Skip to main content
Glama

Check the premises an action relies on

check_premises
Read-only

Call it before EVERY action that changes or sends anything in any system, listed in get_my_setup or not (D-053). Stage 1 of the pre-action guard. Given the action you are about to take and the PINNED subject ids it relies on, returns allow/hold/deny per premise plus a short-lived lease for each cleared one. Takes subject ids, never a natural-language question — retrieval wording cannot change the verdict. To check the DOCUMENT you work from, omit subject_id and pass the document from get_my_setup as relied_on with content_hash ("sha256:" + hex SHA-256 of the exact text you read): allow only when that is the version its owner confirmed. Do the work, then call verify_premises with these leases IMMEDIATELY before acting. Fails closed: an unknown pin, an unreadable record or a premise that cannot be leased all return hold, never allow. Give Rithmo time to answer: wait at least 10 seconds for this call. No answer in that time is not a yes: do not act, as for a hold.

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": {
      "intended_action": "<intended_action>",
      "premises": "<premises>"
    },
    "name": "check_premises"
  }
}

Input Schema

NameRequiredDescriptionDefault
runNoOptional: the run context from start_agent_run, naming WHICH of your agents this check is for. Pass it when your connection serves several agents. A run whose agent was retired, or whose installation was disconnected, is refused here — the governed call is denied, not just hidden.
surfaceNoOptional: the external system the action concerns, forwarded to the guard for observation scoping.
premisesYesThe configured premises the intended action relies on. Pinned ids only; this tool never searches.
resumes_holdNoOptional: the hold_id of YOUR earlier held call that this call checks again (after get_hold_status says a person closed it). Linked only when it is the same request: your own hold, the same intended_action and the same premises (role and subject). The response says whether it was linked. It never changes the verdict -- this is a fresh check.
action_fieldsNoOptional: the action's structured details, by name (e.g. {"amount": 900, "to": ["jane@example.com"]}). An agent whose rules limit a detail must send it; a detail its rules need that is missing is refused. The leases returned are bound to exactly these details: verify_premises must be given the same ones. Optional reserved detail "target": where the action happens, as "<system>:<place>[@<ref>]" (e.g. "github:owner/repo@refs/heads/main"), so what that system later shows was done there can be tied to this check; report the resulting id with report_action_outcome external_ref.
intended_actionYesWhat the caller is about to do, as a short stable string (e.g. "send the weekly report"). Hashed into the lease, so a lease acquired for one action cannot authorize a different one.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed7 schema fields changed
    • addedInput schema / properties / action_fields
      Added value: +{
      +  "additionalProperties": {
      +    "anyOf": [
      +      {
      +        "type": "string"
      +      },
      +      {
      +        "type": "number"
      +      },
      +      {
      +        "type": "boolean"
      +      },
      +      {
      +        "items": {
      +          "type": "string"
      +        },
      +        "type": "array"
      +      }
      +    ]
      +  },
      +  "description": "Optional: the action's structured details, by name (e.g. {\"amount\": 900, \"to\": [\"jane@example.com\"]}). An agent whose rules limit a detail must send it; a detail its rules need that is missing is refused. The leases returned are bound to exactly these details: verify_premises must be given the same ones. Optional reserved detail \"target\": where the action happens, as \"<system>:<place>[@<ref>]\" (e.g. \"github:owner/repo@refs/heads/main\"), so what that system later shows was done there can be tied to this check; report the resulting id with report_action_outcome external_ref.",
      +  "propertyNames": {
      +    "type": "string"
      +  },
      +  "type": "object"
      +}
    • changedInput schema / properties / intended_action / description
      Previous value: -"What the caller is about to do, as a short stable string (e.g. \"reply to customer support email\"). Hashed into the lease, so a lease acquired for one action cannot authorize a different one."New value: +"What the caller is about to do, as a short stable string (e.g. \"send the weekly report\"). Hashed into the lease, so a lease acquired for one action cannot authorize a different one."
    • addedInput schema / properties / premises / items / properties / relied_on
      Added value: +{
      +  "description": "The business resources this premise actually rests on. Rithmo resolves each one independently and decides whether it still represents current resolved context. Send pointers, not content: no claim, no answer text, no summary, no quotes.",
      +  "items": {
      +    "properties": {
      +      "content_hash": {
      +        "description": "Optional fingerprint of the EXACT content you consumed from this resource: \"sha256:\" followed by the lowercase hex SHA-256 of its UTF-8 bytes. Not a revision token and not the content itself. When supplied, Rithmo compares it with the content it currently holds for the resource and holds the premise if the resource has changed since you read it (content_check: mismatched) or if it cannot establish the current content (content_check: unverifiable).",
      +        "pattern": "^sha256:[0-9a-f]{64}$",
      +        "type": "string"
      +      },
      +      "provider": {
      +        "description": "The system the resource lives in. Supported: slack, google_drive, github.",
      +        "maxLength": 64,
      +        "minLength": 1,
      +        "type": "string"
      +      },
      +      "resource_id": {
      +        "description": "Provider-native identity of the exact resource, as that system names it — a Slack permalink or \"<channel id>:<message ts>\", a Google Drive file id, \"owner/repo#number\". Never a Rithmo id.",
      +        "maxLength": 512,
      +        "minLength": 1,
      +        "type": "string"
      +      },
      +      "revision_id": {
      +        "description": "Optional provider-native revision/version token for the exact version you read. Recorded as provenance; what it can prove varies by provider and is reported back as revision_assurance.",
      +        "maxLength": 256,
      +        "type": "string"
      +      }
      +    },
      +    "required": [
      +      "provider",
      +      "resource_id"
      +    ],
      +    "type": "object"
      +  },
      +  "maxItems": 8,
      +  "minItems": 1,
      +  "type": "array"
      +}
    • changedInput schema / properties / premises / items / properties / subject_id / description
      Previous value: -"The SoR subject id this premise is pinned to. Must be a real subject in this org."New value: +"The SoR subject id this premise is pinned to. Omit it to check one of YOUR OWN documents (a file of your instructions or resources, as get_my_setup lists them): relied_on is then that one file, provider \"github\", resource_id \"owner/repo:branch:path\", with revision_id (its git blob sha) or content_hash (sha256 of the text you read)."
    • changedInput schema / properties / premises / items / required
      Previous value: -[
      -  "role",
      -  "subject_id"
      -]New value: +[
      +  "role",
      +  "relied_on"
      +]
    • addedInput schema / properties / resumes_hold
      Added value: +{
      +  "description": "Optional: the hold_id of YOUR earlier held call that this call checks again (after get_hold_status says a person closed it). Linked only when it is the same request: your own hold, the same intended_action and the same premises (role and subject). The response says whether it was linked. It never changes the verdict -- this is a fresh check.",
      +  "format": "uuid",
      +  "pattern": "^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$",
      +  "type": "string"
      +}
    • addedInput schema / properties / run
      Added value: +{
      +  "description": "Optional: the run context from start_agent_run, naming WHICH of your agents this check is for. Pass it when your connection serves several agents. A run whose agent was retired, or whose installation was disconnected, is refused here — the governed call is denied, not just hidden.",
      +  "format": "uuid",
      +  "pattern": "^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$",
      +  "type": "string"
      +}
  2. Added

