| name | test |
| description | Run unit or revdep tests. |
| user-invocable | true |
| allowed-tools | ["Bash","Read"] |
Test
Verify an MR's changes by running the appropriate tests — either local unit
tests or a downstream pipeline in a consumer/revdep repo.
When this fires
- User says "test", "test this", "run tests", "verify this MR"
- User says "test downstream", "check in test.hac", "trigger revdep pipeline"
- After implementing a fix, before declaring it ready to merge
Procedure
1. Determine test strategy
Inspect the repo to decide which testing path applies:
| Signal | Strategy |
|---|
Repo has tests/ or testthat/ or a test framework | Unit tests — run locally |
| Repo is a CI template provider (like HACtions) with no local tests | Downstream — trigger pipeline in a revdep |
| Both exist | Run local tests first, then downstream if relevant |
For HACtions specifically: there are no local unit tests. The test
strategy is to trigger a pipeline in a revdep like test.hac (project 1611)
that exercises the templates.
2. Isolate credentials before testing external writes
Before running a script whose success path can mutate an external service (uploads, releases, deployments, secret rotation), inspect every ambient authentication source it can inherit: environment variables, process options, credential helpers, and keyrings.
Omitting the script's explicit login call is not isolation; an earlier session may already have populated one of those sources.
- If the user reserved the authenticated action for themselves, run only parse/lint and credential-free preflight paths.
Start a clean process that explicitly clears the script's documented auth variables and in-process options, and prove it stops before the network write.
- If end-to-end testing is necessary, get explicit authorization and use a disposable target or a documented dry-run mode.
Verify the remote state before and after.
- Never treat an unchanged remote version as permission to have contacted the service; disclose any accidental authenticated request immediately.
3. Local unit tests (if applicable)
Rscript -e 'devtools::test()'
pytest
npm test
for f in scripts/*.sh scripts/lib/*.sh; do bash -n "$f"; done
Report pass/fail. If tests fail, investigate and fix before proceeding.
Always report SKIP alongside PASS and FAIL
A pass and fail count says nothing about the tests that never executed, so
"43 pass, 0 fail" can be exactly accurate while the suite has not been
verified.
Report all three numbers, and treat a non-zero skip count as an unmeasured
population: say what was skipped and why before citing the run as evidence.
Then ask what made them skip.
A skip is usually keyed on a missing optional dependency
(skip_if_not_installed(), pytest.importorskip), so an invocation that
bypasses environment setup silently shrinks the suite while leaving the
reported numbers looking healthy.
In an R package, R_PROFILE_USER=/dev/null skips renv/activate.R and
therefore the project library, so packages the tests gate on are absent and
their tests skip.
The same applies to any --no-config flag, a bare interpreter, or a
container built without the optional extras.
Before citing a run as verification, confirm it ran in the project's real
environment.
Re-run without the bypass when the skip count is non-zero.
4. Downstream / revdep testing
When the change is to shared infrastructure (CI templates, shared scripts),
test it in a consumer repo.
a. Identify the test bed
- Check
REVDEPS.md for consumer repos
- Prefer a dedicated test bed (e.g.,
test.hac, project 1611) over
production repos
- If no test bed exists, pick the simplest consumer repo
b. Trigger a pipeline on the consumer repo
The consumer must already reference the MR's branch (or the floating tag
must have been slid). Two approaches:
Option A — Trigger pipeline via API (preferred for template repos):
glab api --method POST "projects/<CONSUMER_PROJECT_ID>/pipeline" \
-f "ref=main" 2>&1 | cat
Option B — If the consumer needs to reference the MR branch:
Create a temporary MR on the consumer that points include: ref: to the
feature branch, or use a CI variable override if the consumer supports it.
c. Wait for and check results
glab api "projects/<CONSUMER_PROJECT_ID>/pipelines/<PIPELINE_ID>" \
--jq '.status' 2>&1 | cat
Poll every 30–60 seconds until status is success, failed, or canceled.
5. Report results
Summarize what was tested and the outcome:
Always include the skip count for a unit-test run, per step 3.
✅ Tests passed:
- devtools::test(): 40 pass, 0 fail, 0 skip
- bash -n syntax check: all scripts OK
- test.hac pipeline #NNNN: success (jobs: lint ✓, check-package ✓, claude-review ✓)
— or —
❌ Tests failed:
- test.hac pipeline #NNNN: failed
- Failing job: lint (exit code 1)
- Error: <summary>
If tests fail, investigate the failure and either fix it or report the issue
to the user.
Edge cases
- No test infrastructure at all: Fall back to
bash -n syntax checks
for shell scripts, or a dry-run/lint of whatever the repo contains.
- Consumer repo not on job-token allowlist: The API trigger will fail
with 403. Inform the user they need to add the consumer to the allowlist
(Settings → CI/CD → Job Token Permissions).
- Floating tag already slid: If
v2 was already moved to include the
MR's changes, a simple pipeline trigger on the consumer's main will
pick up the new templates automatically.
- MR not yet merged: The consumer can't see the branch unless it's in
the same project or group. For cross-project testing pre-merge, the
consumer needs a temporary
include: ref: <branch> override.
Known test beds
| Test bed | Project ID | What it exercises |
|---|
| test.hac | 1611 | HACtions CI templates (lint, check-package, claude-review, etc.) |
The row above is the repository owner's own test bed; project ID 1611 is specific to
that GitLab instance. Replace this table with your own downstream revdep
repos and their project IDs.
Anti-patterns
- Don't skip testing because "it's just a template change" — template
changes can break every consumer.
- Don't declare an MR ready to merge without at least one successful
downstream pipeline if the change touches shared CI infrastructure.
- Don't trigger pipelines on production consumer repos if a test bed exists.
- Don't test an upload or deployment script in a process that may inherit ambient credentials when the user reserved the external write for themselves.