable in backlog and detail views.
• Enabled setup-scoped online artifact lookup for missing launcher provisioning while keeping real provisioning evidence stricter.
• Fixed realtime logs for active multi-attempt phases while keeping archived phase versions static and attempt-scoped.
• Hardened execution setup so missing launcher failures must show persistent provisioning attempts before blocking.
• Hardened setup reuse so final testing automatically runs through validated prepared environments.
• Improved workspace setup approval logs so generation starts visibly expanded and collapses once the setup plan is ready.
• Added a one-step rewind from Preparing Workspace Runtime back to setup-plan approval for safe edits and regeneration.
• Fixed runtime setup rewinds so regeneration starts only the requested setup-plan session instead of racing an automatic draft.
• Made future bead commits language-agnostic while blocking setup-created project dirt before coding starts.
• Consolidated concurrent AI status messages and warnings into a premium collapsible activity strip below workspace logs with persistence.
• Fixed stale Current Activity timeout warnings when reviewing phases that already passed.
• Made PRD coverage revisions tolerate safe change-metadata aliases while preserving validated structural review diffs.
• Added tooltip to the CMD log tab in phase and full log views, consistent with all other log tabs.
• Raw tab in artifact views now defaults to the validated variant (or first accepted attempt) instead of the first non-disabled variant. When multiple model sources exist, the first validated model is auto-selected.
• Added dynamically calculated bead implementation time below Completed At in CodingView's bead details panel.
• Improved stall diagnostics with trend-wide process attribution and focused ticket runtime artifact sizing.
• Fixed completed ticket navigation so the historical Implementing row shows the final bead count.

Detailed Changes

Added

• Added trend-wide top system CPU/RSS/read/write attribution and a --ticket-path option to npm run diagnose:stall, so reports can identify memory/I/O spike owners and inspect a specific ticket runtime's logs, largest directories, and large build artifacts.
• Added dynamically calculated bead implementation time (Completed At minus Started At) under the Timeline section in the CodingView's bead details panel, rendered on the fly when both timestamps are available and represent a valid positive duration.
• Added live-only progress wording to the workspace status title: CODING now names the active bead and iteration, coverage phases show pass/version progress when known, and manually retried non-implementation phases show the active retry attempt while the status is live.
• Raw artifact tab now defaults to the validated variant when available. When multiple model sources exist (e.g. council drafts, votes), the first source with a validated variant is auto-selected. Falls back to the first non-disabled variant when no validated variant exists.
• Added an Initial Prompt selector to future model-produced Raw attempt views when the first prompt is persisted, making the model input inspectable beside accepted and rejected attempt outputs without inferring legacy prompts from logs.
• Added a guarded setup-plan rewind while PREPARING_EXECUTION_ENV is active, letting users edit or regenerate the approved setup plan after runtime setup starts while preserving archived runtime evidence.
• Added setup-scoped OpenCode websearch/webfetch access during PREPARING_EXECUTION_ENV, with managed dev OpenCode servers started using OPENCODE_ENABLE_EXA=1, so agents can resolve official launcher artifact metadata when local repository evidence is insufficient.
• Added tool_requirements.provisioning_attempts evidence to execution setup profiles so failed required-launcher setup records distinct temp-root provisioning strategies and commands, or why no safe provisioning path exists.
• Added tooling_probe_commands to execution setup profiles so prepared language/toolchain environments can be verified by model-selected, non-mutating probes before coding starts.
• Added pre-flight and execution-setup worktree cleanliness checks: pre-existing committable project changes now block before setup, ready setup attempts fail if they leave committable project changes, and untracked generated/local noise is reported with .gitignore suggestions.
• Added support for tracking multiple parallel AI model activities and diagnostics concurrently.
• Added state persistence to localStorage for the collapsible activity strip, saving user preferences across reloads.
• Added a "Create and Start" button to the New Ticket modal, letting users create a ticket and immediately trigger the workflow from a single action.

Fixed

• Completed and later-phase tickets now keep the last known bead progress on the left-panel Implementing timeline row, showing labels such as Implementing (Bead 8/8) instead of the generic Bead ?/? fallback.
• Runtime setup rewinds now suppress the restored approval actor's immediate auto-draft when the route is about to save an edited plan or start a commented regeneration, preventing duplicate setup-plan sessions and last-writer-wins overwrites.
• The CMD log tab now shows a descriptive tooltip on hover, matching the tooltip behavior of all other log tabs (ALL, SYS, AI, ERROR, DEBUG).
• Long ticket descriptions are now independently scrollable (300px max-height with overflow) in the Ticket Details modal, DraftView, and PhaseReviewView. Previously, long descriptions expanded indefinitely and pushed other content off-screen.
• Viewing the active (latest) version in multi-attempt phases now keeps using realtime SSE logs while filtering to that active phaseAttempt. Archived phase versions remain static/read-only and load their durable attempt-scoped log snapshot, preventing both missed live runtime setup logs and cross-attempt log mixing.
• Current Activity timeout warnings now render only for the ticket's live status, so revisiting an older phase cannot show an obsolete Approaching timeout banner until refresh.

Changed

• PR merge completion now verifies that origin/<baseBranch> contains the candidate commit and proceeds to cleanup without checking out, fast-forwarding, or requiring cleanliness in the user's main project folder.
• Renamed Raw retry attempt selectors to the clearer Attempt N Output - Accepted/Rejected format.
• Setup-plan edits and regenerations from active runtime setup now stop the runtime session, archive the approved setup contract and runtime attempt, clear stale runtime profile outputs while preserving the tool cache, and require approval again before setup reruns.
• Execution setup prompts now state that wrapper creation, cache inspection, PATH edits, and version probes are discovery/scaffolding only, and do not count as provisioning strategies unless the attempt actually obtains, installs, or activates the missing launcher under approved temp roots.
• Execution setup prompts now treat failed launcher version/info probes as discovery only, replacing single-ecosystem provisioning guidance with non-exhaustive Node, Python, and JavaScript-runtime examples plus permission to use any safe repository-appropriate temp-root provisioning approach.
• Execution setup now requires at least two distinct failed provisioning strategies before accepting terminal failed-tooling evidence, and grants a small bounded persistence retry extension when the base setup budget is exhausted after only one real strategy.
• Final-test command execution now reuses a validated execution-setup wrapper automatically when one is declared, recording both the original command and the effective wrapped command in the report.
• Workspace setup approval now keeps the live log drawer expanded while LoopTroop is generating the setup plan, then automatically collapses it when the plan artifact becomes visible for review.
• Execution setup now fails early when declared wrappers are missing or unusable, probes fail, or wrapper/project command families are declared without probes.
• Changed bead commit capture to be language-agnostic. LoopTroop now commits Git-visible project changes regardless of extension, while still excluding .ticket/**, .looptroop/**, execution-setup roots, and untracked generated/local outputs.
• Execution setup prompts now ask agents to record suggested .gitignore entries in profile cautions instead of editing .gitignore during setup.
• Re-engineered the Current Activity strip below the log tabs to be fully collapsible and display a list of all active model sessions, beads, and warnings.
• Redesigned the strip UI with premium unified severity styling (Error > Warning > Info) and smooth chevron transitions on toggle.