Mb Execute

mb-execute - Implementation Handoff

Purpose

mb-execute implements one scoped JSON task. It is not a mini-scheduler and it does not own task closure. Flow: read task -> minimal preflight -> protocol -> implement -> local gates -> evidence -> handoff.

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.
• Scheduler must write the closure/failure/blocking decision, final task status, and evidence links to the authoritative indexed .memory-bank/tasks/TASK-*.task.json record before /mb-sync.
• /mb-sync records/reconciles already-written task state. It does not decide closure/failure/blocking/promotion and must not sync a decision that exists only in scheduler context.
• 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 for one TASK.
• Manual closure is allowed only when an explicit closure owner exists.
• explicit standalone owner means either the user directly asked the current top-level agent to close the task, or the top-level agent/orchestrator explicitly runs a manual workflow for one TASK and records that it owns closure. Subagents/worker prompts do not silently become closure owners.
• /verify PASS may mark T0 / T1 status: done only when explicit closure ownership is present and completed evidence has been written to the task record verify field and the compact/full protocol required by tier.
• If explicit closure owner is absent, /verify records VERDICT: PASS, evidence, and a closure recommendation, leaves status unchanged, and tells the scheduler/owner to close.
• T2 / T3 manual closure requires /verify PASS plus /red-verify / mb-red-verify SEMANTIC_VERDICT: semantic-pass before status: done or /mb-sync; if semantic-pass is absent, leave closure pending or blocked, not done.
• 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, for example TASK-123
• .memory-bank/tasks/index.json
• indexed .memory-bank/tasks/.task.json
• task tier: T0|T1|T2|T3
• feature or epic docs referenced by the task
• .memory-bank/requirements.md / REQ IDs when relevant
• .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

• richer task fields when present: source_artifacts, normative_inputs,

constraints, invariants, verification_targets

Authoritative SDD spec links are links 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/.

Missing richer fields or absent SDD spec links do not block classic T0 / T1 tasks; fall back to referenced docs. For T2 / T3, missing linked SDD specs are a blocker for serious work.

Minimal Preflight

Stop with an explicit handoff error if:

• task record/indexed file is missing
• the record id does not match TASK_ID
• tier is missing or is not T0|T1|T2|T3
• task status is blocked, failed, or done
• any depends_on task is missing or is not done
• tier is T2 / T3 and neither task richer fields nor linked feature

spec_design_links include relevant SDD spec links

Do not block T0 / T1 only because SDD spec links are absent. Route only by task.tier. Do not use legacy risk / risk.level.

Protocol Routing

Create .tasks// for runtime artifacts.

T0 / T1: use compact protocol:

• .protocols//run.md; record tier, goal, context, plan, changes,

gates, evidence, and handoff notes

• VERDICT: PASS|FAIL|BLOCKED is a local evidence marker only, not task closure

T2 / T3: use full protocol:

• .protocols//context.md, plan.md, progress.md,

verification.md, handoff.md

T3: exact marker lines are closure requirements:

• HUMAN_CHECKPOINT: done
• ROLLBACK_RECOVERY_NOTE: present

Record whether these markers are present or still needed; do not close the task.

Implementation Rules

• Keep scope bounded to the task and its acceptance criteria.
• For any tier, if the task record or linked feature contains authoritative SDD

spec links, read .memory-bank/spec-index.md and all linked authoritative SDD specs before editing; treat them as normative inputs.

• Record goal, non-goals, constraints, touched areas, and gates before broad edits.
• If fan-out is needed, use narrow non-overlapping worker scopes.
• Preserve unrelated changes and do not rewrite generated shared-* files.
• Keep protocol notes factual: what changed, what was checked, where evidence is.

Local Gates

Run relevant local gates from project instructions: lint/typecheck, unit tests, and integration/e2e checks only when relevant.

If a gate cannot run or fails, record command, result, evidence path, and the blocker in the protocol/handoff. Do not convert that into final task status.

Output / Handoff Contract

Finish with:

• changed files summary
• protocol file paths
• local gates run and results
• evidence paths under .tasks//
• verification targets and risk notes for verifier/reviewer
• MB-SYNC handoff notes: what should be synchronized and by whom
• recommended next owner: scheduler, verifier, red-verifier, explicit standalone owner, or human

Non-Ownership Rules

mb-execute never:

• closes tasks
• writes final task status
• runs /mb-sync
• runs /verify or /red-verify
• promotes, blocks, or unblocks dependents
• performs scheduler transitions such as in-progress, failed, or done

Schedulers (/autopilot, /autonomous) or an explicit standalone owner perform verification orchestration, final status decisions, MB-SYNC, and dependent promotion/blocking.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mrvladd-d
• Source: mrvladd-d/memobank
• License: MIT

Install and usage instructions live in the source repository linked above.