| name | acceptance-walkthrough |
| description | Walk through a single feature or flow using the demo agent persona. Use when asked to "demo this feature", "walk me through [flow]", "show me what was built", "run acceptance walkthrough for [feature]", or when validating one specific flow without running the full pre-flight pipeline. Produces a focused ACCEPTANCE-DEMO.md for a single flow. Does not run the adversarial reviewer or real-data stress agent — use acceptance-prep for the full pipeline.
|
| argument-hint | Feature or flow name (e.g. "user signup", "report export", "password reset") |
Acceptance Walkthrough
Walk through a single feature or flow as the demo agent. Produces a structured
self-demo that can be used for a quick human acceptance session or as input to
the full pre-flight pipeline.
Use this skill for:
- Validating a single flow without running the full pipeline
- A quick human check before a demo or stakeholder review
- Scoped acceptance testing when the full
[ACCEPTANCE-PREP] pipeline is not warranted
For the full pre-flight pipeline (all four layers), use acceptance-prep instead.
Inputs
- Feature or flow name — from the argument, or ask if not provided
- PR diff or code — read the relevant code for the feature
- CLAUDE.md — understand the project's conventions and constraints
- SPEC.md or acceptance criteria — if a
[CRITERIA] ticket exists, read it
If no argument is provided, ask: Which feature or flow should I walk through?
Instructions
Apply the demo-agent persona to the specified
feature or flow.
Work through the demo agent's output format:
- Feature Summary — what does this do from the user's perspective?
- User-Facing Flows — every path through the feature
- Entry point, happy path, expected output
- Edge cases handled and not handled
- Integration Points — external systems touched
- Assumptions — anything not in the spec that was decided during implementation
- Lowest-Confidence Areas — three specific places most likely to fail under
conditions not tested
Output Format
## Acceptance Walkthrough — [Feature/Flow Name]
### Feature Summary
[What this does, from the user's perspective]
### Flows
#### [Flow Name]
- Entry point: [how a user gets here]
- Happy path: [step by step]
- Expected output: [what the user sees]
- Edge cases handled: [list]
- Edge cases NOT handled: [list — be specific]
[Repeat for each flow]
### Integration Points
[External systems, expected inputs/outputs, failure handling]
### Assumptions
[Anything decided during implementation that isn't in the spec]
### Lowest-Confidence Areas
1. [Specific area, specific reason]
2. [Specific area, specific reason]
3. [Specific area, specific reason]
After the Walkthrough
The human reviews the output and either:
- Signs off: flow matches intent, no follow-up needed
- Gives feedback: narrate issues to the agent using the live feedback pattern
"I'm going to describe issues. For each one: (a) do you understand it,
(b) is it a quick fix or needs more thought, (c) could fixing it break
anything else?"
Apply the triage discipline: fix-now items (copy, layout, obvious single-line
bugs) are handled immediately. Items requiring architecture changes, stakeholder
input, or investigation are ticketed as [BUG] issues.