openspec-test

name	openspec-test
description	Generic test executor for OpenSpec changes. Use when: running checkpoint verification at gates, executing test.md strategies.
model	sonnet
context	main
allowed-tools	["Glob","Grep","Read","Edit","Bash"]

OpenSpec Test

Generic verification executor for OpenSpec gates. Follows test.md strategy literally — no improvisation.

flowchart LR
    A["Read test.md"] --> B["Validate quality"]
    B --> C{"Lazy patterns?"}
    C -->|Yes| X["BLOCKED"]
    C -->|No| D["Execute verification"]
    D --> E["Write logs"]
    E --> F["Report result"]

    classDef action fill:#C8E6C9,stroke:#388E3C,color:#000
    classDef decision fill:#FFF9C4,stroke:#FBC02D,color:#000
    classDef fail fill:#FFCDD2,stroke:#D32F2F,color:#000
    class A,B,D,E,F action
    class C decision
    class X fail

⚠️ AskUserQuestion Guard

CRITICAL: After EVERY AskUserQuestion call, check if answers are empty/blank. Known Claude Code bug: outside Plan Mode, AskUserQuestion silently returns empty answers without showing UI.

If answers are empty: DO NOT proceed with assumptions. Instead:

1. Output: "⚠️ Questions didn't display (known Claude Code bug outside Plan Mode)."
2. Present the options as a numbered text list and ask user to reply with their choice number.
3. WAIT for user reply before continuing.

checkpoint

Input: $ARGUMENTS = change-id section-number (e.g., add-feature 3)

Workflow:

1. Read openspec/changes/{change-id}/tasks.md → find ## {section-number}. ...
2. Read openspec/changes/{change-id}/test.md → verification strategy. If missing: warn + offer /openspec-plan tasks {change-id} or skip (garage only).
3. If test.md has a **Design**: header → read openspec/changes/{change-id}/design.md for invariants, key flows, and ADRs (informs gate evaluation context)
4. Validate: Scan for lazy patterns (grep "keyword", [ -f file ], wc -l, echo). If found → BLOCKED with suggestion for functional verification.
5. Init logs: Create test-logs/gate-{n}-{yyyyMMddHHmm}.yaml (raw, append-per-step) + test-logs/gate-{n}-{yyyyMMddHHmm}.md (summary). Timestamp = run start time. See reference.md §Log Formats.
6. Execute by type tag (default [auto]):

Tag	Action	Evaluator
[auto]	Run command, assert exit+output	Agent
[smoke]	Run command, display output	Agent + human glance
[manual]	Display instructions, AskUserQuestion (PASS/FAIL/SKIP)	Human

• Order: [auto] + [smoke] first, group [manual] at end
• Per step: Append YAML step block to raw log immediately after execution (not batched)

6. Write summary to gate-{n}-{ts}.md — rendered from raw YAML, not memory. See reference.md §Test Log Format.
7. Display gate summary inline to user.
8. Evaluate:
   • PASS → mark ### GATE {n}: desc [PASS] in tasks.md
   • PARTIAL → report failures with per-step diagnostics
   • BLOCKED → report blockers, suggest options

Log Strategy

Two artifacts per gate run in openspec/changes/{change-id}/test-logs/:

File	Format	Purpose
gate-{n}-{yyyyMMddHHmm}.yaml	YAML	Source of truth — append-per-step, full stdout/stderr (mask secrets)
gate-{n}-{yyyyMMddHHmm}.md	Markdown	Human summary — rendered from raw YAML after all steps

• Timestamped filenames: Each run = new file pair. No overwriting, no array append gymnastics.
• Append-per-step: Raw YAML grows incrementally — each step appended immediately after execution.
• Markdown truncation: stdout >50 lines → first 30 + last 10; stderr >20 lines → first 15

See reference.md §Raw Log Format for YAML schema, §Test Log Format for markdown template.

Mode Behavior

Read openspec/project.md → mode:

• garage: Warn if test.md missing, allow skip
• scale/maintenance: Block until test.md exists + passes

References

See reference.md for: log format templates, YAML schema, test progression strategy, framework-specific patterns, execution tracing, failure diagnostics.