Install
$ agentstack add skill-manmaxmotivation-ai-ai-assisted-qa ✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
About
AI-Assisted QA
This skill is an operating contract for QA work. It is designed to prevent an agent from guessing requirements, shrinking changed-risk scope, or treating a planned or partial check as a result.
Non-Negotiable Session Contract
Run only the stage explicitly requested by the user. The normal sequence is:
- Full analysis.
- Verification algorithm.
- Automated checks, executable test artifacts, and factual evidence.
- Manual test case.
- Final QA report.
These are separate human-gated messages. Do not proceed merely because the previous artifact appears complete. Present the complete artifact for review; when a stage permits saving, persist the same reviewed text without silent compression, rewording, reordering, or format normalization. A report is not a stage-three byproduct: write it only when actual execution evidence exists and the user asks for it.
Use the complete prompts in references/staged-session-prompts.md. They are the public, portable form of the workflow. Do not replace them with a shorter generic checklist.
Required Sources and Preflight
Before selecting coverage, obtain every available source of truth:
- task description, status, acceptance criteria, DoD, and custom fields;
- notes, attachments, screenshots, linked/parent/sibling work;
- merged revisions, pull requests, delivery information, and existing tests;
- actual local implementation and its dependencies;
- approved prior artifacts, route memory, and process barriers.
Identify missing sources. Do not invent task intent, entry points, data, environment state, execution results, access, or a passing outcome.
Apply references/publication-safety.md before reading material for public reuse and before publishing an artifact. A public example must be independently written and anonymized; do not copy proprietary case text, screenshots, URLs, identifiers, logs, or test data.
Route Memory and Barriers
Before new route discovery, search the project's private route atlas and similar approved algorithms. Reuse a matching proven route first. At the end of the algorithm stage, synchronize it as compare -> reuse/update/add.
Only confirmed, reusable routes belong in the atlas. Keep a recurring process barrier in its barrier record. Keep one-off root cause, incident details, and task-specific diagnosis in the task algorithm. Read references/route-memory.md for the full protocol.
Stage Requirements
1. Analysis
Create the full scope contract before execution. It must establish delivery status, implementation facts, prior and intended behavior, requirement/DoD coverage, changed surface, dependencies, state transitions, risks, regression zones, constraints, and the automation/manual boundary. For UI work, decide whether browser automation is both possible and needed, what it can prove, and what remains manual.
2. Verification Algorithm
Create a repeatable route from preconditions to observable proof. First make a read-only preflight of every needed contour and entry point. If a required proof is blocked, record the exact impact and stop that branch. For every route step, name the requirement/risk, user action, agent action, layer/entry point, evidence, expected factual outcome, and stop condition. Keep the route focused but do not omit a changed-risk branch merely to make it shorter.
3. Automated Checks and Evidence
Choose the smallest layer that actually proves each risk. Prefer existing project-native runners. Select shell, Python, browser automation, or a native runner deliberately and preflight the dependencies used. Run an executable artifact end-to-end before presenting it as rerunnable or suitable for CI.
Keep three things distinct: executable checks, their factual outcome report, and their evidence. Preserve failed and blocked scenarios. For browser-visible behavior, leave human-readable proof: action/transition video or focused static screenshot as appropriate. Raw logs and responses supplement, not replace, visual evidence.
4. Manual Test Case
Create one complete human test case from approved evidence. It must cover the original DoD and the human-only or independently required risk. Use explicit, confirmed entry points and test data. A step has one concrete action and one observable expected result. Do not duplicate automation, use placeholders in executable technical steps, or turn internal code names into user-visible expected results. The draft is approved before saving and then saved 1:1.
5. Final Report
Write only from actual execution records. The manual execution section must repeat the approved test-case steps in the same order and detail, including action, expected result, and actual result. Do not collapse steps. Preserve the original DoD wording. List all relevant artifacts and their provenance. The report may say PASS, FAIL, BLOCKED, PARTIAL, or NOT RUN; it may not infer a pass from intent, a green unrelated test, or an unavailable environment. Use references/final-report-template.md.
Quality Gates
Before completing a stage, confirm that:
- all available authoritative sources were reconciled or explicitly missing;
- each selected check has a risk-based reason;
- every requirement has a proof path or an explicit blocker;
- automation and manual work are separate and neither overclaims coverage;
- actual evidence, not prose, supports every reported result;
- route-memory updates contain only proven reusable knowledge;
- the artifact has no credentials, private endpoints, personal data, internal
identifiers, customer data, proprietary code, or copied private text;
- the final report mechanically traces to the approved test case and factual
automated evidence.
Use references/artifact-contract.md as the mandatory detailed contract and copy assets/AGENTS.md into a target project's local instructions when the project needs this workflow.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: ManMaxMotivation
- Source: ManMaxMotivation/AI
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet — be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.