← Agent installation

Agent discovery · local selection

Workflows

Choose a common job, inspect its exact evidence, privacy boundary, tools, limits, steps, and limitations, then explicitly compile a canonical request. Selection is local and inert: nothing is run or sent, and Utilito adds no storage or history.

compile-only catalog

Choose an agent job

Explore 10 common agent jobs in canonical order. Search and filters stay in this page. Compilation creates request files only; nothing is run or sent.

10 jobs shown in canonical order.

Job 1 · run

Decide contract change safety

Decide whether an API/tool contract change is safe.

Outcome: Reconcile 1–5 related submitted contracts, references, and fixtures into a deterministic pass, review, or block packet with explicit unknowns.

Required evidence
  • Contract versions — Previous and current inline contracts, contract kinds, and directions.
  • Submitted fixtures — Bounded fixtures and explicit references.
  • Decision policy — Unknown and failure policy.
Privacy
server-submitted-response-only; browser-local search: true; compiler network: false; compiler storage: false; product collection: false; retention: response_only
Limits
5 steps; 262,144 compiled UTF-8 bytes; tool limits from API_TOOL_MANIFEST
Transport
run
Exact ordered recipe
  1. contract-reliability-workbenchStep 1: contract-review
    {
      "kind": "api-manifest-example",
      "tool_id": "contract-reliability-workbench",
      "example_index": 0
    }
Limitations
  • Conservative supported contract subset only; missing or unsupported semantics remain unknown.
Select exact recipe
Job 2 · run

Find a Utilito workflow

Find the correct Utilito workflow for a natural-language job.

Outcome: Use browser-local catalog search to choose a use case, then compile an inert launch package for an explicitly selected canonical tool and client profile.

Required evidence
  • Local query — Natural-language discovery query remains browser-local.
  • Explicit selection — Selected use-case ID, client profile, tool ID, and language.
  • Safe example (optional) — Optional submitted credential-free example input.
Privacy
browser-local-or-server-submitted-response-only; browser-local search: true; compiler network: false; compiler storage: false; product collection: false; retention: response_only
Limits
5 steps; 262,144 compiled UTF-8 bytes; tool limits from API_TOOL_MANIFEST
Transport
run
Exact ordered recipe
  1. integration-launch-kitStep 1: integration-launch
    {
      "kind": "api-manifest-example",
      "tool_id": "integration-launch-kit",
      "example_index": 0
    }
Limitations
  • Natural-language discovery is browser-local; compilation occurs only after explicit selection and never executes the launch request.
Select exact recipe
Job 3 · run

Diagnose an API or MCP failure

Diagnose a failed API or MCP call from submitted evidence.

Outcome: Diagnose sanitized submitted MCP/client configuration, protocol, capability, and transcript evidence and produce inert fixes and verification requests.

Required evidence
  • Client profile — Client profile and transport.
  • Sanitized configuration — Credential-free bounded configuration.
  • Transcript — Bounded submitted transcript and expected protocol version.
  • Policy — Strict diagnostic policy.
Privacy
browser-local-or-server-submitted-response-only; browser-local search: true; compiler network: false; compiler storage: false; product collection: false; retention: response_only
Limits
5 steps; 65,536 compiled UTF-8 bytes; tool limits from API_TOOL_MANIFEST
Transport
run
Exact ordered recipe
  1. agent-connection-diagnostic-kitStep 1: connection-diagnostic
    {
      "kind": "api-manifest-example",
      "tool_id": "agent-connection-diagnostic-kit",
      "example_index": 0
    }
Limitations
  • Current coverage is MCP/client-connection diagnostics, not the full API status, authentication, and rate-limit taxonomy planned for NX12.
Select exact recipe
Job 4 · run

Translate and validate tool calls

Translate and validate tool calls across OpenAI, Anthropic, and MCP shapes.

Outcome: Validate and convert one submitted OpenAI, Anthropic, MCP, or Utilito tool definition and optional call into another supported shape.

