More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Agentic Entity Resolution Guardrail Planner

    1

    It can propose a MERGE, SPLIT, UNMERGE, or another change to a persistent entity identity.

    $49

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Agentic Entity Resolution Guardrail Planner

    Agentic Entity Resolution Guardrail Planner

    Example session with this skill installed

    I am designing an AI-assisted CRM entity-resolution workflow.

    Our AI agent has identified two customer-account records that may represent the same legal entity and proposes a MERGE.

    Synthetic scenario — all names and identifiers are fictional:

    • Record A: Example Ridge Services Ltd (CRM-ACCT-1042)
    • Record B: Example Ridge Service Holdings Ltd (CRM-ACCT-8891)
    • Entity class: Customer account / legal-entity identity
    • Shared signals: similar company names, the same web domain, one overlapping contact, and a registry extract containing a potentially matching identifier.
    • A contract document also suggests that one company may be a parent or holding entity of the other.
    • The registry extract has not yet been verified for currency, scope, or applicability to the identity proposition.

    The proposed MERGE could affect CRM account records, identity links, contacts, contracts, billing references, search indexes, and downstream integrations.

    Our current documentation does not establish:

    • The complete persistent write-set.
    • Which effects are covered by the same transaction.
    • The actual commit boundary and atomicity guarantees.
    • Whether post-commit recovery or compensation has been successfully tested.
    • The applicable operation-specific approval policy and authorized approvers.

    Please use the Agentic Entity Resolution Guardrail Planner to produce a complete engineering planning package for this scenario.

    Use synthetic assumptions only where necessary and label them explicitly. Do not treat matching signals as proof of legal-entity identity. Distinguish verified facts, customer-provided information, inferences, assumptions, and unresolved decisions.

    Specify the evidence requirements, authorization paths, commit-boundary controls, failure handling, implementation design, and operation-specific validation scenarios.

    Where write-set completeness, transaction coverage, atomicity, or required authorization cannot be established, specify the appropriate HOLD conditions before any persistent production write.

    Do not execute or authorize any mutation. Produce the complete set of deliverables defined by the Skill's output contract.

    • Read your context and instructions
    • Compiled the agentic entity resolution

    Original Claude agent output:

    I've produced the planning package for your CRM entity-resolution scenario: the nine plan files plus the separate ledger sidecar. It is design-time only. No mutation was executed or authorized, and I did not decide whether the two records are the same legal entity.

    Status: not_implementation_ready, with 11 blocking decisions (GAP-01 to GAP-11). A MERGE should not proceed.

    Verified: I confirmed the public reference service is available and retrieved its mutation profile for a production MERGE. I could not verify any of your customer facts; the records, registry extract and contract were described but not supplied, so all of that stays unverified.

    Not run: Schema validation, format validation and the format negative control were not run, because the sandbox had no Draft 2020-12 validator (only Ajv 6, which is Draft-07). I ran my own ad hoc checker and scripted semantic checks instead, which are recorded as separate checks, not as schema validation.

    Sidecar: The ledger records package_coherence: blocked and NOT_ATTESTED. The sandbox had no RFC 8785 canonicalization tool, so its integrity comparison could not run. Exports were not inspected.

    Key conclusions

    The proposition may be false. The contract suggesting a parent or holding entity points toward two distinct legal entities. The plan treats it as an unresolved contradiction (CTRL-04, GAP-07). The alternative of a typed parent link or keeping both accounts would be a separate operation outside this plan.

    Classification: OP-01 is a protected production MERGE (IRREVERSIBLE_WRITE), with factual reversibility unknown. No isolated staging target was described.

    Evidence: Similar names, the same domain and one shared contact are treated as hypotheses only. The registry extract stays unverified until its currency, scope and applicability are checked.

    HOLD conditions: The plan has 14 controls and 31 proposed tests. Nine hold conditions apply before any persistent write, and six are in effect today.

    • Write set: The complete write set is unverified.
    • Atomicity: Transaction coverage, commit point and atomicity are unknown.
    • Approval: No approval policy or authorized approvers are defined.

    Authorization: Neither Path A (documented, tested recovery plus attestation) nor Path B (explicitly accepted non-recovery risk) is established.

    • Evidence: The identity evidence is insufficient and unverified.
    • Interval: No validity interval is defined.

