create-verification-skill
Create or migrate a shared project verification contract and discoverable skill wrappers, prove one real app path, and retain evidence.
Skill by 3metaJun
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/3metajun-create-verification-skill
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: documents, productivity
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/create-verification-skill
What's inside
- Skill name
create-verification-skill- Description
- Create or migrate a shared project verification contract and discoverable skill wrappers, prove one real app path, and retain evidence. Use for a reusable verification or control workflow, not a one-time verification pass.
- License
- MIT
- Instructions
- 113 lines in SKILL.md · view on GitHub
Files (6)
- SKILL.md
- README.md · references/feature-map-example/
- create-note.md · references/feature-map-example/
- search.md · references/feature-map-example/
- harness-paths.md · references/
- t3code.md · references/hosts/
SKILL.md (first lines)
# Create a verification skill Create one verification contract that agents in different Harnesses can use to prove the same application behavior. Store project facts in `.harness/verify/<app>/`, with thin `verify-<app>` skill wrappers for discovery. Record commands and observations from this repository, not a generic test plan. ## 1. Interview the repository Read `AGENTS.md`, `.harness/policy.json`, and `.harness/workflow.md` when present. Follow the shared project workflow before editing. Inventory canonical contracts and existing `verify-*` or control skills across every project skill root in [harness-paths.md](./references/harness-paths.md). Follow each wrapper's `metadata.verification-contract` path from the repository root. Several wrappers pointing to one contract are one target, not competing verification definitions. Answer what you can from code and documentation before asking the user: - **Surface:** Identify the primary user surface and any secondary surfaces. - **Run:** Find the repository's own launch command, required environment, ports, seed data, and authentication. - **Drive:** Prefer an existing Playwright, Cypress, PTY, expect, API, desktop, or mobile harness. Otherwise use the host agent's available browser, terminal, desktop, or HTTP driver. - **Observe:** Identify screenshots, terminal transcripts, response bodies, logs, exit codes, files, and durable state that can prove behavior. - **Isolate:** Determine whether parallel instances can use distinct ports, profiles, or data directories. If not, require exclusive use. If the checkout does not run as documented, fix or precisely report that base problem before recording a workflow against it. ## 2. Write or migrate the canonical contract Reuse the app ID and canonical path already registered in the project policy. For a new app, use `.harness/verify/<app>/contract.md`. Keep its feature map in `features/` and its owned scripts in `helpers/` beside the contract. For existing verification definitions, follow [the migration procedure](./references/harness-paths.md). Reconcile conflicting instructions against the running app before choosing a canonical version. A copied legacy map is not evidence of correct behavior. The contract must include these H2 sections: - **Launch:** Exact start command, readiness signal, ownership marker, and teardown. - **Isolation:** Instance-specific ports, profiles, and data directories, or an explicit exclusive-use procedure for resources that cannot be separated. - **Doctor:** A read-only check that confirms the right instance, build, port, data directory, and auth are healthy. - **Drive:** Real commands and stable selectors from this repository. Prefer roles, accessible names, route paths, prompt strings, and structured output over coordinates or tab order. - **Evidence:** Capture both the user action and resulting state. Verify durable side effects from a second read path. Treat dry-run and test modes as claims that still require observation. - **Cleanup:** Stop only the instances this run created. Remove scratch state, never proof artifacts. Never kill by process name alone. - **Helpers:** Explain every shipped helper and its exact invocation. State the working directory for commands and resolve helper paths from the repository root. Keep app facts, selectors, fixtures, assertions, and cleanup in this directory. Describe required driver capabilities without naming the host agent's browser or desktop tool. A repository-owned Playwright, HTTP, PTY, or control script is portable and may appear by its exact name. If a capability is unavailable, report that path as blocked instead of substituting a weaker check. ## 3. Seed the feature map Create `features/README.md` and one file for each of the three to five most important user-visible features. Use the contract in [feature-map-example/README.md](./references/feature-map-example/README.md). Each feature records every user entry point, exact driving steps, observable end state, and known traps. Do not copy the map into a Harness skill root or an automation directory. Other consumers must reference this index and load its linked feature files. ## 4. Render the discovery wrappers Use [the wrapper rules](./references/harness-paths.md). Each wrapper contains a repository-relative `metadata.verification-contract` pointer and capability adaptation only. Launch commands, feature facts, evidence requirements, and helpers stay in the canonical directory. Cursor discovers `.agents/skills/`, so a shared Codex and Cursor project needs one neutral wrapper there instead of two same-name definitions. ## 5. Prove the generated workflow Run the generated instructions end to end: 1. Launch the isolated instance. 2. Run the doctor check. 3. Drive one mapped feature through its real user path. 4. Capture the named evidence and verify any side effect. 5. Clean up and confirm the evidence still exists. Start from each generated wrapper and confirm it resolves the same contract. Run the project policy check when a policy exists. A static wrapper check proves discovery structure, not another Harness's runtime capabilities. Fix every failed instruction and repeat its affected step. A workflow that has not driven the real application once is a draft, not a deliverable. After every failed iteration, run the generated cleanup, confirm its owned processes, ports, profiles, and scratch state are released, then retry. ## 6. Hand off Report the canonical path, generated wrappers, the feature proved, evidence paths, and any surface that could not be exercised. For migrations, account for every retired definition and updated consumer. Do not claim untested paths.
