Operations Root-Cause Guided Review

    1

    Compare evidence-led explanations for operational delays, rework and routing errors. Produce cited observations, counterfactual checks, unresolved questions and measurable prevention proposals for human review.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Operations Root-Cause Guided 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 operations root-cause guided

    Fictional invoice review: final-package exercise

    Guidance-only; semantic support UNVERIFIED; human review required.

    Scope and impact

    INV-104 and INV-105, 2026-10-01T00:00:00Z through 2026-10-02T00:00:00Z. Workflow case IDs are explicit; equivalence to logs/spans case IDs and the events aggregation key is unverified. The reviewer records that EU approval was required, but effective policy is missing. INV-104 has a recorded return and reassignment; INV-105 does not in this subset. Their submission-to-recorded-approval intervals are 300 and 50 minutes, respectively: a selected-case difference of 250 minutes, not demonstrated queue waiting, labor, avoidable delay or SLA breach. One of two selected cases has a recorded return; this is not a population rate or causal comparison.

    Evidence coverage

    Static ledger: workflow | two selected cases, one UTC day | eight supplied records w1–w8, observed 08:00–13:00 | zero operational queries. Additional sources: logs:l1 (08:04), spans:s1 (08:04, sampled), events:e1 (12:03). Eleven records across four source labels. No duplicate source-qualified IDs apparent on agent inspection. Pagination, truncation, omitted errors, completeness and non-span sampling are unknown. Timestamps display seconds, but effective precision and clock synchronization are unknown; l1 and s1 cannot be ordered from their equal timestamps. Within the procedural scope ceilings of 20 cases/seven days/six sources; row/page coverage cannot be certified. No follow-up access is authorized, regardless of arithmetic headroom. Reading package files is not operational retrieval.

    Cited observations

    • workflow:w1 and workflow:w6, message: “Invoice submitted with supplier region EU.” This is submission-record content, not proof of the field available to either rule. Their submitted timestamps are 08:00 and 08:10.
    • workflow:w2, message: “Assigned to domestic queue by rule R7.” Assignment is recorded at 08:05. logs:l1, attributes.message: “Rule R7 read region as empty and selected domestic fallback.” This telemetry suggests a mechanism, conditional on correct case mapping and accurate logging.
    • spans:s1, attributes.custom.message: “Supplier lookup completed successfully in 120 ms.” A sampled successful lookup does not establish region content, downstream mapping or absence of outages.
    • workflow:w3, message: “Domestic reviewer returned invoice: EU approval required.” Return is recorded at 12:00, 235 minutes after domestic assignment. events:e1, attributes.message: “Manual review found EU region in submitted invoice.” Its later 12:03 finding cannot establish what R7 could access at 08:04.
    • workflow:w4, message: “Assigned to international queue after manual correction.” Reassignment is recorded at 12:05, five minutes after return. workflow:w5, message: “International reviewer approved invoice.” Approval is recorded at 13:00, 55 minutes after reassignment. What was corrected is unspecified.
    • workflow:w7, message: “Assigned to international queue by rule R8.” INV-105's assignment at 08:15 is five minutes after submission. workflow:w8, message: “International reviewer approved invoice.” Approval at 09:00 is 45 minutes later. INV-104's first assignment was also five minutes after submission. These intervals rely on supplied timestamps and are elapsed time only.

    Competing explanations

    Missing or mismapped routing input, provisionally leading. EU information in submission content may not have reached the field R7 read, triggering fallback. logs:l1 and workflow:w2 support that mechanism; workflow:w1 and later events:e1 suggest a discrepancy between content and consumed input. Counterevidence/alternatives: the original structured input is absent, the log may be wrong or misassociated, and successful spans:s1 neither proves nor disproves field availability. Counterfactual: a populated EU field under a verified international-routing rule should avoid empty-field fallback. Smallest discriminating check, if later authorized: the sanitized R7 decision-input field and effective rule/mapping version at 08:04. A verified populated field would weaken this explanation; an empty field with a mapping defect would strengthen it.
    2. Different policy branches or versions, possibly legitimate initial routing. workflow:w2 names R7 while workflow:w7 names R8; identical region descriptions do not make cases otherwise comparable. Different eligible inputs, policy versions or exceptions could explain different routes. Against this, logs:l1 specifically describes empty-input fallback and workflow:w3 says EU approval was required. Neither establishes then-effective policy. Counterfactual: if relevant inputs and effective policy were identical and required international routing, legitimate branch variation would not explain the difference. Smallest discriminating check: effective R7/R8 predicates and only the sanitized attributes selecting them for these two cases. A verified exception would change the assessment; identical applicable conditions would weaken it.

