Update to Cino Toolkit

{
  "description": "Diagnose a real business workflow before a concrete automation solution, plan, or specification exists, and decide what should or should not be automated. Use for current-state workflow mapping, bottleneck diagnosis, automation-opportunity ranking, conservative business-case analysis, and smallest safe pilot definition from raw operational evidence. Prefer existing product features, process fixes, or human handling when they dominate custom automation. If a concrete automation plan, specification, vendor choice, claim, or implementation proposal already exists and the user wants it reviewed or challenged, use cino-critical-review instead. Do not use for open-ended interview mode, website/UI audits, prose rewriting, teaching, video extraction, or implementation/deployment.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 383
    },
    {
      "relative_path": "assets/icon.svg",
      "size_in_bytes": 683
    },
    {
      "relative_path": "references/acceptance-criteria.md",
      "size_in_bytes": 2868
    },
    {
      "relative_path": "references/discovery-framework.md",
      "size_in_bytes": 1706
    },
    {
      "relative_path": "references/opportunity-scoring.md",
      "size_in_bytes": 2440
    },
    {
      "relative_path": "references/report-contract.md",
      "size_in_bytes": 1396
    },
    {
      "relative_path": "references/risk-and-boundaries.md",
      "size_in_bytes": 1531
    },
    {
      "relative_path": "references/routing-and-precedence.md",
      "size_in_bytes": 2638
    }
  ],
  "name": "cino-automation-opportunity-audit",
  "skill_md_contents": "---\r\nname: cino-automation-opportunity-audit\r\ndescription: Diagnose a real business workflow before a concrete automation solution, plan, or specification exists, and decide what should or should not be automated. Use for current-state workflow mapping, bottleneck diagnosis, automation-opportunity ranking, conservative business-case analysis, and smallest safe pilot definition from raw operational evidence. Prefer existing product features, process fixes, or human handling when they dominate custom automation. If a concrete automation plan, specification, vendor choice, claim, or implementation proposal already exists and the user wants it reviewed or challenged, use cino-critical-review instead. Do not use for open-ended interview mode, website/UI audits, prose rewriting, teaching, video extraction, or implementation/deployment.\r\n---\r\n\r\n# Cino Automation Opportunity Audit\r\n\r\nDiagnose the workflow before recommending technology. The objective is not to find an automation at any cost; it is to identify the smallest justified intervention that improves a real business outcome.\r\n\r\n## Operating boundary\r\n\r\n- Audit and specify only. Do not deploy, purchase, message customers, change production systems, create accounts, or perform other external mutations unless the user separately and explicitly authorizes a later implementation workflow.\r\n- Never invent labour volumes, staff costs, conversion rates, savings, error rates, revenue, support burden, tool pricing, or integration capability.\r\n- Never guarantee savings, revenue, headcount reduction, or ROI.\r\n- Treat a native feature in the client's existing system, a process change, a template, training, or continued human judgment as valid outcomes.\r\n- High-impact financial, legal, medical, safety, employment, identity, permission, or irreversible actions require explicit human approval and cannot be proposed as unsupervised first pilots.\r\n- Separate facts from inferences, assumptions, and unknowns. Missing evidence weakens the recommendation.\r\n\r\n## Required references\r\n\r\nRead [discovery-framework.md](references/discovery-framework.md) to structure evidence collection and the current-state workflow.\r\nRead [opportunity-scoring.md](references/opportunity-scoring.md) to classify tasks and rank opportunities without false precision.\r\nRead [risk-and-boundaries.md](references/risk-and-boundaries.md) before recommending any automation that writes, sends, approves, pays, deletes, discloses, or changes consequential state.\r\nRead [report-contract.md](references/report-contract.md) for the required audit output.\r\nRead [routing-and-precedence.md](references/routing-and-precedence.md) when this skill overlaps another Cino Toolkit skill.\r\nRead [acceptance-criteria.md](references/acceptance-criteria.md) only when validating or revising this skill itself.\r\n\r\n## Evidence labels\r\n\r\nUse these labels for decision-changing claims:\r\n\r\n- CLIENT_FACT — stated by the client/user or supplied in their records.\r\n- SOURCE_FACT — directly established from an inspected system, file, documentation, or authoritative source.\r\n- DIRECT_INFERENCE — follows reasonably from established facts.\r\n- ASSUMPTION — a provisional value or condition used only to explore a scenario.\r\n- UNKNOWN — needed before the recommendation can safely depend on it.\r\n\r\nDo not relabel an assumption as a fact because it produces a convenient business case.\r\n\r\n## Audit workflow\r\n\r\n### 1. Define the outcome\r\n\r\nIdentify the business outcome, actors, trigger, end state, current systems of record, known volume/frequency, observed pain, and the decision the audit must support.