Required evidence
  • Tool definition and optional call — One bounded submitted definition and optional matching call.
  • Source and target formats — OpenAI, Anthropic, MCP, or canonical Utilito.
Privacy
browser-local-or-server-submitted-response-only; browser-local search: true; compiler network: false; compiler storage: false; product collection: false; retention: response_only
Limits
5 steps; 262,144 compiled UTF-8 bytes; tool limits from API_TOOL_MANIFEST
Transport
run
Exact ordered recipe
  1. tool-call-interoperabilityStep 1: tool-call-interoperability
    {
      "kind": "api-manifest-example",
      "tool_id": "tool-call-interoperability",
      "example_index": 0
    }
Limitations
  • Supports one definition and optional call through a conservative schema subset; unsupported semantics remain explicit unknowns and no tool is executed.
Select exact recipe
Job 5 · run

Prepare an agent context handoff

Produce a bounded context handoff another agent can trust.

Outcome: Build a dependency-closed Context Pack, evidence report, integrity manifest, and inert next-agent prompt with whole-record omission receipts.

Required evidence
  • Handoff document — Goal, status, decisions, completed work, blockers, next actions, artifacts, evidence, unknowns, constraints, and notes.
  • Byte budget — Explicit maximum Context Pack byte budget.
Privacy
server-submitted-response-only; browser-local search: true; compiler network: false; compiler storage: false; product collection: false; retention: response_only
Limits
5 steps; 262,144 compiled UTF-8 bytes; tool limits from API_TOOL_MANIFEST
Transport
run
Exact ordered recipe
  1. agent-context-handoffStep 1: context-handoff
    {
      "kind": "api-manifest-example",
      "tool_id": "agent-context-handoff",
      "example_index": 0
    }
Limitations
  • Submitted evidence and integrity references are not independently verified; no session, repository, filesystem, artifact reference, provider, model, or network access occurs.
Select exact recipe
Job 6 · run

Prepare retrieval evidence

Prepare retrieval evidence for a grounded model turn.

Outcome: Normalize already-retrieved caller-mapped records into bounded exact citation evidence, omissions, Context Pack, and report.

Required evidence
  • Retrieved records — Submitted records and explicit field mapping.
  • Preparation options — Ordering, dedupe, quote, context-budget, and report options.
Privacy
server-submitted-response-only; browser-local search: true; compiler network: false; compiler storage: false; product collection: false; retention: response_only
Limits
5 steps; 262,144 compiled UTF-8 bytes; tool limits from API_TOOL_MANIFEST
Transport
run
Exact ordered recipe
  1. rag-evidence-preparation-kitStep 1: retrieval-evidence
    {
      "kind": "api-manifest-example",
      "tool_id": "rag-evidence-preparation-kit",
      "example_index": 0
    }
Limitations
  • Evidence is structurally prepared, not judged true; no claim-to-evidence support graph until NX15.
Select exact recipe
Job 7 · run

Reduce a CI failure review

Reduce CI/build failures into an exact review pack.

Outcome: Turn one submitted CI log and run metadata into line-addressable failure evidence, bounded context, patch status, and report.

Required evidence
  • CI log — One submitted CI log and run metadata.
  • Verification status (optional) — Optional submitted patch-verification status and report options.
Privacy
server-submitted-response-only; browser-local search: true; compiler network: false; compiler storage: false; product collection: false; retention: response_only
Limits
5 steps; 262,144 compiled UTF-8 bytes; tool limits from API_TOOL_MANIFEST
Transport
run
Exact ordered recipe
  1. ci-failure-review-kitStep 1: ci-review
    {
      "kind": "api-manifest-example",
      "tool_id": "ci-failure-review-kit",
      "example_index": 0
    }
Limitations
  • One submitted log only; baseline/candidate regression comparison remains NX16.
Select exact recipe
Job 8 · run

Explain and resume an agent run

Explain an agent run and prepare a safe resume/replay handoff.

Outcome: Produce an ordered timeline, submitted-evidence failure review, conservative retry safety, blockers, next actions, and redacted handoff.

