agentleFS
Harry-kpHarry-kp/vortix/.claude/skills/release-qa/SKILL.md

release-qa

Skill757 starsChanged 48 days agoMITView on GitHub

Pre-release QA for vortix — run the P0 release gate (CI gate, smoke set, risk pass from the diff since the last tag, exploratory pass) live in the TUI and CLI on macOS and the Linux lab, fix what breaks with a regression test per finding, log results in P0.md, and end with a go/no-go report. Use whenever the user is about to cut a release, asks to "test everything before release", wants manual or QA-style testing of the live TUI, or asks for release confidence, even if they don't say "P0" or "release-qa".

6 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

12 rules. 0 are common, 5 are distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Check the harness: tmux list-windows -t vxrun on the Mac (ask the user to create it if it is missing), and ssh to the lab.

    line 22
  2. Distinctive:

    Create one task per pass and per P0 area so the user can follow along.

    line 27
  3. Very specific:

    Run sudo scripts/p0.sh on both machines (the smoke set.

    line 44
  4. Very specific:

    Keep a running results table in target/qa-results.md (ID, check, macOS, Linux, note), updated after each area rather than at the end.

    line 52
  5. Distinctive:

    Follow P0.md's "Safety rules" and "Harness" literally.

    line 56
  6. Distinctive:

    After a rebuild, restart the TUI: before retesting. An old binary still running is the usual reason a fix "didn't work".

    line 61
  7. Very specific:

    Frames: use cut -c… to read one panel, and grep -a because frames contain braille, which grep treats as binary.

    line 66
  8. Distinctive:

    Treat every finding as a bug fix, on the release branch:

    line 71
  9. Very specific:

    Write the test that fails for the reported reason where the behaviour is owned (control/, app/tests.rs, render tests.

    line 76
  10. Very specific:

    After the release is published, run P0_TAG=vX.Y.Z P0_SMOKE=0 scripts/p0-vms.sh on the lab: the journey installs the new release through …

    line 91
  11. Distinctive:

    Update P0.md's results log with this run: one row per area, the fixed findings, and the open ones.

    line 99
  12. Very specific:

    Run scripts/ci-local.sh (the full run, with release build) and the lab tests again on the final branch.

    line 102

“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Release QA

The user wants high confidence that the next release works for a real person on both platforms. docs/manual-testing/P0.md is the plan and owns every workflow, pass signal and safety rule. This skill is how to run it. Read CLAUDE.md and P0.md's "Strategy" and "Safety rules" before touching a machine.

0. Before anything

  • Bring up the test VPN servers and their profiles: the scripts/vpn-lab.sh ensure command in docs/live-testing.md "Test VPN servers". Every live pass below needs them.
  • Note the state you must restore on both machines: the kill-switch mode (vortix killswitch), the theme (grep theme ~/.config/vortix/config.toml), and whatever tunnels are up (normally none).
  • Check the harness: tmux list-windows -t vxrun on the Mac (ask the user to create it if it is missing), and ssh to the lab. Never run sudo on the Mac.
  • git describe --tags --abbrev=0 gives the last tag. Then git diff --stat <tag>..HEAD -- crates/ plus P0.md's risk map tell you which areas pass 3 covers.
  • Create one task per pass and per P0 area so the user can follow along.

1. Gate

scripts/ci-local.sh --quick on the Mac. On the lab: sync the branch, then umask 022 && cargo build -p vortix && cargo test -p vortix (the fix-bug skill has the exact lab() function and git apply sync). Stop on red.

Release QA tests the binaries; the release workflow also publishes them, and a job that runs only on a tag fails only on release day (0.5.3 never reached npm that way). For every publish job whose code changed since the last tag (git diff --stat <tag>..HEAD -- .github/ scripts/), its newest run on main must be green: npm-publish.yml and linux-packages.yml dry-run weekly and on their PRs (gh workflow run either one to get a fresh run). A change to the cargo-dist builds (dist-workspace.toml, release.yml) has no dry-run: say so in the report.

2–4. Smoke, risk, explore

Run sudo scripts/p0.sh on both machines (the smoke set; on the Mac, in the root pane), and scripts/p0-vms.sh on the lab (P0.md "Distro VMs": the smoke set and a new user's install-to-uninstall journey on Arch, CachyOS and Fedora). Then walk the risk-mapped workflows, then about 15 minutes per machine trying to break what changed. Run the whole file when the engine, a firewall or DNS backend, or a protocol layer changed broadly.

Keep a running results table in target/qa-results.md (ID, check, macOS, Linux, note), updated after each area rather than at the end. That way a lost context or an interruption loses nothing.

Follow P0.md's "Safety rules" and "Harness" literally. Beyond those:

  • Write multi-step probes as a bash script in $CLAUDE_JOB_DIR/tmp and run it in the root pane, with the restore in a trap … EXIT. A script is one approval, one blocked window and one guaranteed cleanup.
  • After a rebuild, restart the TUI before retesting. An old binary still running is the usual reason a fix "didn't work".
  • Read evidence from logs as well as frames: ~/.config/vortix/logs/ (by date) and ~/.config/vortix/run/*.log (OpenVPN). Count secret-shaped matches, never print them, and redact public IPs in anything you report.
  • Frames: use cut -c… to read one panel, and grep -a because frames contain braille, which grep treats as binary.

When something fails

Treat every finding as a bug fix, on the release branch:

  1. Decide whether it is real: rerun it, and check that the fixture and probe can actually fail. A vacuous or wrong probe is a harness bug; fix the step in P0.md instead.
  2. Write the test that fails for the reported reason where the behaviour is owned (control/, app/tests.rs, render tests; see fix-bug step 4). Confirm it is red.
  3. Fix at the root. If the TUI and CLI disagree, one of them kept its own copy of a rule the engine owns: delete the copy and read the snapshot (P0-39).
  4. Green test, then cargo clippy --workspace --all-targets -- -D warnings, then re-verify live on both machines. Run the mandatory ponytail:ponytail-review and docs-review on the diff, then make one fix: commit per finding.
  5. Log it: FAIL→fixed in the results table, plus the one-line cause.

Anything you decide not to fix (by design, or needing a product call) goes in the report as open, with its user-visible effect. It is never silently dropped.

Finish

  1. After the release is published, run P0_TAG=vX.Y.Z P0_SMOKE=0 scripts/p0-vms.sh on the lab: the journey installs the new release through every channel on each distro. A failure there is a user-facing bug in the release; fix it in a patch release.
  2. Restore both machines to the state noted in step 0 and prove it: no tunnels, the kill-switch mode, empty pf anchor / nft table, the theme. Then stop the droplet's billing with scripts/vpn-lab.sh down --yes; the post-release check makes a new one.
  3. Update P0.md's results log with this run: one row per area, the fixed findings, and the open ones. Fix any workflow whose expectation turned out stale. Add a workflow only for something no automated test can catch.
  4. Run scripts/ci-local.sh (the full run, with release build) and the lab tests again on the final branch.
  5. Push once, open one PR (one commit per finding plus the P0.md update), watch gh pr checks, and rebase-merge only when every row is pass or skipping. Push and merge need the user's go-ahead unless they already gave it for this run.
  6. Report to the user in short points:
    • Verdict: go or no-go.
    • Covered: what was tested on each machine.
    • Fixed: each finding, with its commit.
    • Open: what remains and its impact.
    • Not run: anything skipped, and why.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.