simulate(tui): in-TUI scenario save (no overwrite) + copy-to-clipboard; drop export-scenarios subcommand#867
Open
theomonnom wants to merge 34 commits into
Open
simulate(tui): in-TUI scenario save (no overwrite) + copy-to-clipboard; drop export-scenarios subcommand#867theomonnom wants to merge 34 commits into
theomonnom wants to merge 34 commits into
Conversation
…erwrite), copy scenario to clipboard, print generated scenarios path on exit
…nto scrollActive (shared by keys + wheel)
…n/jobIconStylePtr/jobIcon) into one jobStatusIcon
- Pre-flight: reuse agentfs.CheckSDKVersion (pinned to 1.6.0, the thin-CLI baseline) in startAgent so start/dev/console/simulate fail fast with a clear 'too old, upgrade to ...' instead of a cryptic subprocess error. - On early exit / failed register/connect, surface the agent's OWN output (the real error) plus the full log path - no string-match guessing of the cause. Wired into simulate (CI + TUI) and console. Verified against staging with a broken .venv (shows the ModuleNotFoundError) and an old declared SDK (shows the upgrade message).
…has no agent description (scenario-file runs)
…kers don't survive and hold the port
…gisters but never joins a room
…ingle no-join timeout (which can be cloud connection pacing)
…ough end of log) so tracebacks aren't truncated
…file and surfaced errors stay plain text
The agent logs in colored format for the TUI; the temp log file kept the raw escape sequences.
The header listed every scenario label (bare bullets when labels are empty) and the job list rendered them all again. Replace it with one line: '<check> Loaded N scenarios from <file> <group name>' — the group name only when set.
'Loaded N scenarios from <file> — <name>' renders as the first (already done) step in file mode, alongside Starting agent / Creating simulation, instead of a separate header line.
Log lines were stripped of ANSI codes at scanner ingest, so the TUI rendered everything colorless. Keep raw lines in memory (bubbletea/ lipgloss render ANSI fine) and strip only where plain text is needed: the log file, fatal-marker matching, surfaced error excerpts, and CI stdout.
…e/copy - agentLauncher owns the agent subprocess around the TUI: a fast ctrl+c used to quit before agentStartedMsg was delivered, leaking the worker and breaking the next run with "address already in use" - always save a plain-text run report (the non-TUI output: job results, transcripts, room logs, summary) to a temp file, like the agent log - save scenarios moves to the s key; the prompt is a bordered dialog and copy/save confirmations show as a transient toast instead of a bare footer line - drop the duplicated description hint from the footer (the collapsed block already says "press d to expand")
q during an active run now asks before cancelling the simulation, so a misclick can't kill it. y/enter confirms, any other key dismisses; ctrl+c stays an immediate exit.
The stop-simulation prompt is now a two-button dialog (arrows/tab to select, enter to confirm, esc to dismiss) defaulting to keep running. y still confirms directly.
The report was a snapshot of the final results. Replace it with a runReporter that records the same lines the non-TUI mode prints as the run progresses (setup steps, run id, generating/running/summarizing transitions, job progress), then appends the results section on exit.
simLog is now the single source of every non-TUI output line (setup steps, GitHub group markers, status transitions, results): CI writes it to the terminal and the TUI's runReporter writes the same calls to the report file, so the two can't drift. Exit was slow because Kill gave the worker 5s of SIGINT grace, which a draining worker never meets; and a second ctrl+c during that window killed the CLI before SIGKILL, leaking the worker with its port bound. Shorten the grace to 1s, wait for the post-SIGKILL exit so the port is free before returning, and ignore SIGINT during TUI cleanup.
A second ctrl+c during cleanup now SIGKILLs the worker and exits instead of being ignored, so a slow shutdown can still be broken out of without leaking the worker. writeChatHistory still wrote to stdout after gaining its writer param, so transcripts never reached the report file; also drop the 80-char truncation of tool args/outputs so the report is complete. Rename the CI simLog var (shadowed log/slog), remove the duplicated agent-logs line, and trim comments.
Transliterate output glyphs at the writer level so the report file has no special characters while simLog stays the single source of the strings. Route the Finish trailer through the same writer.
# Conflicts: # cmd/lk/simulate_ci.go # cmd/lk/simulate_tui.go
- staticcheck QF1012: fmt.Fprintf(&b, ...) over WriteString(Sprintf) - drop the gh/plain output split and isGitHubActions gating: group markers are a useful delimiter everywhere, and job ids now appear in the group titles - agent health: tolerate 3 not-joined timeouts, require 2 fatal log markers (a single one can be an isolated job crash), name the marker list pythonFatalMarkers (JS will need its own) - open the log pane automatically when the agent is detected broken - prune obvious comments
rektdeckard
approved these changes
Jun 11, 2026
| } | ||
| if config.mode == modeGenerateFromSource && run != nil { | ||
| if path, err := writeGeneratedScenariosTemp(run); err == nil && path != "" { | ||
| fmt.Fprintf(os.Stderr, "Generated scenarios: %s\n", path) |
Member
There was a problem hiding this comment.
All print statements should use the new Printer interface. There is a global out instance of it with helpers for debug messages (out.Status(...)) and command output (out.Result(...)). This helps us manage output, respect the --quiet flag, and keep random stuff from cluttering stdout.
Member
Author
There was a problem hiding this comment.
Done: all simulate prints now go through out — breadcrumbs via Status/Statusf (silenced by --quiet), results and the gh ::error:: annotations via Resultf, the broken-agent note on out.Err (warning-class, never quieted). For the writer-based streaming (simLog, agent-startup log forwarding) I added Printer.StatusWriter(), which returns io.Discard under --quiet.
All direct stdout/stderr prints in the simulate flows now go through the global Printer: breadcrumbs via Status (silenced by --quiet), results and gh annotations via Result, the broken-agent note on Err. simLog takes out.Out and the new Printer.StatusWriter(), which returns io.Discard under --quiet for writer-based streaming.
WarnWriter and ResultWriter join StatusWriter so streaming code gets the same gating as the print methods, without reaching into the Printer's fields.
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
2 participants
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.
No description provided.