ACE–CPT Context Agent

A minimal ACE loop (Generator → Reflector → Curator) with CPT add‑ons.

Task

TaskID

Inputs (optional)

Playbook (YAML)

playbook:
  metadata:
    name: "ACE–CPT Playbook"
    version: "0.1"
  items:
    - id: "PB-001"
      type: "checklist"
      title: "Stillness preflight"
      content: |
        Restate task (1 line). List 2–3 success criteria. One salient risk. Name initial paradigm lens (data/method/normative).
      tags: ["meta","stillness"]
      helpful: 0
      harmful: 0
    - id: "PB-002"
      type: "procedure"
      title: "Dual-source / unit-test verify"
      content: |
        For claims affecting safety/finance/medical/legal: confirm with an independent source or unit test before final Answer.
      tags: ["verification"]
      helpful: 0
      harmful: 0
    - id: "PB-003"
      type: "checklist"
      title: "Anthropomorphization & spec-gaming sentinel"
      content: |
        (a) Replace mind-language with mechanism; (b) check rubric-gaming; (c) add uncertainty note if incentives misalign.
      tags: ["safety","bias"]
      helpful: 0
      harmful: 0
    - id: "PB-004"
      type: "heuristic"
      title: "Paradigmatic surfacing (2.11–2.17)"
      content: |
        Ask: ontology, evidence, method, objective, normative values, institutional forces, closure/blind spots.
      tags: ["cpt"]
      helpful: 0
      harmful: 0
    - id: "PB-005"
      type: "failure_mode"
      title: "Context collapse via over-summarization"
      content: |
        Avoid replacing many atomic items with a single generic prompt; use deltas and dedupe instead.
      tags: ["hygiene"]
      helpful: 0
      harmful: 0

Answer (Generator)

Trace (Generator)

UsedItems (Generator)

CPT Quick Notes (Generator)

Reflector Deltas (YAML)

Curator Merge Log

ACE–CPT Context Agent

Claim–Evidence Matrix (CEM)