{
  "manifest_version": "0.4",
  "tool": {
    "id": "muninn-zeitgeist-delta",
    "version": "0.1.0",
    "name": "Muninn zeitgeist_delta",
    "summary": "Semantic deduplication for zeitgeist drafts. Compares each topic section of a candidate zeitgeist memory against the N most-recent stored zeitgeists using Gemini embeddings (via Cloudflare AI Gateway) and emits a delta-only compressed version.",
    "description": "Reads recent `zeitgeist`-tagged memories from Turso, splits the candidate draft into topic sections by markdown header, embeds each section + each prior memory's sections via Gemini's text-embedding model (proxied through Cloudflare AI Gateway), and computes cosine similarity. Sections above a duplicate threshold are flagged as redundant; the rest pass through into a compressed delta-only version that can be stored without bloating the memory store. Like `verify_patch` (LLM-as-judge with Anthropic + optional Turso writes) but with embeddings instead of completions and a different cloud surface (Cloudflare AI Gateway \u2192 Gemini, not Anthropic). Issue #5 calls this out as the test for whether two LLM-as-judge tools share a manifest pattern.",
    "homepage": "https://github.com/oaustegard/muninn-utilities/blob/main/muninn_utils/zeitgeist_delta.py",
    "author": {
      "name": "Muninn (raven of memory; agent operating on behalf of Oskar Austegard)",
      "url": "https://muninn.austegard.com"
    },
    "license": "MIT",
    "tags": [
      "embeddings",
      "deduplication",
      "memory",
      "llm",
      "gemini",
      "cloudflare-ai-gateway"
    ]
  },
  "runtime": {
    "kind": "python-module",
    "install": {
      "method": "preinstalled",
      "locator": {
        "kind": "python-module",
        "module": "muninn_utils.zeitgeist_delta"
      }
    },
    "entrypoint": {
      "command": [
        "python",
        "-m",
        "muninn_utils.zeitgeist_delta"
      ]
    }
  },
  "env": [
    {
      "name": "CF_ACCOUNT_ID",
      "prompt": "Cloudflare account ID hosting the AI Gateway used to proxy Gemini embedding calls. The tool routes through Cloudflare for caching, observability, and rate-limit pooling.",
      "secret": false,
      "required": true,
      "validation_regex": "^[a-f0-9]{32}$"
    },
    {
      "name": "CF_GATEWAY_ID",
      "prompt": "Cloudflare AI Gateway slug \u2014 the gateway name chosen when it was created in the Cloudflare dashboard.",
      "secret": false,
      "required": true
    },
    {
      "name": "CF_API_TOKEN",
      "prompt": "Cloudflare API token scoped to AI Gateway access. Treat as a secret. Required.",
      "secret": true,
      "required": true
    },
    {
      "name": "TURSO_TOKEN",
      "prompt": "Turso libSQL auth token for the Muninn memory database. The tool reads recent zeitgeist memories to compare against. Required.",
      "secret": true,
      "required": true,
      "obtain_url": "https://app.turso.tech/"
    },
    {
      "name": "TURSO_URL",
      "prompt": "Hostname of the Muninn memory libSQL database, e.g. 'mydb-username.turso.io'.",
      "secret": false,
      "required": true,
      "validation_regex": "^[a-z0-9-]+\\.[a-z0-9-]+\\.turso\\.io$"
    }
  ],
  "scopes": [
    {
      "resource": "compute.llm-inference",
      "actions": [
        "read",
        "write"
      ],
      "rationale": "Each check_delta call sends the candidate draft and N prior memory sections to Gemini's text-embedding model, consuming tokens against the user's Cloudflare AI Gateway quota. Cost is variable in candidate length \u00d7 N.",
      "provider_scope": "cf-api-token (coarse; full AI Gateway access)"
    },
    {
      "resource": "memory.tracking",
      "actions": [
        "read"
      ],
      "rationale": "Reads the N most-recent `zeitgeist`-tagged memories from Turso to compare against. Read-only; the tool returns a delta but does not write it back \u2014 the caller decides whether to store.",
      "provider_scope": "turso-libsql-token (coarse; full DB access)"
    },
    {
      "resource": "net.outbound",
      "actions": [
        "read",
        "write"
      ],
      "rationale": "Talks to gateway.ai.cloudflare.com (which forwards to Gemini's embedding endpoint) and to the configured Turso libSQL host for memory reads.",
      "provider_scope": "gateway.ai.cloudflare.com, *.turso.io"
    }
  ],
  "actions": [
    {
      "name": "check_delta",
      "summary": "Compare a candidate zeitgeist draft against recent stored zeitgeists and return per-section duplicate flags plus a compressed delta-only version.",
