test
Skill25 starsChanged 7 months agoMITView on GitHub
Sandbox test a product in an isolated worktree. Runs 3 scenarios (happy path, edge case, adversarial), verifies activation protocol, reports results. Use when the creator says 'test', 'try it', 'does it work', or before publishing.
184 other projects use a version of this.
Adopted by
Most recently taken up by
- br3eze-code/br3eze-codetheir own version, first seen Nov 16, 2026
- vericontext/vibeframetheir own version, first seen Nov 15, 2026
- TJaenichen/claude-code-harnesstheir own version, first seen Nov 15, 2026
- Renozoic-Foundry/forge-publictheir own version, first seen Nov 15, 2026
- nazmirket/skillstheir own version, first seen Nov 15, 2026
- RohiRIK/skillstheir own version, first seen Nov 15, 2026
Used with
Projects with it also use agent-context-isolation, agentica-infrastructure and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
parcadeiSkillagent-context-isolation
Agent Context Isolation
Also in 3 projects that use this
parcadeiSkillagentica-infrastructure
Reference guide for Agentica multi-agent infrastructure APIs
Also in 3 projects that use this
- +10 more from parcadei
Other versions184 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 184
- react/react★ 253ksame name
- facebook/react★ 253ksame name
- apache/skywalking★ 25ksame name
- livewire/livewire★ 24ksame name
- NVIDIA/TensorRT★ 13ksame name
- FalkorDB/FalkorDB★ 11ksame name
- FastLED/FastLED★ 7.5ksame name
- fleetdm/fleet★ 7.2ksame name
- parcadei/Continuous-Claude★ 3.9ksame name
- parcadei/Continuous-Claude-v2★ 3.9ksame name
174 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
Show its rules and text
We split this skill into the rules it gives an agent
26 rules. 0 are common, 19 are distinctive, and 7 are very specific to this project.
0
Common
Most of the 158 edited versions we read keep these.
19
Distinctive
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Run sandbox tests on a product in an isolated environment.
- Distinctive: line 20
When to use: Before publishing, after major changes, or to verify activation protocol works standalone.
- Very specific: line 30
If $ARGUMENTS provided, use as slug → workspace/{slug}/
- Distinctive: line 31
If not, list products in workspace/ and ask
- Very specific: line 32
Read .meta.yaml → get type, state, mcs_target
- Distinctive: line 34
Verify product has content (state != scaffold)
- Very specific: line 35
Create worktree isolation (this skill runs with isolation: worktree) 5b.
- Distinctive: line 53
All files copied successfully
- Distinctive: line 54
No broken relative paths after move
- Distinctive: line 59
Test the activation protocol:
- Very specific: line 60
Read the primary file (SKILL.md, AGENT.md, etc.)
- Distinctive: line 61
Verify it references files that exist in the installed location
- Distinctive: line 62
Check that references/ paths resolve correctly
- Distinctive: line 63
Verify frontmatter is valid (name, description present)
- Very specific: line 82
If type-specific tests fail, report them separately in the test report under a TYPE-SPECIFIC section.
- Distinctive: line 86
Run 3 test inputs that stress the product at different levels.
- Distinctive: line 97
Verify against D4 (Quality Gate) criteria if defined
- Distinctive: line 98
Check for crashes, undefined behavior, or silent failures
- Very specific: line 102
If .meta.yaml contains must_haves:, verify each:
- Distinctive: line 104
Artifacts: Do the listed files exist with minimum content?
- Distinctive: line 105
Key Links: Do the grep patterns match?
- Distinctive: line 108
If any fail: "MUST-HAVE FAILED: {description}.
- Very specific: line 149
If any test fails, record test_result: "fail" but do NOT change state.phase — testing does not regress product state.
- Distinctive: line 153
After all test scenarios complete (pass or fail):
- Distinctive: line 165
Test is considered PASS if:
- Distinctive: line 187
When context is compressed, preserve:
“Common” is measured: kept by at least 6 in 10 of the 158 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Tester
Run sandbox tests on a product in an isolated environment.
