rldyourterm

DEPRECATED. This repository is archived and no longer maintained. It is kept read-only for reference; its GDS anchor records lifecycle: archived under portfolio:archived-projects.

Crash-intolerant AI terminal runtime with v1.0 priorities locked as:

1. stability,
2. AI CLI compatibility (Claude Code, Codex, Gemini CLI),
3. speed.

License

Licensed under the GNU Affero General Public License v3.0 or later (AGPL-3.0-or-later). See LICENSE. Repository governance validates AGPL license consistency in CI and Jenkins.

MVP runtime status

This repository now includes a Rust workspace with an MVP bootstrap path:

• rldyourterm-app: CLI bootstrap + shell resolution + diagnostics events.
• rldyourterm-ui: single-window runtime scaffold with command-driven control hooks.
• scripts/mvp/*: executable compatibility harness for Claude/Codex/Gemini scenarios.

Single-window baseline is explicit and mandatory in v1.0 (window_count=1).

Quick checks

cargo check --locked -p rldyourterm-ui
cargo check --locked -p rldyourterm-app

Benchmarking

terminal_benchmark/ now provides two benchmark suites:

• canonical-headless for deterministic CI-safe coverage across core, services/session, ui, features/settings, features/shell-integration, features/font, features/render-gpu policy helpers, and features/render-cpu
• live-display for real winit window plus wgpu/softbuffer presentation timing on a live display session

The canonical headless JSON report still marks app, foundation, foundation-platform, and features/diagnostics as correctness-only layers. The live-display suite is intentionally local/manual and does not run as a required PR CI gate.

Quick run:

cargo run --locked -p rldyourterm-terminal-benchmark -- --scenario all --scale standard

Structured JSON output:

cargo run --locked -p rldyourterm-terminal-benchmark -- \
  --suite canonical-headless \
  --scenario all \
  --scale stress \
  --format json \
  --output /tmp/rldyourterm-terminal-benchmark.json

Headless CI-parity smoke run:

bash scripts/ci/run_terminal_benchmark_smoke.sh

Full benchmark suite:

bash scripts/ci/run_terminal_benchmark_full.sh

Live-display local smoke run:

bash scripts/ci/run_terminal_display_benchmark_smoke.sh

Live-display local full run:

bash scripts/ci/run_terminal_display_benchmark_full.sh

Controlled live-display calibration run:

bash scripts/ci/run_terminal_display_benchmark_controlled.sh

Controlled live-display calibration run with baseline refresh:

bash scripts/ci/run_terminal_display_benchmark_calibration.sh \
  target/terminal-benchmark/live-display-controlled-report.json \
  terminal_benchmark/baselines/live-display.controlled.json \
  target/terminal-benchmark/live-display-controlled-calibration.json

Use TERMINAL_DISPLAY_BENCHMARK_COMPARISON_MODE=enforced only when you intentionally want the calibrated baseline to become a hard gate.

There is also a manual self-hosted GitHub workflow for the same controlled calibration flow:

• .github/workflows/display-benchmark.yml
• runner labels required: self-hosted, display-benchmark, and the selected OS label
• it uploads four artifacts: runner readiness report, controlled report, controlled baseline, calibration report
• artifact upload is always()-style, so readiness and partial evidence survive failed calibration attempts

Optional threshold validation against versioned baselines:

TERMINAL_BENCHMARK_BASELINE=terminal_benchmark/baselines/canonical-headless.standard.json \
  bash scripts/ci/run_terminal_benchmark_full.sh

TERMINAL_DISPLAY_BENCHMARK_BASELINE=terminal_benchmark/baselines/live-display.quick.json \
  bash scripts/ci/run_terminal_display_benchmark_smoke.sh

Headless baselines are enforced. Live-display baselines are advisory by default: regressions are reported to stderr but do not turn the command into a hard failure unless you deliberately promote that policy.

Benchmark baseline scope is fail-closed:

• portable-headless baselines apply only to canonical headless reports
• local-display-session baselines apply only to generic local live-display reports
• controlled-display-session baselines can only be refreshed from, and applied to, monitor-aware controlled live-display reports
• controlled baselines now also carry exact environment requirements for the calibrated host profile, including display server hint and per-scenario CPU monitor metadata

Canonical local system suite:

bash scripts/ci/run_terminal_system_suite.sh

Canonical local system suite with optional live-display coverage:

bash scripts/ci/run_terminal_system_suite.sh --with-live-display smoke

Canonical local system suite with controlled live-display validation:

bash scripts/ci/run_terminal_system_suite.sh --with-live-display controlled

Canonical local system suite with baselines:

bash scripts/ci/run_terminal_system_suite.sh \
  --benchmark-baseline terminal_benchmark/baselines/canonical-headless.standard.json \
  --with-live-display smoke \
  --live-display-baseline terminal_benchmark/baselines/live-display.quick.json

The system suite emits a machine-readable JSON report and validates the referenced full benchmark report and, when requested, the live-display benchmark report and benchmark baselines before returning success.

