Fictional Referral Metadata Review

    1

    Practice reconciling fictional referral packet metadata, immutable revisions, checklist categories, receipts and acknowledgments. No actual documents, patient data or clinical judgment.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Fictional Referral Metadata Review

    Example session with this skill installed

    Complete the task in the included fictional example, using only its supplied evidence and taking no external actions.

    • Read your context and instructions
    • Compiled the fictional referral metadata

    Fictional referral metadata brief

    DRAFT / human review required. Guidance-only; semantic support UNVERIFIED. Local host-agent exercise, not original CLI or live Agensi execution.

    Scope, policy and gate

    Entirely invented REF-DEMO-101 revision 1; submission sender:EV-101; 2026-10-05T08:00:00Z–12:00:00Z. No documents or patient identities were supplied or inspected.
    REFERRAL-DEMO-POLICY v1 requires administrative_form and routing_sheet; delivery and acknowledgment each 3600 seconds; max lag 300 seconds. Approval approval:EV-001 dated 2026-10-04T12:00:00Z says: “For this fictional exercise I approve REFERRAL-DEMO-POLICY v1 as stated above.” Content binding/authenticity UNVERIFIED. Human review must compare exact content/version; this is provisional exercise approval, not authenticated authorization.
    Sender and receiver manifests cover exactly this case, available, untruncated, no offline sync, complete through 12:00Z, uncertainty zero. Sender count 1; receiver count 2; counts match manually. Ingestion is explicitly occurrence plus one minute, within allowed lag. Cutoff is 12:00Z; conservative submission age = 12:00−08:10−2×0 = 3h50m = 13,800 seconds. Unknown or blocked cases: none found within these provisional fictional facts.

    Category ledger

    • administrative_form: sender:EV-101 at 08:10Z lists version 1, fingerprint A (64 a characters). receiver:EV-102 at 08:30Z explicitly links the submission and records the same category/version/fingerprint. Supplied metadata matches; actual content is not verified.
    • routing_sheet: sender:EV-101 lists version 1, fingerprint B (64 b characters). No matching receiver document_received event is observed in the available records. No received unexpected category or conflicting metadata is supplied.
      Both required categories are listed in submission metadata, so there is no required-category-not-listed observation. The receipt comparison includes every listed category.

    Acknowledgment and precise observation

    receiver:EV-103 at 08:45Z explicitly acknowledges sender:EV-101, same case/revision, with no documents. Acknowledgment is recorded after 35 minutes (2100 seconds), within the 3600-second rule. There is no unobserved-acknowledgment finding.
    The acknowledgment does NOT settle routing_sheet receipt. At the 12:00Z cutoff, submission age 13,800 seconds strictly exceeds delivery threshold 3600 seconds: receipt of the listed routing_sheet version 1/fingerprint B is not observed in available records. This is an administrative receipt-evidence observation, not proof that delivery did not happen. No cancellation is supplied and no pending inference is suppressed by cancellation.

    Alternatives, human questions and limits

    Possible explanations include incomplete delivery logging, mismapped category/version metadata, delayed ingestion, or off-system handling. These remain alternatives despite source-asserted completeness. A human coordinator could check only fictional receipt links and manifest completeness, not open actual documents; any new revision must be separately scoped. Matching metadata and acknowledgment cannot establish document validity, consent validity, clinical completeness, correct destination/provider, patient safety or compliance. No urgency, treatment or coverage judgment is made. No external action, notification or automatic RCA occurred.

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

    About this skill

    What you get

    Practice reconciling fictional referral packet metadata, immutable revisions, checklist categories, receipts and acknowledgments. No actual documents, patient data or clinical judgment.

    Includes a tailored evidence/decision guide, report template and bounded fictional practice input. The host agent reviews source-qualified observations and uncertainty, then prepares a provisional result for a human decision.

    Requirements and scope

    Use an Agensi-connected agent that can read delivered text. Wholly fictional training metadata only; no real records. No Python, CLI, installation or external tool is needed for the included example.

    This edition is restricted to entirely invented training metadata. It is not a live administrative, clinical or inventory tool.

    Limitations

    This is an installation-free, guidance-only adaptation of the original public MIT-licensed skill. It does not include the original deterministic runtime, automatic redaction, schema/quote validation, integrity hashing or enforced budgets. It does not authenticate approvals, prove causes, certify compliance or carry out recommendations.

    Source and license

    Source: https://github.com/evgenytkachenko/operational-quality-skills/tree/d330c2956c56c810e6cb8fceb95fe5b456b1b76a/skills/referral-packet-auditor

    Original author: Evgeny Tkachenko (evgenytkachenko). Copyright (c) 2026 evgenytkachenko. MIT license included in the package.

    Verification

    Historical demonstration: a local text-loaded exercise using the included fictional input. This was not a behavioral test using Agensi-delivered instructions or validation of the original CLI. The listing is live, but behavior using Agensi-delivered instructions and independent buyer access have not been tested.

    Guidance-only; semantic support UNVERIFIED; human review required.
    Synthetic-only. Do not use real employer/customer/patient/employee records; no clinical, insurance, urgency or safety decisions.
    No original CLI, deterministic schema/quote/arithmetic checks, automatic redaction, verified hashes or machine-enforced limits.
    No source/approval authentication, causal proof, compliance certification or demonstrated operational effectiveness.
    No system changes, implementation, schedules, notifications, external communications or new permission grants.
    This listing is live. Behavior using Agensi-delivered instructions has not been tested. Historical local text-loaded exercises do not establish reliability across other agents or operational effectiveness.

    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

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 10 days ago

    • Free with an Agensi account

    Requires

    Python

    Permissions

    Read Files

    Read only the delivered package text and entirely invented training metadata. No real records, Python, shell, installation, additional permissions or integrations, system changes or external actions.

    Listed10 days ago

    What's inside

    Frequently Asked Questions

    Practice reconciling fictional referral packet metadata, immutable revisions, checklist categories, receipts and acknowledgments. No actual documents, patient data or clinical judgment. All findings are provisional and require human review.

    Use entirely invented training metadata only. Do not provide real employer, customer, employee or patient records. No clinical, insurance, urgency, safety or live operational decisions. The host only needs to read the delivered text. Use an Agensi-connected text-reading agent. No Python, CLI or installation is required.

    The free package includes SKILL.md, a tailored guide, a report template, a fictional example, README and the original MIT license. Original author: Evgeny Tkachenko (evgenytkachenko). It excludes the original runtime, automatic redaction, deterministic checks and enforced budgets. The displayed example is a historical local text-loaded exercise. Behavior using Agensi-delivered instructions has not been tested. Neither the local exercise nor listing approval proves causal validity or operational effectiveness.

    New in Business & Operations