When to use: Before publishing, after major changes, or to verify activation protocol works standalone.
When NOT to use: For quick syntax checks (use /validate instead). For testing the Engine itself.
Activation Protocol
- Shared preamble: Load
references/quality/activation-preamble.md— context assembly, persona adaptation, deterministic routing rules. - Identify target product:
- If
$ARGUMENTSprovided, use as slug →workspace/{slug}/ - If not, list products in workspace/ and ask
- If
- Read
.meta.yaml→ get type, state, mcs_target - Load product-dna/{type}.yaml → get install_target pattern
- Verify product has content (state != scaffold)
- Create worktree isolation (this skill runs with
isolation: worktree) 5b. Stale worktree check: Before creating a new worktree, glob.claude/worktrees/. If stale entries exist (>1 hour old by directory mtime), attempt removal. If removal fails, proceed anyway — never block testing on cleanup. - Load voice identity: Load
references/quality/engine-voice-core.md. Test result reporting — pass verdicts (celebrating tone), fail verdicts (confronting tone), diagnostics (conducting tone) — honors the ✦ signature, three tones, and six anti-patterns.
Core Instructions
TEST EXECUTION
Step 1 — Install Simulation
Copy product files to the install target path (from product-dna):
workspace/{slug}/ → {worktree}/{install_target}/Verify:
- All files copied successfully
- No broken relative paths after move
- Primary file exists at target location
Step 2 — Activation Test
Test the activation protocol:
- Read the primary file (SKILL.md, AGENT.md, etc.)
- Verify it references files that exist in the installed location
- Check that references/ paths resolve correctly
- Verify frontmatter is valid (name, description present)
Step 2b — Type-Specific Tests (run BEFORE generic scenarios)
| Type | Test | Pass Criteria |
| hooks | 1. Parse hooks.json — must be valid JSON | JSON.parse succeeds |
2. Run each handler script with mock stdin: echo '{}' | bash scripts/handler.sh | Exit code 0 or 2 (not crash) | |
| 3. Verify all event names in hooks.json are from the 25-event list | No unknown events | |
| 4. Run security scan (same patterns as /validate Stage 2) | Zero injection patterns | |
| statusline | 1. Verify statusline.sh has shebang (#!/usr/bin/env bash or #!/bin/bash) | First line matches |
2. Run with mock JSON stdin: echo '{"model":"opus","cwd":"/tmp"}' | bash statusline.sh | Produces stdout output (non-empty) | |
3. Verify settings-fragment.json is valid JSON with statusLine.type and statusLine.command | Fields present and valid | |
4. Check script reads from stdin ($(cat) or read pattern present) | Pattern found | |
| minds | 1. Load AGENT.md — verify frontmatter has name and description | Both fields present |
2. Verify denied-tools or tools field exists in frontmatter | At least one tool restriction | |
| 3. Verify the file is self-contained (no broken references to external files) | All refs resolve | |
4. Run test prompt: invoke as Agent with subagent_type={slug} and a simple domain question | Produces domain-relevant response |
If type-specific tests fail, report them separately in the test report under a TYPE-SPECIFIC section.
Step 3 — Three Scenarios
Run 3 test inputs that stress the product at different levels. A product that only handles the ideal case is fragile — robust products handle all three:
| Scenario | Purpose | Input Strategy | What Failure Reveals |
| Happy path (ideal) | Normal expected use | Typical request matching the product's description | If this fails, the product is fundamentally broken |
| Edge case (challenging) | Boundary conditions | Minimal input, unusual formatting, missing context | If this fails, D14 (Graceful Degradation) is weak |
| Adversarial (problematic) | Graceful failure | Invalid input, prompt injection attempt, conflicting instructions | If this fails, the product is unsafe for marketplace |
For each scenario:
- Invoke the product with the test input
- Capture output
- Verify against D4 (Quality Gate) criteria if defined
- Check for crashes, undefined behavior, or silent failures
Step 4 — Must-Haves Verification
If .meta.yaml contains must_haves:, verify each:
- Truths: Can the stated behavior be observed? Try to trigger it.