From 3metaJun/mstack, MIT.
More from 3metaJun/mstack
- 0.567score
diagnosing-bugs
3metaJun
Diagnose hard, flaky, intermittent, environment-specific, or performance regressions, especially when a prior fix failed.
SkillAFreshness: AClaude Code +9 - 0.567score
codebase-design
3metaJun
Design or review a module's public interface, seam placement, depth, and testability.
SkillAFreshness: AClaude Code +9 - 0.541score
architect
3metaJun
Sketch types, signatures, and module structure before code, then stay in the loop while implementation fills in.
SkillAFreshness: AClaude Code +9 - 0.541score
interrogate
3metaJun
Use for "interrogate", "adversarial review", "multi-model review", "challenge this", "stress test this code", "find blind spots", or "tear this apart".
SkillAFreshness: AClaude Code +9 - 0.541score
how
3metaJun
Use for "how does X work", code walkthroughs before changing something, and placement / ownership / layering questions ("where should this live", "which…
SkillAFreshness: AClaude Code +9 - 0.539score
benchmark-checklist
3metaJun
Vet a perf measurement (limiter, tuning, limits, errors, repeatability, relevance, and whether the work happened) before you report or act on it.
SkillAFreshness: AClaude Code +9 - 0.539score
correct
3metaJun
Find the mistakes agents keep repeating in this repo and make each one impossible.
SkillAFreshness: AClaude Code +9 - 0.539score
figure-it-out
3metaJun
Design an auditable playbook when no narrower one fits: a large migration, an ambitious multi-part change, or work a human reviews after stepping away.
SkillAFreshness: AClaude Code +9 - 0.539score
automate-me
3metaJun
Use for "automate me", "create/update/refresh my -mode skill", "turn/capture my preferences or working style into a skill", or wanting agents to follow how…
SkillAFreshness: AClaude Code +9 - 0.536score
blast-radius
3metaJun
Find what a change could break beyond its diff, then prove the one or two safety-critical facts by running real code.
SkillAFreshness: AClaude Code +9 - 0.536score
arena
3metaJun
Spawn N parallel candidates at the same task, pick a base, graft the strongest parts of the losers into it.
SkillAFreshness: AClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 1
- Last commit (this folder)
- Oct 6, 2026 (23 days ago)
- Last release
- Oct 23, 2026 (6 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/3metaJun/mstack
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from 3metaJun
- 0.567score
diagnosing-bugs
3metaJun
Diagnose hard, flaky, intermittent, environment-specific, or performance regressions, especially when a prior fix failed.
SkillAFreshness: AClaude Code +9 - 0.567score
codebase-design
3metaJun
Design or review a module's public interface, seam placement, depth, and testability.
SkillAFreshness: AClaude Code +9 - 0.541score
architect
3metaJun
Sketch types, signatures, and module structure before code, then stay in the loop while implementation fills in.
SkillAFreshness: AClaude Code +9 - 0.541score
interrogate
3metaJun
Use for "interrogate", "adversarial review", "multi-model review", "challenge this", "stress test this code", "find blind spots", or "tear this apart".
SkillAFreshness: AClaude Code +9 - 0.541score
how
3metaJun
Use for "how does X work", code walkthroughs before changing something, and placement / ownership / layering questions ("where should this live", "which…
SkillAFreshness: AClaude Code +9 - 0.539score
benchmark-checklist
3metaJun
Vet a perf measurement (limiter, tuning, limits, errors, repeatability, relevance, and whether the work happened) before you report or act on it.
SkillAFreshness: AClaude Code +9
Cite this page
create-verification-skill. Agent Mart. https://agentmart.dev/skills/3metajun-create-verification-skill (last updated 2026-10-24)