mikeparcewski/scenario-executor

Scenario Executor

Runs acceptance test scenarios end-to-end. Parses scenario markdown, executes each step (bash, slash-command, or prose), captures output, and reports structured results. This is a runner tool — the decision of what a result means is made by the test-designer agent (Phase 3: ANALYZE / VERDICT).

Quick Start

Invoke this skill when you have a scenario file and need to execute it:

Skill(
  skill="wicked-garden:qe:scenario-executor",
  args="{path_to_scenario_file}"
)

Typical output:

## Results: {scenario name}
**Status**: PASS | FAIL | PARTIAL
**Duration**: {total}s
**Steps**: {N} passed, {M} failed, {K} skipped

| Step | Type | Status | Duration | Details |

When to Invoke

• You need to run a wicked-garden acceptance scenario
• You need to run a wicked-scenarios format file (api/browser/perf/infra/security/a11y)
• A scenario mixes bash commands and /wicked-garden:* slash commands
• A scenario has prose-only steps that need interpretation

QE Pipeline Context

Before executing, detect whether this skill is being invoked inside a QE trio pipeline (via the test-designer agent) or standalone.

Detection: If the invoking context contains a ## Test Plan header, this skill is running inside the test-designer agent's Phase 2 (EXECUTE). Skip further QE checks and proceed with direct execution.

Standalone: If no test plan header, the caller is executing a scenario directly. Consider recommending they dispatch the test-designer agent for evidence-gated acceptance testing instead.

Execution Process

For each scenario file:

1. Read — use Read tool
2. Parse YAML frontmatter — name, description, tools (required/optional), env, timeout
3. Execute ## Setup — bash via Bash tool; slash commands via Skill tool
4. Execute each ### Step N in order
5. Execute ## Cleanup — always runs, even on failure
6. Report results in the standard format

Slash-Command Parsing

Slash commands in scenario files follow:

/wicked-garden:{domain}:{command} [args...]

Shortened form:

/wicked-{domain}:{command} [args...]

For the Skill tool always use the full form. Examples:

• /wicked-garden:mem:store --type decision "chose PostgreSQL" → Skill(skill="wicked-garden:mem:store", args='--type decision "chose PostgreSQL"')
• /wicked-garden:crew:start "Add OAuth2" → Skill(skill="wicked-garden:crew:start", args='"Add OAuth2"')

PASS/FAIL Rules

• Bash steps: exit code 0 = PASS, non-zero = FAIL
• Slash-command steps: Skill tool returns without error = PASS; error or failure message = FAIL
• Mixed steps: all must succeed for PASS; any failure = FAIL
• Prose steps: execute interpreted action + verify expected outcome
• Missing external CLI tool: SKIPPED (not FAIL) — only for external CLIs not bundled with the plugin
• Per-scenario: all PASS → PASS; any FAIL → FAIL

See refs/prose-interpretation.md for the full prose-step decision tree and interpretation patterns.

Output Format

## Results: {scenario name}

**Status**: {PASS|FAIL|PARTIAL}
**Duration**: {total}s
**Steps**: {pass} passed, {fail} failed, {skip} skipped

| Step | Type | Status | Duration | Details |
|------|------|--------|----------|---------|
| {name} | bash | PASS | 0.5s | |
| {name} | skill | PASS | 2.1s | |
| {name} | skill | FAIL | 1.0s | Error: ... |
| {name} | prose | PASS | 1.5s | Verified field X = Y |

Active Project Handling

Before executing scenarios that invoke crew:start, check for an active crew project:

Skill(skill="wicked-garden:crew:status")

If a project is active, archive it first:

Skill(skill="wicked-garden:crew:archive", args="{active-project-slug}")

Rules

• Sequential execution — run steps in order, don't parallelize
• Continue on failure — record FAIL but keep going
• Setup/Cleanup always run — even on failure
• Respect timeouts — use timeout for bash commands if specified
• Capture evidence — save stdout/stderr snippets for failed steps
• Be honest — don't mark PASS if output indicates an error, even if exit code is 0
• Interpret prose — never skip a step with an executable action just because it lacks a code fence

See Also

• refs/prose-interpretation.md — prose-step decision tree
• test-designer agent — owns the write + execute + verdict pipeline when running evidence-gated acceptance testing
• /wicked-garden:qe:run — wicked-scenarios CLI-tool runner (for scenarios with category frontmatter)