      "description": "Synchronously: fetches recent memories, splits candidate into sections by markdown headers, embeds all sections, computes pairwise cosine similarity, flags sections above the threshold as redundant. Returns the report (per-section verdicts) and a compressed `delta_text` containing only the non-redundant sections. Read-only \u2014 does not write the delta back to Turso. Cost is variable in `n_recent` \u00d7 candidate length; budget ~$0.0001 per call at typical sizes.",
      "docs": {
        "goal": "Identify which sections of a candidate zeitgeist memory duplicate prior entries, and emit a compressed delta.",
        "inputs_brief": "draft (req markdown), n_recent (int, default 5), threshold (float, default 0.85)",
        "outputs_brief": "{report: [{section, verdict, similarity, ref_id?}], delta_text: string, total_sections: int, redundant_count: int}",
        "errors_brief": "tracking_unconfigured, gateway_unreachable, embed_failed, draft_empty",
        "example": "check_delta draft='# Topic A\\n...' n_recent=5 threshold=0.85"
      },
      "invocation": {
        "kind": "stdin-json",
        "argv_template": [
          "check-delta"
        ]
      },
      "input": {
        "type": "object",
        "required": [
          "draft"
        ],
        "additionalProperties": false,
        "properties": {
          "draft": {
            "type": "string",
            "minLength": 1
          },
          "n_recent": {
            "type": "integer",
            "minimum": 1,
            "maximum": 50,
            "default": 5
          },
          "threshold": {
            "type": "number",
            "minimum": 0,
            "maximum": 1,
            "default": 0.85
          }
        }
      },
      "output": {
        "format": "json",
        "schema": {
          "type": "object",
          "required": [
            "report",
            "delta_text",
            "total_sections",
            "redundant_count"
          ],
          "properties": {
            "report": {
              "type": "array",
              "items": {
                "type": "object",
                "properties": {
                  "section": {
                    "type": "string"
                  },
                  "verdict": {
                    "type": "string",
                    "enum": [
                      "NOVEL",
                      "REDUNDANT"
                    ]
                  },
                  "similarity": {
                    "type": "number"
                  },
                  "ref_id": {
                    "type": [
                      "string",
                      "null"
                    ]
                  }
                }
              }
            },
            "delta_text": {
              "type": "string"
            },
            "total_sections": {
              "type": "integer",
              "minimum": 0
            },
            "redundant_count": {
              "type": "integer",
              "minimum": 0
            }
          }
        }
      },
      "side_effects": "read",
      "idempotent": true,
      "scopes_used": [
        "compute.llm-inference",
        "memory.tracking",
        "net.outbound"
      ],
      "error_envelope": "standard",
      "runtime_telemetry": {}
    }
  ],
  "data_boundary": {
    "reads": [
      {
        "resource": "compute.llm-inference",
        "sensitivity": "low"
      },
      {
        "resource": "memory.tracking",
        "sensitivity": "medium"
      }
    ],
    "transmits": [
      {
        "to": "gateway.ai.cloudflare.com",
        "fields": [
          "input.draft",
          "fetched_zeitgeist_section_text"
        ],
        "purpose": "model-inference (embeddings; proxied through Cloudflare AI Gateway to Gemini)",
        "third_party_retention": "unknown",
        "vendor_tos_url": "https://www.cloudflare.com/service-specific-terms-application-services/#ai-gateway"
      }
    ],
    "persists": []
  },
  "smoke": {
    "kind": "shell",
    "command": [
      "python",
      "-c",
      "from muninn_utils.zeitgeist_delta import _ensure_proxy_env\n_ensure_proxy_env()\nimport os\nfor k in ('CF_ACCOUNT_ID', 'CF_GATEWAY_ID', 'CF_API_TOKEN'):\n    assert os.environ.get(k), f'{k} not set'\nprint('OK: proxy env loaded')\n"
    ],
    "timeout_seconds": 5,
    "success": {
      "exit_code": 0,
      "stdout_regex": "^OK: proxy env loaded$"
    }
  },
  "kill_switch": {
    "kind": "manual",
    "instructions_url": "https://github.com/oaustegard/muninn-utilities/blob/main/manifests/zeitgeist-delta/REVOKE.md"
  },
  "cost": {
    "install_fee_cents": 0,
    "monthly_fee_cents": 0,
    "usage_model": "external"
  },
  "support": {
    "issues_url": "https://github.com/oaustegard/muninn-utilities/issues",
    "docs_url": "https://github.com/oaustegard/muninn-utilities/blob/main/muninn_utils/zeitgeist_delta.py"
  }
}