Skip to main content
Glama
Verigent-AI

@verigent/mcp-server

Official
by Verigent-AI

continue_run

Drive a verification run to completion by submitting battery answers and evaluation responses as phases require.

Instructions

Drive a verification run to completion — the ONE tool to loop after start_verification. Verigent drives the test; you just do what each response's next_action says and call continue_run again. Phases it walks you through: it returns the battery tasks (answer them in chunks of ~10 as each is ready, rather than waiting to collect them all), then the multi-turn evaluation scenarios (respond to each in character — this is where memory, governance-under-pressure and sycophancy-resistance are measured), then done: true. Grading happens server-side IN THE BACKGROUND per chunk and completes on its own within a few minutes (a backstop drives it whether or not you poll) — you do NOT need to loop or set timers waiting for it. Note: the FIRST call starts the battery, so it's best to share the live tracker link from start_verification with your operator first, so they can watch grading progress there. The first (battery) response returns tasks grouped one content block per dimension rather than one giant block (K-43a) — read every block, not just the first. Supply { answers } after a 'battery' phase (a partial chunk is fine — call again with more as they're ready; idempotent per task_id) and { eval_responses } after each 'eval' phase — when a phase returns several scenarios at once, send every ready response together in the same eval_responses array in one call, since they're graded concurrently server-side and there's no need for one call per scenario. Once your answers and all scenarios are in, the run finishes on its own — call continue_run just ONCE more after a few minutes to confirm completion, rather than polling repeatedly. Each call's combined answers/eval_responses payload is capped around 32 KB (32768 bytes) (K-43a) — a call over that limit gets back a 413 naming the exact cap; split into smaller chunks and resend only what didn't go through, never the same oversized payload unmodified. run_token is optional: omit it and this tool falls back to the run_token this server saved locally when start_verification last ran (/.verigent/state.json) — so a cold session can call continue_run directly with no other setup. If nothing was saved, pass run_token explicitly or call resume_run.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
answersNoBattery answers — supply after the 'battery' phase, one entry per task_id
run_tokenNoRun token from start_verification. Optional — omitted, falls back to the run_token this server saved locally at start_verification.
eval_responsesNoMulti-turn evaluation responses — supply after an 'eval' phase, one entry per scenario_id. Send every ready response together in a single array call; they're graded concurrently server-side, so there's no need to call once per scenario.

Schema Changelog

Changes observed during successful MCP inspections.

  1. First observedv0.7.15

TDQS

A5/5.0
Behavior5/5

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

With no annotations provided, the description carries the full burden and exceeds it. It discloses that grading happens server-side in the background, that polling is unnecessary, that the first call starts the battery, that answers are idempotent per task_id, and that there's a payload cap with a 413 error. Also explains the fallback behavior for run_token. This is thorough and transparent.

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

Conciseness5/5

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

The description is long but every sentence earns its place. It's front-loaded with the core loop instruction, then systematically covers phases, payload limits, and fallback. Despite length, it is organized and free of fluff, effectively guiding the agent through a complex interaction.

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?

Covers all necessary context: phases, what to send when, idempotency, background grading, payload cap with error handling, run_token fallback, and even advises sharing the tracker link. It also mentions the output behavior (done: true) and the multi-block response format. Nothing an agent needs to call this tool correctly is missing.

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

Parameters5/5

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

Schema coverage is 100%, so the schema already documents parameters well. However, the description adds crucial semantics beyond the schema: explains how answers and eval_responses are structured in practice, the chunking behavior, idempotency, concurrent grading of eval_responses, and the payload cap. It clarifies the optional run_token fallback. This significantly enriches the parameter meaning.

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 the tool's purpose precisely: 'Drive a verification run to completion' and explicitly names itself as 'the ONE tool to loop after start_verification'. Clearly distinguishes from siblings by referencing start_verification and resume_run, so an agent knows exactly which tool this is among the list.

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?

Provides explicit when-to-use guidance: 'loop after start_verification', describes the phases and what to do in each (battery tasks, eval scenarios, done:true). Also names alternatives like resume_run for cases where no token is saved, and explains when to call once more after completion. No ambiguity.

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