# Execution Alignment Gate > Detects materially ambiguous or under-specified requests, selects the right clarification target, and enforces bounded alignment before execution, including spec-governed continuations and approval-gated manager handoffs. Use when ambiguity could cause wrong deliverables, wrong scope, wrong implementation path, avoidable rework, or token waste from repeated follow-up. - **Type:** Skill - **Install:** `agentstack add skill-mwillbanks-agent-skills-execution-alignment-gate` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [mwillbanks](https://agentstack.voostack.com/s/mwillbanks) - **Installs:** 0 - **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills) - **Latest version:** 0.1.0 - **License:** Apache-2.0 - **Upstream author:** [mwillbanks](https://github.com/mwillbanks) - **Source:** https://github.com/mwillbanks/agent-skills/tree/main/.agents/skills/execution-alignment-gate ## Install ```sh agentstack add skill-mwillbanks-agent-skills-execution-alignment-gate ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # execution-alignment-gate Use this skill when ambiguity is material enough that proceeding without alignment is likely to produce the wrong result, force rework, or waste tokens through repeated follow-up. Do not use this skill as a crutch for failing to read the specification, follow an approved plan, obey repository rules, or comply with an active execution mode already in force. ## Primary objectives * Reduce materially harmful ambiguity before implementation starts. * Reduce wrong-path execution. * Reduce token waste from avoidable rework and repeated clarification. * Improve clarity and alignment while preserving forward progress. * Keep clarification bounded, concise, and high leverage. ## Non-objectives * Maximizing certainty. * Asking every plausible question. * Replacing execution discipline. * Interrupting obvious continuation messages. * Adding approval ceremonies to trivial or low-risk work. ## Default posture Default posture is **conservative**. Only activate when ambiguity is likely to cause one or more of the following: * wrong deliverable * wrong scope * wrong implementation path * wrong validation target * avoidable rework * materially increased token usage from likely follow-up and correction Posture may be elevated to **moderate** or **aggressive** only through explicit invocation or a higher-priority instruction. Do not self-escalate posture by agent whim. ## Workflow Follow this sequence: 1. Perform a silent ambiguity assessment before asking anything. 2. Check [activation rules](references/ACTIVATION_RULES.md) and [bypass rules](references/BYPASS_RULES.md). 3. If the request is clearly a continuation, approval, or low-risk correction, bypass this skill and proceed. 4. If alignment is required, route to the correct target mode using [target modes](references/TARGET_MODES.md). 5. Ask the fewest high-leverage questions needed using [question patterns](references/QUESTION_PATTERNS.md). 6. After answers arrive, synthesize a concrete plan, lock any remaining material assumptions, and apply [approval flow](references/APPROVAL_FLOW.md) when required. 7. If the user is unavailable, apply [unattended behavior](references/UNATTENDED_BEHAVIOR.md). 8. If clarification reveals no true blocker, proceed. Do not reopen broad clarification loops. 9. Apply [token economics](references/TOKEN_ECONOMICS.md) and avoid spending more clarification tokens than the expected rework would cost. ## Hard limits * Prefer 1 to 3 questions. * Maximum 5 questions in a clarification round. * Maximum 4 options per question, plus freeform when helpful. * Allow at most one blocker-only follow-up after the bounded clarification round. * Do not open repeated wait windows in unattended mode. * Do not retain verbose per-run narrative history. ## Required behaviors * Clarify only when answers materially affect execution. * Treat approved spec or plan updates as governing context; do not reopen alignment for the same decision unless a new material contradiction appears. * Prefer safe defaults when ambiguity is minor and risk is low. * Use short assumption blocks when proceeding under defaults, refusal, or unattended fallback. * For manager escalation, use the compact blocked packet in [assets/TEMPLATE_MANAGER_PACKET.md](assets/TEMPLATE_MANAGER_PACKET.md). * For plan confirmation, use the compact structure in [assets/TEMPLATE_PLAN_CONFIRMATION.md](assets/TEMPLATE_PLAN_CONFIRMATION.md). * For assumption locking, use [assets/TEMPLATE_ASSUMPTIONS_BLOCK.md](assets/TEMPLATE_ASSUMPTIONS_BLOCK.md). ## Prohibited behaviors * Do not blame the user for ambiguity. * Do not defend the model. * Do not mirror hostile tone. * Do not ask the user to restate the whole task. * Do not ask questions already answered by context. * Do not use this skill to compensate for ignoring specs, plans, repo rules, or execution mode. * Do not ask low-value questions that do not change execution materially. * Do not demand approval when it adds no value. * Do not ask the user for a manager-owned decision simply because a subagent is approval-gated or unavailable; route through manager mode instead. * Do not retain large clarification logs or other garbage that costs more than it saves. ## Complement to execution mode Apply this skill before implementation when material ambiguity would otherwise force low-confidence execution. Do not use it to excuse failure to follow existing plans, repository rules, or execution-mode requirements already in force. See also: * [activation rules](references/ACTIVATION_RULES.md) * [target modes](references/TARGET_MODES.md) * [bypass rules](references/BYPASS_RULES.md) * [question patterns](references/QUESTION_PATTERNS.md) * [approval flow](references/APPROVAL_FLOW.md) * [unattended behavior](references/UNATTENDED_BEHAVIOR.md) * [failure modes](references/FAILURE_MODES.md) * [token economics](references/TOKEN_ECONOMICS.md) * [examples](references/EXAMPLES.md) ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [mwillbanks](https://github.com/mwillbanks) - **Source:** [mwillbanks/agent-skills](https://github.com/mwillbanks/agent-skills) - **License:** Apache-2.0 Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-mwillbanks-agent-skills-execution-alignment-gate - Seller: https://agentstack.voostack.com/s/mwillbanks - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.