Required evidence
  • Run evidence — Failed or paused run metadata and ordered events.
  • Handoff evidence — Decisions, blockers, next actions, artifacts, evidence, and pack/report options.
Privacy
browser-local-or-server-submitted-response-only; browser-local search: true; compiler network: false; compiler storage: false; product collection: false; retention: response_only
Limits
5 steps; 131,072 compiled UTF-8 bytes; tool limits from API_TOOL_MANIFEST
Transport
run
Exact ordered recipe
  1. agent-run-forensics-handoff-kitStep 1: run-forensics
    {
      "kind": "api-manifest-example",
      "tool_id": "agent-run-forensics-handoff-kit",
      "example_index": 0
    }
Limitations
  • Resume safety is conservative and evidence-based; no checkpoint transition validator until NX17.
Select exact recipe
Job 9 · run

Gate web release evidence

Combine submitted web-release evidence into a deterministic gate.

Outcome: Build a six-domain release matrix from one submitted page artifact, image metadata, text artifacts, regression checks, and explicit policy.

Required evidence
  • Release identity — Submitted release identity.
  • Artifacts — Optional page HTML and headers, image metadata, and text artifacts.
  • Checks and policy — Regression checks and required or informational domain policy.
Privacy
browser-local-or-server-submitted-response-only; browser-local search: true; compiler network: false; compiler storage: false; product collection: false; retention: response_only
Limits
5 steps; 262,144 compiled UTF-8 bytes; tool limits from API_TOOL_MANIFEST
Transport
run
Exact ordered recipe
  1. web-release-qa-kitStep 1: release-gate
    {
      "kind": "api-manifest-example",
      "tool_id": "web-release-qa-kit",
      "example_index": 0
    }
Limitations
  • No baseline/candidate release delta until NX18.
Select exact recipe
Job 10 · batch

Deliver verifiable artifacts

Deliver artifacts with checksums, manifests, verification instructions, and generated conformance fixtures.

Outcome: Independently assemble a release evidence packet, compare submitted checksum metadata, and generate inert schema, provider, and conformance artifacts.

Required evidence
  • Release packet — Release checks and artifact manifest.
  • Digest metadata — Expected and submitted digest metadata.
  • Tool schema — Strict tool schema and typed failures.
Privacy
browser-local-files-or-server-submitted-metadata-only; browser-local search: true; compiler network: false; compiler storage: false; product collection: false; retention: response_only
Limits
5 steps; 262,144 compiled UTF-8 bytes; tool limits from API_TOOL_MANIFEST
Transport
batch
Exact ordered recipe
  1. release-evidence-kitStep 1: release-packetCaller ID: release-packet
    {
      "kind": "api-manifest-example",
      "tool_id": "release-evidence-kit",
      "example_index": 0
    }
  2. file-verification-handoff-kitStep 2: checksum-handoffCaller ID: checksum-handoff
    {
      "kind": "api-manifest-example",
      "tool_id": "file-verification-handoff-kit",
      "example_index": 0
    }
  3. tool-schema-publishing-kitStep 3: conformance-fixturesCaller ID: conformance-fixtures
    {
      "kind": "api-manifest-example",
      "tool_id": "tool-schema-publishing-kit",
      "example_index": 0
    }
Limitations
  • Hosted checksums are caller-submitted and unverified. Real digest recomputation remains browser-local; full NX19 binary delivery is not claimed.
Select exact recipe

Explicit local compilation

Compile a selected request

No recipe selected. Use an ordinary “Select exact recipe” link above; opening a link does not compile anything.

Select a recipe to begin. No automatic compilation occurs.

Not compiled.
Download JSONDownload TypeScript

Complete generated catalogs: workflow JSON · generated TypeScript. These are data-only catalogs, not execution clients.

submitted-input products

Reconcile evidence, runtime state, configuration, data, and provider requests

These browser-local workbenches use the same strict pure cores as their HTTP and MCP tools. Loading a route or sample remains inert until explicit submit.

