Skip to main content
Glama

set_human_gate

DestructiveIdempotent

Arm a human gate on a task to freeze it and record who is waiting, what was asked, and what happens if nobody answers.

Instructions

Arm the human gate on a task: freeze it and record WHO is waiting, WHAT was asked, and WHAT you will do if nobody answers. To end your turn on a question to a person, use ask_human: it posts the question, arms this gate and records your wait in one step; this tool arms a gate without ending your turn. A wait on an agent is ask_agent. Records the whole ask on the task (who is waiting, what was asked, the default), so nothing has to re-read the thread; the '❓ Blocking @' comment marker is still recognised. recommended_default is REQUIRED: a gate with no stated default can only ever be resolved by finding a human. The call takes four answers (credential_exists / reversible / blocked_by_other_task / customer_visible_now), each with one line of justification. The server refuses the arm when the answers show nobody needs to be asked: the credential is held, the action is reversible, and nothing a customer sees or pays changes right now.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "task_id": "<task_id>",
      "reason": "<reason>",
      "recommended_default": "<recommended_default>",
      "addressee": "<addressee>",
      "credential_exists": "<credential_exists>",
      "credential_reason": "<credential_reason>",
      "reversible": "<reversible>",
      "reversible_reason": "<reversible_reason>",
      "blocked_by_other_task": "<blocked_by_other_task>",
      "blocked_reason": "<blocked_reason>",
      "customer_visible_now": "<customer_visible_now>",
      "customer_reason": "<customer_reason>"
    },
    "name": "set_human_gate"
  }
}

Input Schema

NameRequiredDescriptionDefault
classNo'hard' (default, never auto-released) or 'soft' (released by timeout — the release does NOT answer the question).
reasonYesThe question itself, in your own words.
task_idYesTask ID to gate.
deadlineNoRFC3339 timestamp when recommended_default applies. Omit for no deadline.
addresseeYes@name of the PERSON being asked. Must be a human in the workspace; an agent or unknown name is refused (gate_addressee_not_human). Agents are reached by assign_task, not by a gate.
copy_tierNoOnly when this ask is about VISIBLE PRODUCT COPY (a label, a page's prose, a message users read) — otherwise omit entirely. Answer three questions: (1) is this the company's voice going OUT, or a caption inside the interface? (2) would someone who never opened this screen notice the change? (3) does the text carry a promise — legal, price, product? Any "yes" → tier "A" (external/legal/marketing copy, a promise) and the workspace owner decides. All "no" → tier "B" (a field caption, menu item, system or validation message, section name, an existing dictionary string) and the product lead ships it without asking — the workspace owner sees it after the fact. The server refuses tier "B" outright: ship it yourself, tag the task `copy:b`, quote the exact string in your closing comment. Omitting copy_tier on an ask that reads as copy-approval is refused too, naming this field — state the tier and retry.
reversibleYesIs there a rollback anchor — git revert, backup, snapshot, image tag? If you can MANUFACTURE one (take a backup first), the answer is true.
supersedesNoOnly when the previous question on this task already has an answer: say what that answer does not cover. Without it the server refuses a repeat ask (repeat_ask_refused).
canonical_keyNoOptional stable key naming this question. If a human decision is already recorded under the same key, the server refuses the ask (repeat_ask_refused).
blocked_reasonYesOne line: which card, or why none.
customer_reasonYesOne line: what the customer would see, or why nothing changes for them now.
credential_existsYesDo you ALREADY hold the credential or access this needs? Check your team's credential store before answering false — a service account your team created for its agents is yours to use.
credential_reasonYesOne line: which credential, and where you checked.
reversible_reasonYesOne line: the exact rollback path, or why none exists.
recommended_defaultYesWhat you will do if nobody answers. Required — an ask with no default cannot time out.
customer_visible_nowYesDoes this change what a customer SEES or PAYS right now? A disabled gateway, an inactive flag or a reversible migration is NOT customer-visible; a rate that prints on invoices people already download is.
blocked_by_other_taskYesIs the thing you are waiting on actually ANOTHER card? If yes, the answer is add_dependency, not a gate — a blocks edge freezes the feed without adding anything to a human's queue.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed4 schema fields changedv0.1.6
    • addedInput schema / properties / addressee
      Added value: +{
      +  "description": "@name of the PERSON being asked. Must be a human in the workspace; an agent or unknown name is refused (gate_addressee_not_human). Agents are reached by assign_task, not by a gate.",
      +  "type": "string"
      +}
    • addedInput schema / properties / canonical_key
      Added value: +{
      +  "description": "Optional stable key naming this question. If a human decision is already recorded under the same key, the server refuses the ask (repeat_ask_refused).",
      +  "type": "string"
      +}
    • addedInput schema / properties / supersedes
      Added value: +{
      +  "description": "Only when the previous question on this task already has an answer: say what that answer does not cover. Without it the server refuses a repeat ask (repeat_ask_refused).",
      +  "type": "string"
      +}
    • changedInput schema / required
      Previous value: -[
      -  "task_id",
      -  "reason",
      -  "recommended_default",
      -  "credential_exists",
      -  "credential_reason",
      -  "reversible",
      -  "reversible_reason",
      -  "blocked_by_other_task",
      -  "blocked_reason",
      -  "customer_visible_now",
      -  "customer_reason"
      -]New value: +[
      +  "task_id",
      +  "reason",
      +  "recommended_default",
      +  "addressee",
      +  "credential_exists",
      +  "credential_reason",
      +  "reversible",
      +  "reversible_reason",
      +  "blocked_by_other_task",
      +  "blocked_reason",
      +  "customer_visible_now",
      +  "customer_reason"
      +]
  2. First observedv0.1.1

TDQS

A4.6/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Beyond the annotations (destructiveHint=true, idempotentHint=true) it discloses real behavior: the task is frozen, the whole ask is persisted on the task, the server refuses the arm under stated conditions, and a gate without a default 'can only ever be resolved by finding a human.' The one gap is that it never says how a gate is released or undone, which matters for a tool flagged destructive.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Well front-loaded: the core action comes first, then the sibling routing, then the recording behavior, then the required-default rule and the refusal logic. Dense but each sentence carries information; the closing sentence about the four answers is slightly compressed for how much rule it encodes.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a 12-required-parameter destructive mutation with no output schema, the description supplies the missing decision logic: what the tool freezes, what it persists, the refusal conditions, and why the default is mandatory. Remaining optional parameters are documented in the schema, so nothing an agent needs to call it correctly is absent.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so the baseline is 3, but the description adds genuine modeling on top: recommended_default is framed as REQUIRED with the consequence of omitting it, and the four boolean answers are explained as a set whose combination triggers a server refusal. It does not touch class, deadline, supersedes, or canonical_key, which the schema covers on its own.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a precise verb and resource: 'Arm the human gate on a task: freeze it and record WHO is waiting, WHAT was asked, and WHAT you will do if nobody answers.' It immediately distinguishes itself from ask_human ('arms a gate without ending your turn') and ask_agent ('A wait on an agent is ask_agent'), so an agent can pick it without reading either sibling's schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicit routing rules: use ask_human to end your turn on a question to a person, use this tool to arm without ending the turn, use ask_agent for agent waits, and use add_dependency when the blocker is another card (stated in the blocked_by_other_task schema text and reinforced by the 'nobody needs to be asked' refusal rule). It even defines when NOT to arm: credential held, action reversible, nothing customer-visible.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.