CPU live-display reports now include:

• suite-level environment metadata: session_type and display_server_hint
• phase-level timing for redraw_dispatch, which captures the wait between request_redraw and RedrawRequested
• phase-level timing for buffer_acquire, raster, and present
• cpu_buffer_age_counts, which shows how often the softbuffer framebuffer arrives with age=0, age=1, age=2, or age>=3
• pacing_mode and monitor_refresh_rate_millihz, which show whether the live run used monitor-driven cadence or the same event-driven fallback that production uses when monitor timing is unavailable
• monitor_name and monitor_scale_factor, which show whether winit exposed concrete monitor metadata for the run

Controlled live-display validation adds monitor-aware requirements:

• steady-redraw-cpu and resize-cycle-cpu must use pacing_mode=monitor-cadence
• monitor refresh and scale metadata must be present for those scenarios
• optional session/display-server expectations can be provided through TERMINAL_DISPLAY_BENCHMARK_REQUIRED_SESSION_TYPE and TERMINAL_DISPLAY_BENCHMARK_REQUIRED_DISPLAY_SERVER_HINT

That keeps local display regressions explainable before touching production render code. The current CPU display path is explicitly age-aware and replays a two-frame damage history when softbuffer returns age=2.

CI/CD profile

• Current PR-visible check suite includes CI, CodeQL, ClusterFuzzLite PR fuzzing, Scorecard, Semantic PR, and PR Automation.
• CI gate is strict for critical fan-out jobs (check/test/clippy/fmt/msrv/audit/deny must be success); only coverage may be skipped when disabled.
• CodeQL publishes Rust extraction diagnostics artifact and fails closed when diagnostics telemetry is missing/invalid; actionable diagnostics (ExtractionErrors > 0 or non-benign extraction warnings) block merges, while known generated build-output warnings remain observable but non-blocking.
• Release governance remains manual-only (workflow_dispatch) with enforced preflight (terminal system suite, security gates, AI CLI compatibility matrix).
• Weekly non-blocking soak lane: .github/workflows/soak.yml runs scripts/mvp/run_matrix.sh and publishes artifacts for long-run compatibility evidence.

Fuzzing

The repository includes cargo-fuzz targets and ClusterFuzzLite workflows.

Local quick run:

cargo install cargo-fuzz
cargo fuzz run parser_feed -- -max_total_time=30

Interactive Run

Run the binary directly as the terminal runtime (outside the MVP harness):

cargo run -q -p rldyourterm-app -- --mode auto --shell fish --window-count 1

Default non-harness runtime now launches the GUI window terminal path (single-window MVP baseline). If GUI initialization fails, runtime emits a warning and falls back to TTY mode. To force TTY mode explicitly:

cargo run -q -p rldyourterm-app -- --mode auto --shell fish --window-count 1 --tty

Or build once and run the binary directly:

cargo build -p rldyourterm-app
./target/debug/rldyourterm-app --mode auto --shell fish --window-count 1

MVP compatibility harness

MVP harness scenarios continue to run from scripts/mvp/* and do not replace interactive mode.

Required manual release gate (covers R-11, R-12, R-13):

bash scripts/mvp/run_matrix.sh 3

Extended compatibility soak:

bash scripts/mvp/run_matrix.sh 5

Run a single profile with optional extra commands:

bash scripts/mvp/run_profile.sh codex 3 recoverable:pty-write tick mode:cpu

Harness logs are written to scripts/mvp/output/. Each required gate run must produce these evidence lines:

• MVP_PROFILE_START ... single_window_required=1 release_governance=manual-only ...
• MVP_HARNESS ... single_window_required=1 release_governance=manual-only
• MVP_HARNESS_CMD ... --window-count 1 ...
• MVP_RESULT ... windows=1 ... single_window_required=1 single_window_enforced=yes ... release_governance=manual-only ...
• MVP_GATE_RESULT ... status=pass ...
• MVP_PROFILE_RESULT ... status=pass ...
• MVP_MATRIX_RESULT ... status=pass ...