Claim-to-Evidence Coverage WorkbenchReconcile explicit claims and prepared evidence through caller-authored mappings.claim-evidence-coverageCI Failure Regression Diff WorkbenchCompare baseline and candidate CI failure signatures.ci-regression-diffAgent Run State and Resume ValidatorValidate finite run transitions and checkpoint references without replay.agent-run-state-validatorRelease Evidence Delta and Promotion GateCompare compatible release evidence through an explicit promotion policy.release-evidence-deltaArtifact Delivery Bundle and Agent Conformance KitBuild truthful integrity facts and deterministic fixtures from submitted metadata and inline text.artifact-delivery-manifestConfiguration Drift Reconciliation WorkbenchCompare two caller-owned configuration snapshots.configuration-drift-reconciliationAPI Fixture Matrix and Negative-Case PackGenerate deterministic valid and one-fault invalid fixtures.api-fixture-matrixBatch Retry Safety PlannerBuild a conservative dependency-aware manual retry plan.batch-retry-safety-plannerStructured Data Reconciliation WorkbenchCompare bounded structured datasets through explicit keys.structured-data-reconciliationProvider Request Envelope Production KitAssemble one complete inert provider request envelope.provider-request-envelopeJSON Schema Evolution Compatibility WorkbenchCompare strict submitted schema versions conservatively.json-schema-compatibility-diffOpenAPI Surface Delta WorkbenchCompare allowlisted submitted OpenAPI 3.1 operations.openapi-surface-deltaTool-Call Trace Conformance WorkbenchValidate submitted ordered tool-call traces without execution.tool-call-trace-conformanceAgent Capability Policy AuditorAudit caller-declared capabilities against explicit policy.agent-capability-policy-auditorCitation and Quote Integrity WorkbenchCheck exact quotes and Unicode code-point ranges in submitted sources.citation-quote-integrityEvaluation Run Delta WorkbenchCompare submitted scalar evaluation runs without execution or semantic grading.evaluation-run-deltaDependency Manifest Reconciliation WorkbenchReconcile bounded submitted dependency manifests without registry access or installation.dependency-manifest-reconciliationEnvironment Contract CompilerCompile inert environment declarations without accepting or reading values.environment-contract-compilerPolicy Decision Table CompilerCompile finite submitted rules into an inert deterministic decision table.policy-decision-tableAPI Pagination Contract PlannerPlan bounded inert pagination clients from placeholder contracts.api-pagination-contract-plannerHTML to Agent Markdown WorkbenchConvert submitted inert HTML into deterministic semantic Markdown without rendering or fetching it.html-to-agent-markdownContract Conformance Repair WorkbenchApply only caller-declared JSON Pointer moves, defaults, and coercions to a finite closed target contract.contract-conformance-repairUntrusted Input Admission WorkbenchAssess opaque serialized JSON against a strict structural profile without reflecting submitted keys or values.untrusted-input-admissionSensitive Log Redaction Evidence WorkbenchRedact supported credential signatures browser-locally and validate only raw-free evidence metadata through hosted surfaces.sensitive-log-redaction-evidenceTool Catalog Context Budget AuditorAudit a submitted tool catalog and build a deterministic greedy capability-covering subset under explicit context budgets.tool-catalog-context-budget-auditorThree-Way Structured Merge WorkbenchMerge submitted baseline, ours, and theirs JSON with explicit conflict resolutions and reversible patches.three-way-structured-mergeAgent Plan Dependency & Parallelism AuditorAudit a submitted task graph for dependencies, declared concurrency conflicts, approval serialization, and critical path without execution.agent-plan-parallelism-auditorChange-Impact & Test Selection PlannerPlan impact closure and deterministic greedy tests from submitted dependency, coverage, risk, and mandatory mappings.change-impact-test-selection-plannerTest Run Stability MatrixReconcile ordered submitted test runs into stability classifications, transitions, exact durations, and a deterministic review queue.test-run-stability-matrixSARIF Findings Reconciliation WorkbenchReconcile closed submitted SARIF 2.1.0 fragments through exact fingerprint and fallback rule-location identities.sarif-findings-reconciliationStreaming Response & Tool-Call AssemblerAssemble finite submitted stream fragments into provider-neutral text and parsed tool-call evidence without provider calls or execution.streaming-tool-call-assemblerSubagent Delegation Contract CompilerCompile finite submitted task, ownership, acceptance, budget, stop, and escalation declarations without spawning workers.subagent-delegation-contract-compilerArtifact Lineage Integrity AuditorAudit submitted artifact lineage graphs while separating computed inline-text digests from unverified caller claims.artifact-lineage-integrity-auditor