Test and validate

Skill dayvidpham/aura-plugins/.claude/skills/test-and-validate

Use when running tests, validation, CI checks, coverage, race detection, vet, build, or verifying strike-cli changes. Trigger on make test, go test, lint, verify, validate, regression, or before claiming work done.From its SKILL.md

Install

npx -y skills add dayvidpham/aura-plugins --skill test-and-validate

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.4 KB, as published. Nobody here has run it

Test and validate (strike-cli)

Read-only verification skill. Observe and report — do not fix failures here (use built-in /verify or implement fixes under issue-handler when owning a branch).

Single source of truth for gates: root AGENTS.md → Verification tiers. This skill runs those tiers; do not invent softer or harder local suites.

CI mirror (order matters)

Match .github/workflows/ci.yml:

gofmt -l . must be empty
go generate ./internal/tui (TUI flatten; required before build/test if _src changed or generate is stale)
make web-check when web/ is touched or web/package.json exists and UI may be affected
go build ./... or make build
make vet
go test ./... — CI uses go test -race ./... on every PR

Local convenience: make test && make vet && make build after gofmt (+ generate/web when needed).

Risk tiers (pick one per change)

Tier	When	Local gate
A	Docs, skills, comments, markdown-only, no Go/web	`test -z "$(gofmt -l .)"` (skip if no `.go` touched); no full suite required
B	Normal Go/web/TUI code (default)	gofmt → generate if TUI `_src` → `make web-check` if `web/` → `make test && make vet && make build`
C	Trust boundary: `internal/tool`, `permission`, `auth`, `session`, `engine` concurrency/turn loop, `protocol` wire, sandbox/workspace	Tier B + `go test -race ./... -count=1` + focused package tests first

CI still runs race on every PR. Do not pay full local race on Tier A/B unless reproducing a CI failure.

Optional: make cover / make cover-check (soft in CI). Offline product smoke: load skill smoke when user-visible startup/input/session/auth paths change.

Commands

Check	Command
Format	`test -z "$(gofmt -l .)"`
TUI generate	`go generate ./internal/tui`
Web	`make web-check`
Unit suite	`make test` or `go test ./...`
Fresh run	`go test ./... -count=1`
Race	`go test -race ./... -count=1`
Coverage	`make cover` / `make cover-check`
Package focus	`go test ./internal/tool/ -count=1 -v`
Single test	`go test ./internal/permission/ -run TestEvaluate -count=1 -v`
Vet / build	`make vet` / `make build`
Offline boot	`make run-echo`

Required workflow

Discover what changed (git diff --stat, package paths) → assign tier A/B/C.
Run focused package tests for touched packages first (Tier B/C).
Run the tier gate above.
If Tier C or CI-red reproduction: full race suite.
If user-visible CLI/TUI/session/auth: load smoke (or note gap if smoke not run).

Report format

Verdict — pass/fail with tier.
Commands run — exact shell lines.
Failures — verbatim output, never paraphrased.
Gaps — packages or flows not exercised and why.
Flakes — see below; never silent skip.

Flake policy

A flake fails intermittently or only on one OS/env while CI (or 3 local reruns) disagrees.

Capture: platform, command, pass/fail of 3 reruns, snippet.
Open or link a wave: 0 / priority: bugs-first issue with repro.
Do not block unrelated feature merges on a known env-only flake.
Quarantine only with t.Skip("…") + issue link — never delete coverage silently.
Prefer fixing root cause (e.g. EvalSymlinks path compare, mtime granularity) over skip.

Package risk map

Area	Higher risk signals
`internal/tool`	filesystem mutation, shell, sandbox, filestate freshness, workspace roots
`internal/permission`	last-match-wins, always grants, reject cascade
`internal/auth`	credentials mode 0600, env precedence, OAuth
`internal/session` + `protocol`	transcript integrity, replay
`internal/provider`	HTTP/SSE, cancellation
`internal/engine`	turn state machine, tool loop, prune/compaction, interrupt
`internal/tui`	Bubble Tea update loops (existing harnesses)
`internal/history`	concurrency + path security

Rules

Never edit source or tests to make verification green.
Never skip a failing package without marking it FAIL with output.
Prefer project Make targets over inventing new scripts.
Prefer tiered gates over always-max ceremony.