    Failure handling: Abort before commit is kept separate from post-commit compensation or quarantine, with idempotent reconciliation and no blind retry.

    Caveats

    • All identifiers are synthetic. The call arguments I sent to the reference service held abstract labels only.
    • The evidence-promotion and commit-boundary contracts were not retrieved separately, so those statements rely only on fields inside the profile.
    • My first sidecar assembly crashed on a bug in my own checker, after the README freeze write but before the sidecar was written. No frozen bytes changed, but the sidecar records that the post-freeze re-hash was repeated afterwards.

    Everything is in /mnt/user-data/outputs/: the nine plan files are in the agentic-identity-guardrail folder, and the ledger sits beside it.

    Planning deliverables

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Turn an AI-proposed identity change into an engineering decision

    Your AI agent can identify likely duplicate entities. It can propose a MERGE, SPLIT, UNMERGE, or another change to a persistent entity identity.

    But a similarity score, an LLM explanation, or a confident match is not an implementation decision.

    Agentic Entity Resolution Guardrail Planner helps engineering teams turn an agentic entity-resolution workflow into a structured, reviewable engineering plan for the controls required before a proposed identity change can reach persistent production state.

    It is built for teams designing AI-assisted CRM, Master Data Management (MDM), entity registry, corporate-entity resolution, and knowledge-graph workflows.

    What does it produce?

    Provide a description of your agentic entity-resolution workflow. The Skill produces a structured engineering package covering:

    • Mutation scope: application, entity class, identity proposition, operation type, and persistence boundary
    • Evidence and provenance: what must be evidenced, attributed, current, applicable, and sufficient for the specific operation
    • Authorization and recovery paths: operation-specific approval requirements, separation of duties, and recovery or compensation options
    • Commit-boundary controls: write-set completeness, transaction boundaries, atomicity, state and policy version checks, revalidation, approvals, and pre-commit HOLD conditions
    • Failure and recovery design: pre-commit abort, post-commit compensation or quarantine, ambiguous outcomes, reconciliation, and avoidance of blind retries
    • Implementation guidance: a stack-neutral design first, with technology-specific mapping only where supported by the customer's stated environment
    • Validation scenarios: operation-specific tests and explicit unresolved implementation decisions

    A complete run produces nine core planning deliverables:

    • 00-README.md — package overview, provenance, validation status, and handoff information
    • 01-mutation-scope.md — application, entities, operations, and mutation scope
    • 02-guardrail-specification.md — operational controls and guardrail requirements
    • 03-evidence-requirements.md — evidence claims, sources, and requirements
    • 04-commit-boundary-design.md — pre-persistence controls and commit-boundary design
    • 05-implementation-blueprint.md — stack-neutral implementation guidance
    • 06-validation-scenarios.md — operation-specific validation scenarios
    • 07-assumptions-and-gaps.md — unresolved decisions, assumptions, and missing dependencies
    • guardrail-spec.json — machine-readable specification for downstream implementation

    An additional artifact-creation-ledger.json may be produced as an auxiliary handoff record documenting artifact creation and provenance. It is not a core planning deliverable.

    Built for engineering handoff

    The output is designed to support actual engineering review and downstream implementation work.

    • Reviewable documentation: Markdown artifacts support engineering, governance, security, and product review.
    • Machine-readable specification: JSON provides a structured basis for downstream implementation work.
    • Decision provenance: distinguishes customer-provided facts, inferences, assumptions, and unresolved decisions.
    • Fail-closed planning: requires a HOLD before persistent production writes when write-set completeness, transaction boundaries, or atomicity remain unverified.
    • Explicit implementation gaps: identifies customer decisions, dependencies, and validation work that remain before implementation.

    This can substantially reduce the manual effort required to design identity-mutation guardrails. Rather than coordinating a new specification across AI engineering, data governance, security review, and platform implementation, teams receive a structured starting point that can be reviewed, adapted, and handed off.

    Why is a confident match not enough?

    An entity-resolution agent may reasonably propose that two records represent the same customer, company, account, or graph node.

    That proposal still does not answer the questions that determine whether a production identity change is safe to implement:

    • What identity proposition must be established?
    • Which evidence is admissible, current, attributable, and applicable?
    • How should conflicting, stale, incomplete, or uncertain evidence be handled?
    • Who is authorized to approve a protected identity-changing operation?
    • Is the complete write-set known?
    • Are transaction boundaries and atomicity established?
    • What happens if the operation fails before or after commit?
    • How should an ambiguous commit outcome be reconciled without a blind retry?
    • Which validation scenarios should be designed before deployment?