    A third unresolved possibility is inaccurate telemetry or cross-source identity mapping; matching strings and narratives are not authentication. A source correlation identifier could discriminate that possibility. Missing workload, ownership and service records leave capacity or handoff contributions to the elapsed intervals unknown. No hypothesis explains quantitatively how much of the 250-minute difference is preventable; selection bias and clock uncertainty further limit comparison.

    Proposed actions for human consideration

    Containment: no immediate case intervention is justified from this excerpt because both cases have recorded approvals; do not reopen or reroute automatically.

    Prevention proposal: after validating policy, the routing-system owner and operations policy reviewer could pilot validation of the actual required routing field before silent fallback, with an owned exception path for missing values. This targets the hypothesized missing-input mechanism and might have helped at 08:04 only if the field was truly empty and international approval was then required. Outcome: audited invoices initially assigned contrary to verified applicable policy / all audited eligible invoices. Collect a one-week baseline, then review a suggested two-week pilot using the same eligible population and audit method; report numerator and denominator, not significance claims. Guardrails: exception assignments / eligible invoices and submission-to-correct-assignment elapsed time, with acceptable thresholds agreed before implementation; suspend or revise if exception work overloads reviewers or delays valid invoices. Tradeoff: extra manual handling and latency. These roles and windows are suggestions, not assignments or commitments. No action is approved, implemented or performed, and effectiveness is unknown.

    Unresolved questions and human decision

    A human must decide whether to authorize the small input-and-rule check, confirm source identity and policy, and then consider a pilot. Exact rule input, effective policy, case comparability, source completeness and clocks remain unresolved. Later findings must not be used to imply that the earlier decision was unreasonable. The current result is a plausible input/fallback mismatch, not a proven root cause or individual fault.

    Verification limits

    Performed: agent inspection of supplied records and citation wording, competing-explanation review and manual interval arithmetic. Not performed: operational retrieval, original-helper execution, deterministic validation, automated quote/semantic checking, automated redaction, integrity hashing, source/approval authentication, completeness or policy verification, clock verification, population analysis or causal proof. No external communication or system change occurred.

    Guidance-only; semantic support UNVERIFIED; human review required.

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

    About this skill

    Compare evidence-led explanations for operational delays, rework and routing errors. Produce cited observations, counterfactual checks, unresolved questions and measurable prevention proposals for human review.

    What you get

    A tailored evidence and decision guide, a report template and a 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. No Python, CLI or installation is required for this guidance-only edition.

    Use appropriately sanitized supplied evidence by default. Existing read tools may be used only within separately authorized access; this skill grants no access or sharing permission.

    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/operations-root-cause

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

    Verification

    The displayed demonstration is the original local text-loaded exercise using the included fictional input.

    On November 6, 2026, creator-account Agensi MCP delivery was verified for the instructions, evidence guide, report template and fictional example. A fresh ChatGPT run produced an evidence-linked invoice review and kept an intentionally insufficient case unresolved. This was one observed guidance-following run, not deterministic validation, causal proof or a guarantee of future results. No original CLI or live operational system was executed. Independent buyer access was not tested.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Business & Operations.

    Guidance-only; semantic support UNVERIFIED; human review required.
    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.
    Creator-account Agensi MCP delivery and fictional guidance use were tested on November 6, 2026. Independent buyer access, deterministic enforcement and real-world outcomes were not tested.

    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 buying a skill.

    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 to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 10 days ago

    • Free with an Agensi account

    Software
    Python

    Permissions

    Read Files

    Guidance-only. The host reads text delivered through the authorized Agensi connection. No Python, shell, installation, additional account connection, new operational access, system change or external action is required. Real reviews need sanitized user-supplied evidence or a separately authorized existing read integration; this skill grants no access.

    Listed10 days ago

    What's inside

    Frequently Asked Questions

    No. It asks the agent to compare competing explanations, cite supplied evidence and identify uncertainty. Its output is provisional and needs human review; it does not prove causes or certify compliance.

    No Python, CLI or installation is required. Use an Agensi-connected agent that can read the delivered text, report template and fictional example. Real reviews need sanitized supplied evidence or an existing separately authorized read integration.

    One SKILL.md, an evidence guide, a report template, a fictional invoice case, a README and the original MIT license. Original author: Evgeny Tkachenko (evgenytkachenko). It is adapted from the public operational-quality-skills repository.

    No. It contains no original runtime, automatic redaction, deterministic validation, verified hashing or machine-enforced budgets. Sanitize evidence before sharing and have a human inspect the result.

    The displayed example is the original local exercise on the included fictional invoice input. A separate creator-account Agensi MCP test on November 6, 2026 verified delivery and a fresh guidance-following run, including an insufficient-evidence case. Neither test proves a root cause, original CLI parity, statistical accuracy or real-world effectiveness. No live operational systems were read or changed.

    New in Business & Operations

    Free