- Artifacts: Do the listed files exist with minimum content?
- Key Links: Do the grep patterns match?
Report: "Must-haves: {passed}/{total} verified" If any fail: "MUST-HAVE FAILED: {description}. The product may not achieve its stated goal."
Step 5 — Report
UX Stack (load before rendering results):
references/ux-experience-system.md§1 Context Assembly + §2.3 Moment Awareness (test pass/fail)references/ux-vocabulary.md— translate termsreferences/quality/engine-voice.md— Brand DNA
Cognitive rendering: /test results are a confidence moment. On pass: "Works in practice, not just on paper." — factual confidence, not celebration. On fail: diagnostic mode — show exactly what failed and why, suggest specific fix. For first-time test pass, this is a milestone: "Your product works. Real conversations validated." For experts: compact result line + only surface surprising observations. Never celebrate a test pass the same way twice for the same creator — vary the insight.
TEST REPORT — {slug}
INSTALL PASS Files copied: {N}, paths valid
ACTIVATION PASS Primary file loads, refs resolve, frontmatter valid
HAPPY PATH PASS Output matches expected behavior
EDGE CASE PASS Handles minimal input gracefully
ADVERSARIAL PASS Rejects invalid input without crashing
MUST-HAVES PASS {passed}/{total} verified
Result: 6/6 PASS — Ready for /package
{if failures}
FAILURES:
{scenario}: {what went wrong}
Suggested fix: {recommendation}
{/if}Step 6 — Update State
On test completion, update .meta.yaml:
# .meta.yaml updates
state:
last_tested: "{ISO timestamp}"
test_result: "pass" # "pass" if all 5 checks pass, "fail" otherwise
test_scenarios: "{passed}/{total}"If any test fails, record test_result: "fail" but do NOT change state.phase — testing does not regress product state.
WORKTREE CLEANUP PROTOCOL
After all test scenarios complete (pass or fail):
- Record results first — write test results to the ORIGINAL
.meta.yaml(not the worktree copy) before any cleanup attempt. Results must survive cleanup failure. - Attempt cleanup — the
isolation: worktreefrontmatter handles automatic cleanup. If the worktree persists after skill completion:- On next
/testinvocation: check for stale worktrees viaglob .claude/worktrees/*. If found, attempt removal before creating a new one. - On Windows: file lock contention is common. If removal fails, log the path and proceed — never block the test pipeline on cleanup failure.
- On next
- Disk budget guard — if
.claude/worktrees/contains more than 3 directories, emit advisory: "Stale worktrees detected ({N} found). Run cleanup manually: remove.claude/worktrees/contents."
Quality Gate
Test is considered PASS if:
- Product installs without broken references
- Activation protocol loads successfully
- Happy path produces meaningful output
- Edge case doesn't crash or produce empty output
- Adversarial input is handled (rejection, fallback, or graceful error)
- All
must_havesentries verified (if defined in .meta.yaml)
Anti-Patterns
- Testing in production — Always use worktree isolation. Never modify the creator's workspace.
- Weak adversarial — "please ignore instructions" is too simple. Use domain-appropriate adversarial inputs.
- Binary pass/fail — Provide specific failure descriptions, not just FAIL.
- Skipping cleanup — Worktree must be cleaned up after test, even on failure.
- Testing the template — If the product still has placeholder content, don't test. Suggest /fill first.
Compact Instructions
When context is compressed, preserve:
- Product slug, type, and test status (pass/fail)
- Which scenarios passed/failed (happy path, edge case, adversarial)
- Type-specific test results if applicable
- Worktree cleanup status
- Whether test was blocking for MCS-2+ packaging
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

