A release gate is useful only when it says which decision it controls, what evidence is required, and what happens when evidence is missing or invalid. The Quality Gate Design Skill turns those rules into a reviewable contract.
When this Skill helps
Use it to define or assess entry criteria, evidence requirements, owners, exception paths, and rollback conditions for a delivery or release gate. It reviews supplied policy and evidence design; it does not execute a pipeline, approve an exception, or decide that a release is safe.
Build the gate contract
First define one decision per gate: merge, deploy, expand traffic, or stop. List each required obligation and the evidence version and freshness needed to satisfy it. Assign an operator and keep the states distinct:
| State | Meaning for the decision |
|---|---|
| PASS | Fresh, valid evidence satisfies the stated obligation. |
| FAIL | Evidence shows that an obligation was not met. |
| BLOCKED | Execution could not proceed; name the cause and next attempt. |
| UNKNOWN | Evidence is absent, stale, conflicting, or inconclusive. |
Absence never means PASS. Apply the project’s aggregation and fail-closed rules; do not let an average erase a required failure or unknown. Where policy is missing, record the decision needed from a Human instead of inventing a threshold.
Exception and failure paths
An exception should record its scope, reason, authorizer, approval, expiry, compensating controls, and revocation condition. An invalid or expired exception does not rewrite a failure. For every blocked or failed obligation, specify who investigates, whether a rerun is allowed, and what stops or rolls back the rollout.
Keep gate design, comparison against supplied evidence, and pipeline implementation as separate activities. A designed rule or a dry run is not proof that the real gate ran.
During review, write the rules as state transitions and pay particular attention to stale evidence, reruns, and exception expiry. The project contract must define the response; the gate should not invent a default.
| Changed condition | Gate response to specify | What to verify |
|---|---|---|
| Previously valid evidence expires | Whether it becomes UNKNOWN, BLOCKED, or another declared state, and how the current decision handles it | Whether an old PASS remains visible, with evidence time and affected scope |
| A failed check may be rerun | Who can start it, when to stop rerunning, and which build and data version to use | Whether the new run links to the original failure without overwriting its record |
| An exception expires | When the original obligation resumes and whether new human approval is required | An expired exception cannot renew itself or erase a failure |
| A required check is blocked | Whether the decision pauses, who is notified, and when to stop or roll back | BLOCKED is not counted as a pass, and stop conditions are explicit |
Evidence marked PASS still may not satisfy the current gate if its source, build, or time window does not match. Name the specific linkage gap and leave acceptance of substitute evidence to an authorized owner.
Worked example
Consider a staged checkout release with unit, contract, browser, and production-observation obligations. Before deployment, the contract might require current unit and contract evidence. Before traffic expands, it might require browser evidence. Full rollout might depend on an agreed production observation window. These are candidate stages only; the project owner must set the actual obligations and thresholds.
If browser execution is BLOCKED by unavailable infrastructure, the gate should not silently convert that state to PASS. The owner can keep the rollout stopped or authorize a scoped, expiring exception with a compensating control. Record which choice was made, by whom, and what evidence will close the exception.
Keep the quality loop visible
In day-to-day testing, connect each obligation to its outcome: requirement or risk → reviewed test design → execution on a named build and environment → evidence-backed gate state → owner action → fix and linked rerun. AI-generated test cases enter as drafts. Review them against the requirement and expected result before execution; generation alone does not satisfy a gate. Preserve the original failure when a rerun happens, and count a result as PASS only when its evidence is fresh, valid, and tied to the stated obligation.
What to hand off
Use QGD-## for a missing or ambiguous gate rule. Include the object or obligation, source, trigger, expected concern, evidence state, impact and priority, owner role, close condition, and validation method. Also name the stop, escalation, and rollback conditions. State residual risk and keep Human approval separate from technical evidence.
Best practices
- Start with obligations that could cause business loss, a quality regression, or a blocked decision.
- Keep every rerun linked to the original result; require an owner, scope, expiry, and compensating control for each exception.
Boundaries
A gate controls a decision only within its stated scope. It cannot choose business risk tolerance or turn stale evidence into current proof. Do not claim all checks passed, release approval, or pipeline execution without the matching run record and authorized decision.
Related Skills
- Quality Dashboard Design: https://inaodeng.com/en/qaskills/quality-dashboard-design/ presents signals for people making decisions.
- Release Testing Workflow: https://inaodeng.com/en/qaskills/release-testing-workflow/ places checks across release stages.
- Quality Risk Analysis: https://inaodeng.com/en/qaskills/quality-risk-analysis/ helps rank the risks a gate should control.
Install and invoke
Install the individual Skill from the Awesome QA Skills repository:
npx skills add https://github.com/naodeng/awesome-qa-skills/tree/main/skills/en --skill quality-gate-design
Invoke quality-gate-design with the target decision, gate policy, obligations, evidence sources, owners, exceptions, and rollback rules.
Use the quality-gate-design Skill.
Decision controlled: [merge, deploy, expand traffic, or stop].
Policy and obligations: [approved rules and required evidence].
Evidence: [source, version, freshness, and current state].
Owners and exceptions: [roles, approval path, expiry, and compensating controls].
Map missing, stale, blocked, and failed evidence to the supplied policy. Flag undefined rules for an owner; do not invent thresholds or approve a release.
FAQ
Can missing evidence count as PASS?
No. Keep it in the state defined by the project contract, such as UNKNOWN or BLOCKED. If the contract does not define a state, record the gap and ask the owner to decide.
Can this Skill choose the release threshold?
No. It can expose missing or conflicting rules. An accountable owner must set the threshold and decide whether an exception is acceptable.
Source
Read the Quality Gate Design prompt: https://github.com/naodeng/awesome-qa-skills/tree/main/skills/en/testing-types/quality-gate-design/prompts/quality-gate-design.md before using the Skill. The source directory: https://github.com/naodeng/awesome-qa-skills/tree/main/skills/en/testing-types/quality-gate-design contains the entry point, references, and evaluation contract.
- Skill details: https://inaodeng.com/en/qaskills/quality-gate-design/
More Resources
- Personal website: https://inaodeng.com/en/
- AI-generated test detection project: https://github.com/AI-Native-QA-Lab/ai-test-auditor
- DeepSeek QA Workbench: https://github.com/naodeng/dsh-qa
- More AI testing skills: https://inaodeng.com/en/qaskills/
- More testing wiki: https://inaodeng.com/en/wiki/
- More AI Wiki: https://inaodeng.com/en/AIWiki/