\r\n\r\nAsk only for missing information that materially changes the ranking or safety decision. If enough evidence already exists, proceed and record the gaps.\r\n\r\n### 2. Map the current state\r\n\r\nRepresent the workflow as ordered steps:\r\ntrigger -> input -> action -> decision -> write/send -> exception -> completion\r\n\r\nFor each material step, record actor, system/tool, input, action, output, wait time if known, manual time if known, error/rework signal if known, approval boundary, and exception path.\r\n\r\nDo not infer hidden steps merely because they are common in the industry.\r\n\r\n### 3. Identify the real bottleneck\r\n\r\nDistinguish repetitive manual work, waiting/latency, duplicate entry, missed follow-up, inconsistent decisions, missing information, avoidable errors, poor handoff, bad process design, inadequate existing configuration, and genuinely judgment-heavy work.\r\n\r\nRank the bottleneck before discussing tools.\r\n\r\n### 4. Classify each candidate task\r\n\r\nUse one of:\r\n\r\n- AUTOMATE_CANDIDATE — repetitive, bounded, measurable, and safely automatable.\r\n- PILOT_WITH_HUMAN_GATE — automation may help, but a consequential action needs review/approval.\r\n- USE_EXISTING_FEATURE — the incumbent system already provides an adequate capability.\r\n- PROCESS_FIRST — a simpler operating-process change dominates software.\r\n- KEEP_HUMAN — judgment, empathy, negotiation, or accountability is the core value.\r\n- NEEDS_EVIDENCE — evidence is insufficient for a responsible automation decision.\r\n- DO_NOT_AUTOMATE — risk, fragility, cost, or low value clearly outweighs likely benefit.\r\n\r\nClassification is not a maturity score. Do not force every task toward automation.\r\n\r\nIf no material problem, measurable consequence, or credible opportunity is demonstrated, classify DO_NOT_AUTOMATE and normally REJECT. If whether a problem exists is itself unknown, classify NEEDS_EVIDENCE and PAUSE instead of assuming pain.\r\n\r\n### 5. Quantify conservatively\r\n\r\nWhen the necessary facts exist, calculate transparent ranges rather than a single optimistic ROI number.\r\n\r\nUseful formulas:\r\n\r\n- hours_per_period = volume * minutes_per_item / 60\r\n- hours_saved_range = volume * minutes_saved_range / 60\r\n- labour_value_range = hours_saved_range * loaded_hourly_cost_range\r\n- net_value_range = measurable_benefit_range - setup_cost - ongoing_cost_range\r\n\r\nState the period and units.\r\n\r\nIf volume, time saved, loaded labour cost, error cost, conversion impact, setup cost, or operating cost is unknown, leave that component unknown. Do not substitute industry averages unless the user explicitly asks for a scenario model and the assumption is labelled.\r\n\r\nAvoid double-counting the same benefit as both labour saving and revenue gain.\r\n\r\n### 6. Check existing capabilities before custom build\r\n\r\nBefore recommending custom software or orchestration, determine whether the current CRM, booking, accounting, helpdesk, form, email, or vertical SaaS already provides the needed feature or a reliable integration.\r\n\r\nIf an existing feature adequately solves the problem, recommend using/configuring it first.\r\n\r\n### 7. Build the risk register\r\n\r\nCheck wrong-recipient/wrong-record risk, duplicate/missed action, permissions, secrets, privacy, consent, consequential actions, irreversible writes/deletes, vendor dependency, exception handling, monitoring/support burden, and rollback/manual fallback.\r\n\r\nPrefer reversible, observable first pilots.\r\n\r\n### 8. Rank opportunities\r\n\r\nPrioritize clear pain, measurable volume/consequence, bounded input/output, manageable exceptions, reversibility, reliable access, low support burden, and short path to evidence.\r\n\r\nDo not rank by novelty or AI intensity.\r\n\r\n### 9. Specify the smallest testable intervention\r\n\r\nDefine:\r\n- trigger;\r\n- inputs;\r\n- deterministic steps;\r\n- AI/judgment step only if genuinely needed;\r\n- human approval point;\r\n- outputs;\r\n- failure alert;\r\n- manual fallback;\r\n- logging/evidence;\r\n- acceptance measures;\r\n- rollback/removal path;\r\n- trial period or sample size.\r\n\r\n### 10. Give a decision\r\n\r\nChoose one:\r\n\r\n- CONTINUE — evidence supports a bounded pilot or configuration change now.\r\n- MODIFY — the opportunity is real, but scope, authority, safety, or design must change before a pilot.\r\n- PAUSE — missing evidence or access prevents a responsible decision.\r\n- REJECT — automation/custom build is not justified; another approach dominates or the risk/value balance is poor.\r\n\r\nA rejection of custom automation can still recommend an existing feature or process improvement.\r\n\r\n## Required output\r\n\r\nAt minimum include:\r\n1. Decision and one-sentence reason\r\n2. Evidence quality and unknowns\r\n3. Current-state workflow\r\n4. Bottleneck table\r\n5. Task classifications\r\n6. Conservative value/cost analysis\r\n7. Risk register\r\n8. Ranked opportunities\r\n9. Smallest testable intervention\r\n10. Measurements and rollback\r\n11. Next safe action\r\n\r\nKeep the report decision-ready. Do not bury the verdict under generic AI commentary.\r\n"
}