Test Engineering
Repository continuity composition
For repository-scoped work, compose maintain-repository-continuity in
Resume mode before selecting work. After an authorized repository change
passes domain validation, compose Refresh and Verify immediately before
presenting the pull request, and include the reconciled root CONTINUITY.md in
the same change. A policy-permitted no-change or exemption result must be
documented instead of fabricating an edit.
- Contribute: Behavior under test, deterministic coverage, exact commands and outcomes, remaining gaps, and residual risk
- Never claim: That tests prove behavior, environments, or integrations they did not exercise
Purpose
Increase confidence with deterministic tests that cover meaningful behavior,
risk, and boundaries without weakening assertions or overfitting to
implementation trivia.
Required Inputs
Resolve as much of the following as possible:
- behavior or contract at risk
- failure impact and the boundary under test
- existing coverage and why it is insufficient
- repository test conventions, infrastructure, and validation commands
- sources of nondeterminism such as time, randomness, network, or environment
Workflow
model the risk and choose the lowest sufficient test layer
design deterministic fixtures, seams, and assertions using:
./references/determinism-checklist.md
./templates/TEST_PLAN.template.md
assert observable behavior and meaningful state transitions
add success, failure, boundary, and recovery coverage proportional to risk
write a regression test when addressing a defect
validate with the smallest focused command first, then broader required checks
report covered behavior, remaining risk, and any required follow-up testability work
Constraints
- Do not chase coverage percentages by weakening assertions.
- Do not paper over flakiness with sleeps, retries, or disabled tests.
- Do not test implementation trivia when observable behavior suffices.
- Do not duplicate the same assertion across many layers without added value.
- Do not update expected output blindly after a failing test.
Completion Criteria
Provenance
This canonical skill is first-party Ego Hygiene content curated from the staged
candidate at .staging/skills/test-engineering/SKILL.md.
Source Delta
- Adopted: the staged focus on risk modeling, layer selection, deterministic
design, and anti-flakiness constraints.
- Rewritten: canonical metadata, deterministic-testing resources, and explicit
update/regression eval coverage.
- Rejected: the narrower
breakdown-test and pytest-coverage overlaps as
canonical identities because the core workflow must stay general-purpose and
behavior-focused rather than framework- or coverage-metric-centric.
1---2name: test-engineering3description: Designs, implements, repairs, or reviews deterministic automated tests that protect meaningful behavior without chasing superficial coverage. Use when confidence is insufficient and stronger regression or validation evidence is needed.4license: MIT5---67# Test Engineering89<!-- aether-continuity-disposition: reader-writer -->1011## Repository continuity composition1213For repository-scoped work, compose `maintain-repository-continuity` in14**Resume** mode before selecting work. After an authorized repository change15passes domain validation, compose **Refresh** and **Verify** immediately before16presenting the pull request, and include the reconciled root `CONTINUITY.md` in17the same change. A policy-permitted no-change or exemption result must be18documented instead of fabricating an edit.1920- **Contribute:** Behavior under test, deterministic coverage, exact commands and outcomes, remaining gaps, and residual risk21- **Never claim:** That tests prove behavior, environments, or integrations they did not exercise2223## Purpose2425Increase confidence with deterministic tests that cover meaningful behavior,26risk, and boundaries without weakening assertions or overfitting to27implementation trivia.2829## Required Inputs3031Resolve as much of the following as possible:3233- behavior or contract at risk34- failure impact and the boundary under test35- existing coverage and why it is insufficient36- repository test conventions, infrastructure, and validation commands37- sources of nondeterminism such as time, randomness, network, or environment3839## Workflow40411. model the risk and choose the lowest sufficient test layer422. design deterministic fixtures, seams, and assertions using:4344 - `./references/determinism-checklist.md`45 - `./templates/TEST_PLAN.template.md`46473. assert observable behavior and meaningful state transitions484. add success, failure, boundary, and recovery coverage proportional to risk495. write a regression test when addressing a defect506. validate with the smallest focused command first, then broader required checks517. report covered behavior, remaining risk, and any required follow-up testability work5253## Constraints5455- Do not chase coverage percentages by weakening assertions.56- Do not paper over flakiness with sleeps, retries, or disabled tests.57- Do not test implementation trivia when observable behavior suffices.58- Do not duplicate the same assertion across many layers without added value.59- Do not update expected output blindly after a failing test.6061## Completion Criteria6263- [ ] The chosen test layer is justified by risk and cost.64- [ ] Tests are deterministic and assert meaningful behavior.65- [ ] Important success, failure, or boundary paths are covered proportionally.66- [ ] Validation commands and outcomes are explicit.67- [ ] Remaining risk and follow-up seams are visible.6869## Provenance7071This canonical skill is first-party Ego Hygiene content curated from the staged72candidate at `.staging/skills/test-engineering/SKILL.md`.7374## Source Delta7576- Adopted: the staged focus on risk modeling, layer selection, deterministic77 design, and anti-flakiness constraints.78- Rewritten: canonical metadata, deterministic-testing resources, and explicit79 update/regression eval coverage.80- Rejected: the narrower `breakdown-test` and `pytest-coverage` overlaps as81 canonical identities because the core workflow must stay general-purpose and82 behavior-focused rather than framework- or coverage-metric-centric.