Mb Verify by mrvladd-d — AI Skill for Any LLM

Overview

Publisher	mrvladd-d
Repository	`memobank`
Skill name	`mb-verify`
Stars	56
Forks	11
Bundled files	1
License	MIT
Links	Skill source Repository

Markdown instructions

A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

Works with any LLM

AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

1 bundled files

Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

Open source

Published by mrvladd-d on GitHub. Read the source before you install it.

Installation

Install the Mb Verify AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

Open the app and go to Plugins → Skills.
Choose "Install from GitHub".
Paste the skill folder URL below and confirm.
https://github.com/mrvladd-d/memobank/tree/main/skills/mb-verify
Enable the skill in any chat where you want it available.

Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills

git clone --depth 1 https://github.com/mrvladd-d/memobank.git /tmp/memobank
mkdir -p .claude/skills
cp -r /tmp/memobank/skills/mb-verify .claude/skills/mb-verify

Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Mb Verify in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Mb Verify on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Mb Verify is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

Loaded only when it is needed

The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

Switch models mid-chat

Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

mb-verify — Verifier loop (acceptance → evidence → verdict)

What it does: checks a completed task against acceptance criteria and records the result with evidence.
Use it when: implementation is done and you want an explicit PASS, FAIL, or partial verdict.
Input: TASK_ID, acceptance criteria sources, and the task protocol files.
Output: verification.md, evidence artifacts, verification verdict, and recommended next status/follow-up bugs when criteria fail.

Goal

Independent-ish verification so we don’t “trust without verify”.

This is not the adversarial semantic pass. If a task may satisfy AC/REQ while still being wrong in substance, follow with /red-verify / mb-red-verify.

Status Transition Modes

Status transitions have two modes.

Scheduler mode:

/autopilot and /autonomous own task status transitions.
Scheduler decides closure/failure/blocking eligibility.
/execute returns scoped implementation handoff; it does not close tasks.
/verify gives functional verdict/evidence; in scheduler mode it does not close/fail/block/promote.
/red-verify gives semantic verdict for T2/T3; in scheduler mode it does not close/fail/block/promote.
/mb-sync records/reconciles state after the scheduler-provided closure/failure/blocking decision. It does not decide closure itself.
T0/T1 scheduler closure may use compact evidence / functional PASS according to tier policy.
T2/T3 scheduler closure requires VERDICT: PASS plus SEMANTIC_VERDICT: semantic-pass before scheduler marks done.
T3 scheduler closure also requires exact markers HUMAN_CHECKPOINT: done and ROLLBACK_RECOVERY_NOTE: present.

Manual mode:

Expected T0/T1 simple flow: /execute -> /verify.
Manual closure is allowed only when an explicit closure owner exists.
T0/T1 may be marked done after functional VERDICT: PASS and completed evidence.
T2/T3 must not treat /verify PASS alone as final done; run /red-verify and require SEMANTIC_VERDICT: semantic-pass before final closure//mb-sync.
If required T2/T3 /red-verify returns anything other than semantic-pass, leave closure pending or blocked, not done; optional T0/T1 red-verify does not make their normal verify-based closure stricter.
semantic-concern in manual mode means do not trust the existing done state without human review / follow-up.
Do not mix scheduler mode and manual mode inside one task run.
No persisted mode field is used.

Inputs

TASK_ID (e.g. TASK-123)
Authoritative task record via .memory-bank/tasks/index.json and .memory-bank/tasks/<TASK_ID>.task.json
Mandatory tier: T0|T1|T2|T3 in that task record
Links to acceptance criteria:
- .memory-bank/features/FT-* and/or
- .memory-bank/requirements.md (REQ IDs)
Link to protocol plan: .protocols/<TASK_ID>/plan.md

If present, also use:

linked authoritative SDD specs for any tier
verification_targets
normative_inputs
constraints
invariants
task record references to source artifacts

Authoritative SDD spec links are links in task richer fields or linked feature spec_design_links that point to .memory-bank/spec-index.md, .memory-bank/tech-specs/, .memory-bank/architecture/, .memory-bank/contracts/, .memory-bank/domains/, .memory-bank/states/, .memory-bank/adrs/, .memory-bank/testing/, or .memory-bank/runbooks/.

Preconditions

Implementation is done and gates were run (or failures recorded).
.memory-bank/tasks/index.json lists the target task record, and the indexed .task.json validates the requested TASK_ID.
Authoritative verification routing is only task.tier; the old risk / risk.level model is invalid.

Required outputs

T0 / T1: verification may be recorded in compact .protocols/<TASK_ID>/run.md.
T2 / T3: update (or create) .protocols/<TASK_ID>/verification.md using:
- ./references/shared-protocols-verification-template.md
Store evidence in .tasks/<TASK_ID>/:
- logs, screenshots, videos, reproduction steps
Add completed evidence entries to the task record verify field; evidence_required and verification_targets remain requirements/targets, not proof by themselves.
Before any command sets status: done, the task record verify field must contain completed verification/evidence entries.

