Ejecuta cualquier Skill en Manus
con un clic

Ejecuta cualquier Skill en Manus con un clic

hil-regression

Estrellas2

Forks1

Actualizado12 de julio de 2026 a las 22:40

Diagnose a HIL CI failure on real WS63/BS2X silicon — compare the failed CI run against the last passing run, identify whether the regression is caused by a code change, toolchain update, or rig issue, and output a triage report. Use when a HIL CI job fails on the main branch or a PR.

Instalación

Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.

Ejecutar en Manus

Fuente

hispark-rs

hispark-rs/hisi-riscv-rs

Abrir repositorio de GitHub Ver repositorios del creador

Descarga

Ejecutar en Manus

Ocupaciones relacionadasSOC

Basado en la clasificación ocupacional SOC

Analistas de garantía de calidad de software y probadoresOcupaciones informáticas y matemáticas·SOC 15-1253

Explorador de archivos

2 archivos

SKILL.md

readonly

name	hil-regression
description	Diagnose a HIL CI failure on real WS63/BS2X silicon — compare the failed CI run against the last passing run, identify whether the regression is caused by a code change, toolchain update, or rig issue, and output a triage report. Use when a HIL CI job fails on the main branch or a PR.
disable-model-invocation	true

HIL Regression Diagnosis Skill

Diagnoses HIL CI failures on real silicon. Run when a hil.yml CI job fails on a commit that should be safe (e.g. a docs-only change fails HIL → rig issue; a DMA change fails HIL → likely code regression).

User-invoked (/hil-regression). It reads CI logs, does NOT flash hardware.

Usage

# Diagnose the latest failed HIL CI run (autodetects run ID):
bash .agents/skills/hil-regression/regression.sh

# Diagnose a specific run:
bash .agents/skills/hil-regression/regression.sh --run 28502888903

# Show pass/fail history for the last 30 days:
bash .agents/skills/hil-regression/regression.sh --history

Diagnosis methodology

Phase 1 — Gather data

# 1. Find the latest failed HIL CI run:
gh run list --workflow hil.yml --limit 5 --json databaseId,conclusion,headBranch,displayTitle

# 2. Fetch the failed job's logs:
gh run view <run-id> --log --job <job-name> > /tmp/hil-fail.log

# 3. Find the last passing run on the same branch:
gh run list --workflow hil.yml --branch main --limit 10 --json databaseId,conclusion
gh run view <passing-run-id> --log > /tmp/hil-pass.log

# 4. Check what code changed between the two runs:
git log --oneline <passing-sha>..<failing-sha>
git diff --stat <passing-sha>..<failing-sha>

Phase 2 — Classify failure

Symptom	Likely cause	Action
No UART output at all	Boot hang: wrong loaderboot, flash address, or power	Check `ADDRESS` and `LOADERBOOT` in the CI config. Compare with hil-smoke --preflight.
Garbled UART	Baud mismatch: clock-tree change affects UART divider	Check `TIMER_CLOCK_HZ` / UART clock source in the diff.
Specific test FAILED	Code regression in the exercised driver	`git diff <passing-sha>..<failing-sha> -- crates/hisi-hal/src/<driver>.rs`
All tests FAIL (incl. ones not touched)	Rig issue: board disconnected, probe-rs failed, power cycled	Check the CI runner logs for hardware errors, USB resets, timeout on `probe-rs run`.
Intermittent (passes locally)	Board contact / wiring / timing edge	Re-run HIL CI (`gh run rerun <run-id>`). If 2/3 passes → wiring flaky, not code.

Phase 3 — Produce report

# HIL Regression Report

## Commit
<failing-sha> by <author> on <date>
<commit-message>

## CI Run
<url-to-run>
<workflow> / <job> — <conclusion>

## Diff vs last passing
<files changed, summary>

## Diagnosis
**<RIG / CODE / TOOLCHAIN / INTERMITTENT>**

<detailed evidence>

## Next step
- If RIG: <what to check on the board / CI runner>
- If CODE: <which file/line is suspect, what to fix>
- If TOOLCHAIN: <pin toolchain version, revert to known-good>
- If INTERMITTENT: <re-run CI, add retry logic if flaky>

Más de este repositorio

mismo repositorio

pac-svd-register-access

hispark-rs/hisi-riscv-rs

Maintain hisi-hal register access through SVD/PAC truth sources. Use when changing HAL MMIO/register code, reviewing raw PAC usage, fixing check-register-access CI failures, adding or correcting SVD/PAC fields/access types, regenerating ws63-pac or bs2x-pac, deciding whether a register should be modeled in SVD versus handled in HAL, or auditing read/modify/write behavior for write-only or dynamically masked registers.

2026-07-132

submodule-commit

hispark-rs/hisi-riscv-rs

Commit + push changes that span ws63-rs git submodules in dependency order, then bump the parent repo's submodule pointers. Use when you've edited files inside one or more submodules (ws63-pac/hal/rt/examples/guide, chips/ws63/rf/ws63-RF, ws63-pac/ws63-svd) and need to land them correctly.

