qa-session
Run a QA session for a unit of work (milestone, task, diff, PR) against any surface — Website, Docker Compose, HTTP API, gRPC, MCP, GraphQL. Derives the user experience the work should deliver, confirms it, probes the running surface, delivers a verdict report. Triggers include "QA this", "QA session", "verify the milestone", "verify this PR", "acceptance check", "does this deliver", "check user experience".
- License 许可
- MIT
- 最近更新
- 2026-09-29
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- qa-session
- description
- Run a QA session for a unit of work (milestone, task, diff, PR) against any surface — Website, Docker Compose, HTTP API, gRPC, MCP, GraphQL. Derives the user experience the work should deliver, confirms it, probes the running surface, delivers a verdict report. Triggers include "QA this", "QA session", "verify the milestone", "verify this PR", "acceptance check", "does this deliver", "check user experience".
02 资源文件 6
- 参考文档references/compose.md1.2 KB
- 参考文档references/graphql.md1.1 KB
- 参考文档references/grpc.md995 B
- 参考文档references/http.md1.0 KB
- 参考文档references/mcp.md1.0 KB
- 参考文档references/website.md1.2 KB
03 SKILL.md 正文
Run the QA session as four steps. Each ends on its completion criterion; do not advance early.
1. Bound the work
Identify the work unit and what changed in it: read the milestone/task/diff/PR description and the actual changes (diff, files, commits). Ground facts in the repo; note claims the description makes that the diff does not show.
Completion: one paragraph — what changed, what it claims to deliver.
2. Derive the UX contract
From the work unit, state the user experiences that should be observable on the surface. For each expectation write: who the user is, what they do, what they must see or get. Expectations are outcome statements, not implementation steps.
Then confirm with the user via ask: present the contract, ask them to correct, add, or drop expectations. Do not probe before confirmation — a QA session against the wrong contract produces a confident wrong verdict.
Completion: user-confirmed list of expectations. This list is the source of truth for the verdict.
3. Probe the surface
Load the reference for the surface the user named, read it, then probe:
- Website →
skill://qa-session/references/website.md - HTTP API →
skill://qa-session/references/http.md - gRPC API →
skill://qa-session/references/grpc.md - MCP server →
skill://qa-session/references/mcp.md - GraphQL API →
skill://qa-session/references/graphql.md - Docker Compose →
skill://qa-session/references/compose.md
If the surface is not running, bring it up first (compose.md covers startup for stack surfaces). Probe every confirmed expectation. Each expectation MUST end with a verdict and captured evidence — a screenshot, a response body, a log line, a measured value. An expectation without evidence is unprobed, not passed.
Completion: every expectation has verdict + evidence, or is explicitly marked blocked with the reason (surface unreachable, prerequisite missing).
4. Verdict report
Deliver:
## QA Verdict: <work unit> — <N>/<total> pass
PASS <expectation> — <evidence one-liner>
FAIL <expectation> — <what diverged>, <evidence one-liner>
BLOCKED <expectation> — <why>
Blocking: <count> — <the failures that break the core experience>
Classify a failure blocking when it breaks the experience a normal user hits on the happy path. Divergences on edge cases, cosmetics, or contract details are non-blocking unless the work unit explicitly claims them. Cite evidence verbatim; never summarize a response you did not capture.
04 同仓库更多 Skills
writing-for-agents
Writing documents for agents. Use when creating or editing skills, or modifying AGENTS.md or CLAUDE.md.
write-skill
Use when creating, designing, or fixing an agent skill (SKILL.md pack) — triggers include "write a skill", "make me a skill", "create a skill", skill directories, and skills that fail to trigger or fire on the wrong tasks.
write-agent
Use when creating, designing, or fixing an omp subagent (task agent) — triggers include "write an agent", "make me an agent", "create a subagent", agent `.md` definitions, and agents that misbehave when spawned via task.
wizard
Generate an interactive bash wizard that walks a human through steps only they can perform. Use when provisioning infrastructure, setting up credentials or CI secrets, walking an unfamiliar third-party dashboard, or running a one-off migration or cutover. Don't invoke this for steps the agent can perform itself.
wayfinder
Plan a huge chunk of work (more than one agent session can hold) as a shared map of decision tickets on your issue tracker, and resolve them one at a time until the way to the destination is clear.
wait-what
Stop. That last message did not land: re-pitch it.