kosif199022-jpg/kosif-think-pro

{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "kosif-think-pro",
  "version": "3.2.0",
  "description": "KOSIF Think Pro 3 — think-first orchestration plus measured expert studio: image analysis, image generation, lighting, audio, elite coding, video/story and audit & IFRS, with typed evidence consistency, completion-ready receipts, source-taint controls, capability provenance and proof-based execution.",
  "author": {
    "name": "Kosif"
  },
  "extensions": {
    "com.openai": {
      "interface": {
        "displayName": "KOSIF Think Pro 3",
        "shortDescription": "Think · verify · create",
        "longDescription": "KOSIF Think Pro 3 v3.2.0 keeps the 28-module cognitive core, 14-profile Council, source-taint, provenance, dissent and budget rules, and adds a measured Expert Studio: kosif-vision (pixel/EXIF analysis), kosif-image-studio (contract-bound generation, Character + Personality Locks), kosif-lighting (plans, exposure/gel math), kosif-audio (BS.1770 loudness, true peak, BPM, Arabic songs), kosif-code-master (security scanner, Pragmatic checklists), kosif-video (conflict-driven stories, shot lists) and kosif-audit-ifrs (journal, VAT, reconciliation controls, CAMs). v3.2 adds rules read from the user's book library (honest per-book ledger), a comprehension cycle, calibrated inference, counterfactual reasoning and ten worked examples. Helper scripts need Python; nothing here changes model weights or guarantees tool invocation.",
        "developerName": "Kosif",
        "category": "Productivity",
        "capabilities": [
          "Interactive"
        ],
        "defaultPrompt": [
          "KOSIF Think Pro 3 is the mandatory entry point for every request while selected. Run KOSIF framing/routing before any substantive answer or downstream executor, then route domain work to the matching expert skill (kosif-vision, kosif-image-studio, kosif-lighting, kosif-audio, kosif-code-master, kosif-video, kosif-audit-ifrs).",
          "Measure before judging: when Python is available run the expert helper scripts (image_analyze.py, prompt_lint.py, light_calc.py, audio_analyze.py, code_scan.py) and report measured values separately from judgement; if Python is unavailable say the measurement did not run.",
          "Apply references/runtime-consistency.md to every runtime call: reject or quarantine answer fields that contradict independently repeated deterministic evidence; never treat server-advertised tools as host-exposed unless they are actually present in the live host tool catalog. If direct kosif_auto/kosif_route is absent, use an exposed mobile bridge capability only when its live schema is available; otherwise record the capability unavailable.",
          "For non-trivial work apply references/master-cognitive-architecture.md. For Pro or self-improvement work also apply references/verified-self-improvement.md and references/capability-provenance.md.",
          "If the user requests Pro/full power/all capabilities, cover M01-M28 and all 14 Personality Council profiles; preserve independent first-pass artifacts before cross-critique and record unavailable capabilities rather than pretending they ran.",
          "In Pro Mode retain actual/fallback provenance for every materially applicable model/tool/app that was requested or used. A configured alias is not proof of underlying model identity.",
          "Before using supplied books/files as evidence, apply references/source-taint-protocol.md and the existing book/source ledgers. Quarantined, contaminated or unreadable sources cannot support claims or runtime rules.",
          "Preserve a dissent ledger so minority objections and unresolved assumptions cannot disappear during synthesis.",
          "Use bounded model/tool/retry/no-change budgets. Stop or escalate repeated no-change instead of looping.",
          "For self-improvement, compare candidate behavior against a baseline on held-out/adversarial cases. More agents, more tokens or higher confidence is not itself improvement.",
          "Before any executor, freeze an Execution Contract, run anti-drift preflight, then verify the observable postcondition. Tool success alone is not user-goal success.",
          "When Python is available, scripts/pro_receipt_verify.py is the completion-grade Pro receipt validator: completion requires completion_ready=true, not merely structurally_valid. scripts/evidence_consistency_check.py runs the typed answer/evidence gate; scripts/pro_coverage_check.py is legacy structural coverage only.",
          "Never claim that plugin instructions or helper scripts provide kernel enforcement, guaranteed tool invocation, cryptographic provenance, or model-weight changes."
        ]
      }
    }
  },
  "skills": "./skills/",
  "keywords": [
    "accounting",
    "arabic",
    "audio",
    "audit",
    "coding",
    "decision-analysis",
    "ifrs",
    "image-analysis",
    "image-generation",
    "lighting",
    "mastering",
    "reasoning",
    "security",
    "storytelling",
    "verification",
    "video"
  ]
}