TDQS

A4.8/5.0
Behavior5/5

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

Goes well beyond the annotations (readOnly/idempotent/destructive hints), declaring fail-closed behavior for unknown pins and unreadable records, lease lifetimes, the 10-second wait, and the crucial 'no answer is not a yes' rule. It also warns that retrieval wording cannot change the verdict and that leases are bound to intended_action and action_fields.

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?

Front-loaded with the imperative call condition and the verdict shape, and nearly every sentence carries an operational rule. It is dense and jargon-heavy (D-053, 'PINNED', revision_assurance), and the document-check sentence packs multiple conditions into one long clause, costing some readability.

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 six-parameter, nested-schema tool with no output schema, the description still tells the agent what comes back (allow/hold/deny per premise, a lease per cleared premise, content_check and revision_assurance statuses) and what to do with it. Nothing needed to call it correctly is left implicit.

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 already 100%, so baseline would be 3; the description earns more by explaining the interactions the schema cannot: subject_id omitted means checking your own document (with provider 'github' and an 'owner/repo:branch:path' resource_id), content_hash vs revision_id semantics, and what resumes_hold links against.

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 specific verb and resource: 'Stage 1 of the pre-action guard', returning 'allow/hold/deny per premise plus a short-lived lease'. It also explicitly separates itself from verify_premises (stage 2) and from get_my_setup (pinned ids only, 'this tool never searches').

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?

Gives explicit when-to-use ('before EVERY action that changes or sends anything... listed in get_my_setup or not') and the follow-up sequencing ('Do the work, then call verify_premises with these leases IMMEDIATELY before acting'). It also carves out the alternative mode: to check your own source document, omit subject_id and pass relied_on with content_hash.

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

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.

Resources