2026-07-132

typed-config

hispark-rs/hisi-riscv-rs

Make a hisi-hal driver's CONFIG surface "if it compiles, it runs on silicon" — design, review, or tighten a Config / new() / configure() so a value you can WRITE is a value that RUNS, with no parameter that compiles but is silently clamped, truncated, left without a clock, or dependent on an unenforced precondition. Use when adding a driver, reviewing/PR-ing a config API, tightening one for the 0.5.0 config pass, or diagnosing a "compiles but wrong/dead on hardware" bug. Applies the project's primary API convention (see root Agent Instructions "Typed config").

2026-07-132

embedded-test-hil

hispark-rs/hisi-riscv-rs

Create, modify, audit, build, or run on-target embedded-test HIL tests for hisi-hal and tests-hil. Use when adding HAL driver silicon evidence, writing crates/hisi-hal/tests/hil/*.rs cases, registering

2026-07-122

hil-smoke

hispark-rs/hisi-riscv-rs

Build a ws63-rs example for a HiSilicon chip, flash it to a REAL board via the repo HIL flash path, read UART, and assert the example marker — the silicon twin of qemu-smoke. Use for example-level hardware smoke; HAL driver-level evidence uses embedded-test via hil/embedded-test-runner.sh instead.

2026-07-122

qemu-smoke

hispark-rs/hisi-riscv-rs

Build a ws63-rs example for a HiSilicon chip (ws63/bs21/bs21e/bs22/bs20) and boot it in the hisi-riscv-qemu fork, asserting the expected UART banner / GPIO toggle / embassy interleave / IRQ delivery. Use to validate firmware end-to-end in emulation, or after touching a driver/example/the QEMU model.

2026-07-122

name	hil-regression
description	Diagnose a HIL CI failure on real WS63/BS2X silicon — compare the failed CI run against the last passing run, identify whether the regression is caused by a code change, toolchain update, or rig issue, and output a triage report. Use when a HIL CI job fails on the main branch or a PR.
disable-model-invocation	true

HIL Regression Diagnosis Skill

User-invoked (/hil-regression). It reads CI logs, does NOT flash hardware.

Usage

# Diagnose the latest failed HIL CI run (autodetects run ID):
bash .agents/skills/hil-regression/regression.sh

# Diagnose a specific run:
bash .agents/skills/hil-regression/regression.sh --run 28502888903

# Show pass/fail history for the last 30 days:
bash .agents/skills/hil-regression/regression.sh --history

Diagnosis methodology

Phase 1 — Gather data

# 1. Find the latest failed HIL CI run:
gh run list --workflow hil.yml --limit 5 --json databaseId,conclusion,headBranch,displayTitle

# 2. Fetch the failed job's logs:
gh run view <run-id> --log --job <job-name> > /tmp/hil-fail.log

# 3. Find the last passing run on the same branch:
gh run list --workflow hil.yml --branch main --limit 10 --json databaseId,conclusion
gh run view <passing-run-id> --log > /tmp/hil-pass.log

# 4. Check what code changed between the two runs:
git log --oneline <passing-sha>..<failing-sha>
git diff --stat <passing-sha>..<failing-sha>

Phase 2 — Classify failure

Symptom	Likely cause	Action
No UART output at all	Boot hang: wrong loaderboot, flash address, or power	Check `ADDRESS` and `LOADERBOOT` in the CI config. Compare with hil-smoke --preflight.
Garbled UART	Baud mismatch: clock-tree change affects UART divider	Check `TIMER_CLOCK_HZ` / UART clock source in the diff.
Specific test FAILED	Code regression in the exercised driver	`git diff <passing-sha>..<failing-sha> -- crates/hisi-hal/src/<driver>.rs`
All tests FAIL (incl. ones not touched)	Rig issue: board disconnected, probe-rs failed, power cycled	Check the CI runner logs for hardware errors, USB resets, timeout on `probe-rs run`.
Intermittent (passes locally)	Board contact / wiring / timing edge	Re-run HIL CI (`gh run rerun <run-id>`). If 2/3 passes → wiring flaky, not code.

Phase 3 — Produce report

# HIL Regression Report

## Commit
<failing-sha> by <author> on <date>
<commit-message>

## CI Run
<url-to-run>
<workflow> / <job> — <conclusion>

## Diff vs last passing
<files changed, summary>

## Diagnosis
**<RIG / CODE / TOOLCHAIN / INTERMITTENT>**

<detailed evidence>

## Next step
- If RIG: <what to check on the board / CI runner>
- If CODE: <which file/line is suspect, what to fix>
- If TOOLCHAIN: <pin toolchain version, revert to known-good>
- If INTERMITTENT: <re-run CI, add retry logic if flaky>