    The Planner makes these questions explicit, structured, and reviewable before implementation.

    Methodology: created by the author of DIO/ODI

    Agentic Entity Resolution Guardrail Planner was created by Daniel Beránek, author of Digital Identity Optimization (DIO) and the Ontology of Digital Identity (ODI).

    The Skill is not a generic prompt template. It operationalizes an evolving research and design framework for reasoning about digital identity, evidence, authorization, and identity-changing operations.

    Its planning logic is designed around the central distinction between an AI agent’s proposed entity match and the engineering controls required before that proposal can be connected to a persistent production change.

    The author also maintains the optional DIO/ODI Reference MCP used in reference-backed mode. This creates a coherent planning workflow: the Skill, its output contracts, and the reference service are designed to work together.

    Related DIO/ODI work

    Explore other DIO/ODI research and publications

    The Skill translates the central idea of that work into a practical planning workflow: separate the evidence available for agent reasoning from the evidence and controls required before a persistent identity-changing mutation.

    Optional reference-backed mode

    The Skill can operate in a customer-context-only mode without an external reference service.

    It can also use the public, read-only DIO/ODI Reference MCP when the buyer connects it separately in a compatible agent environment. The MCP provides reference profiles, contracts, concepts, and mappings. It does not host customer application data or authorize customer writes.

    Do not send real customer records, personal data, proprietary evidence, credentials, or secrets to the public reference MCP. Use abstract context only.

    Who is it for?

    This Skill is intended for:

    • AI agent developers and engineering teams
    • Entity-resolution engineers
    • MDM architects and data-platform engineers
    • Data governance architects
    • Technical integrators and solution architects
    • Teams building AI-assisted CRM, entity registry, or knowledge-graph workflows

    It is particularly useful when an AI agent can propose—or may eventually initiate—persistent changes to entity identity.

    Get a structured starting point

    Use Agentic Entity Resolution Guardrail Planner to turn an agent-proposed MERGE, SPLIT, UNMERGE, or other identity-changing workflow into a reviewable engineering package covering evidence, authorization, commit-boundary controls, failure handling, implementation guidance, and validation scenarios.

    This is a design-time planning and specification Skill. It does not:

    • Execute database mutations or write to customer production systems
    • Authorize, commit, revert, or deploy live operations
    • Act as a runtime guardrail, authorization engine, database firewall, or production write blocker
    • Implement a customer-side runtime Supervisor
    • Automatically detect, clean, or merge duplicate records
    • Replace your domain policies, legal identity criteria, approval authorities, or production controls

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 day ago

    Requires

    Python
    Public MCP
    No MCP
    Node.js
    Derived MCP
    PCHK 01 MCP

    Permissions

    Read Files
    Write Files
    Network Access
    Listed1 day ago

    What's inside

    Frequently Asked Questions

    It covers MERGE, SPLIT, UNMERGE, and other identity-changing operations. The required evidence, authorization, commit-boundary controls, and recovery strategies depend on the operation and its risk.

    No. The Skill works with workflow descriptions, customer-provided context, and supplied files. It does not require direct access to customer production systems.

    No. The Skill supports a customer-context-only mode. The optional public, read-only DIO/ODI Reference MCP provides additional reference profiles, contracts, concepts, and mappings. It must be configured separately in a compatible agent environment.

    No. The Planner produces engineering plans, guardrail specifications, implementation guidance, and validation scenarios. It does not execute, authorize, or commit production database mutations.

    No. It assumes an AI agent or another entity-resolution system has proposed an identity change. The Planner focuses on designing the controls required before that proposed change can reach persistent production state.

    Add a remote MCP server in your MCP-compatible agent client using Streamable HTTP and the following endpoint: https://mcp.danielberanek.cz/mcp Save the connection and make the server available to the agent running the Skill. The exact configuration steps depend on your client. Some clients provide a graphical setup, while others require an MCP configuration file. Once connected, the Planner can use reference-backed mode. If your client does not support this connection method, you can still use the Skill in customer-context-only mode. The MCP server is read-only. It provides reference material only; it does not access customer CRM/MDM data or authorize or execute production mutations.

    New in Agents & Orchestration