Status ownership

mb-verify owns verification evidence and VERDICT: PASS|FAIL|NEEDS-CLARIFICATION.
When invoked by /autopilot or /autonomous, it must not close the task, set status: done, set status: failed, block dependents, or promote dependents. It reports a recommended next status to the scheduler.
In standalone/manual mode, it may mark a T0 / T1 task done after functional VERDICT: PASS only with explicit closure ownership.
For T2 / T3, mb-verify records functional evidence and closure recommendation, but final closure requires mb-red-verify semantic-pass first.

Process

1) Prime only what you need

Read:

.memory-bank/tasks/index.json
indexed .memory-bank/tasks/<TASK_ID>.task.json
.protocols/<TASK_ID>/context.md
.protocols/<TASK_ID>/plan.md
.protocols/<TASK_ID>/progress.md
acceptance criteria source docs
.memory-bank/spec-index.md and all linked authoritative SDD specs when the task record or linked feature contains SDD spec links, for any tier

Before verifying, validate the authoritative task record:

the task is present in .memory-bank/tasks/index.json
the indexed record id matches TASK_ID
required fields for verification are present (status, feature, reqs, depends_on, gates, verify)
tier is present; if missing, stop
for T2 / T3, linked SDD specs are present in task richer fields, feature spec_design_links, or spec-index.md; if absent, stop and route back to /spec-improve or /spec-auto

Do not block T0 / T1 only because SDD spec links are absent. If the authoritative task record is missing or invalid, stop and report the issue instead of verifying from protocol docs alone.

Priority:

linked authoritative SDD specs for any tier, when present
explicit Verification Targets
explicit Normative Inputs
classic feature acceptance criteria and RTM
evidence in .tasks/<TASK_ID>/

Missing richer fields or absent SDD spec links must not block verification of a classic T0 / T1 task.

2) Verify acceptance criteria

For each AC / REQ:

run the smallest meaningful check
prefer deterministic checks (tests/CLI) over “looks OK”
record what you did and link the evidence

If the task changes UI or browser behavior:

prefer Playwright / agent-browser / CDP-driven verification
capture screenshots/videos/traces when useful
store artifacts in .tasks/<TASK_ID>/
do not use “I clicked around manually” as the main evidence when browser automation is available

3) Verdict

If anything fails:

set VERDICT: FAIL
create a bug doc in .memory-bank/bugs/BUG-<short>.md
add a follow-up .task.json and update .memory-bank/tasks/index.json (if needed)
recommend current task status: failed
in scheduler mode, do not block downstream dependents directly; return that recommendation to the scheduler

If all pass:

VERDICT: PASS
add completed verification/evidence entries in verify
apply status by tier:
- scheduler mode: recommend the scheduler decision; do not close/fail/block/promote
- manual mode: may set T0 / T1 status: done after functional VERDICT: PASS with explicit closure ownership; for T2 / T3, leave closure pending /red-verify SEMANTIC_VERDICT: semantic-pass
in scheduler mode, final T2 / T3 closure is eligible only after /red-verify / mb-red-verify returns semantic-pass

4) Sync recommendations

Record RTM/feature lifecycle recommendations for /mb-sync
Do not independently perform scheduler closure or dependent block/promotion

Definition of done

Verification output exists and is evidence-backed: compact run.md for eligible T0 / T1, full verification.md for T2 / T3.
PASS verification has updated RTM/task evidence; T2 / T3 tasks are not closed until /red-verify / mb-red-verify produces semantic-pass.
FAIL tasks have a bug doc and next steps.

Frequently asked questions

What does the Mb Verify AI skill do?

Verify one TASK-* against acceptance criteria and record reproducible evidence.

Why use Mb Verify on TypingMind?

Because you install it once and use it with any model. Mb Verify is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Mb Verify in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/mrvladd-d/memobank/tree/main/skills/mb-verify. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Mb Verify?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Mb Verify?

As many as you like. As long as a model supports skills, you can use Mb Verify with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Mb Verify AI skill free?

Yes. It is published on GitHub by mrvladd-d under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

Overview

Installation

Install in TypingMind

Install in another agent

Use it in TypingMind

Works with any AI model

Skill instructions

mb-verify — Verifier loop (acceptance → evidence → verdict)

Goal

Status Transition Modes

Inputs

Preconditions

Required outputs

Status ownership

Process

1) Prime only what you need

2) Verify acceptance criteria

3) Verdict

4) Sync recommendations

Definition of done

Bundled files

Frequently asked questions

Related AI Skills

Set up your own AI workspace now