@ifsoul/core 0.3.0 → 0.5.0

package/specs/v0.2/README.md

@@ -1,421 +0,0 @@
-# Character and host contracts: draft-0.2.0
-
-This revision describes portable characters in games and chat, including hosts
-that already control NPC behavior. It is **documentation, JSON Schema, and
-synthetic examples only**. There is no character controller, model client, game
-bridge, locomotion executor, storage engine, archive importer, or migration.
-The words MUST, MUST NOT, SHOULD, and MAY describe future conformance.
-
-The historical [draft-0.1.0](../README.md) schemas and example bytes remain unchanged. This revision is an
-explicit new contract, not an automatic upgrade, replacement of every independent
-contract, or promise that old documents can be imported. In particular,
-[ModelProvider](../model-provider/README.md) retains `schemaVersion: "draft-0.1.0"`
-and its existing `urn:ifsoul:protocol:draft-0.1.0:model-provider` ID. A v0.2
-controller may use that independent provider boundary after explicit composition;
-a v0.2 character or plan is not a v0.1 model request. No v0.2 provider schema or
-implicit version coercion is introduced. Historical v0.1 permission-policy
-messages are also not embedded into these new envelopes; current trusted
-policy and host restrictions still require explicit evaluation.
-
-## Registry and shared conventions
-
-All schema files are in
-[`packages/protocol/schemas/v0.2/`](../../packages/protocol/schemas/v0.2/).
-The `$id` prefix is `urn:ifsoul:protocol:draft-0.2.0:` plus the listed suffix.
-Register all eight local files by `$id`, then compile with Ajv 2020,
-`strict: true`, and `ajv-formats` for `date-time`. URNs are identifiers, not
-hosted endpoints; no network resolution, type coercion, default insertion,
-unknown-field removal, or union-type relaxation is needed.
-
-| File / ID suffix                                                                                           | Root `kind` values                                                                                             |
-| ---------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------- |
-| [common.schema.json](../../packages/protocol/schemas/v0.2/common.schema.json) / `common`                   | Definitions only; deliberately rejects exchange documents                                                      |
-| [character.schema.json](../../packages/protocol/schemas/v0.2/character.schema.json) / `character`          | `character-template`, `character-instance`                                                                     |
-| [environment.schema.json](../../packages/protocol/schemas/v0.2/environment.schema.json) / `environment`    | `environment-profile`                                                                                          |
-| [context.schema.json](../../packages/protocol/schemas/v0.2/context.schema.json) / `context`                | `context-report`                                                                                               |
-| [planning.schema.json](../../packages/protocol/schemas/v0.2/planning.schema.json) / `planning`             | `plan-proposal`, `plan-decision`, `task-report`, `plan-cancel-request`, `plan-cancel-result`                   |
-| [conversation.schema.json](../../packages/protocol/schemas/v0.2/conversation.schema.json) / `conversation` | `conversation-request`, `conversation-response`, `conversation-delivery`                                       |
-| [memory.schema.json](../../packages/protocol/schemas/v0.2/memory.schema.json) / `memory`                   | `memory-event`, `working-context`, `memory-episode`, `memory-fact`, `memory-relationship`, `intent-checkpoint` |
-| [archive.schema.json](../../packages/protocol/schemas/v0.2/archive.schema.json) / `archive`                | `archive-manifest`                                                                                             |
-
-Every envelope requires `schemaVersion: "draft-0.2.0"` and `kind`. Objects are
-closed unless explicitly designated as a trait/argument map or extension object.
-Optional fields are omitted; `null` is not a substitute. JSON is UTF-8 and field
-names are case-sensitive camelCase. IDs match
-`[A-Za-z0-9][A-Za-z0-9._:-]{0,127}`: opaque, case-sensitive names rather than
-credentials, paths, proof of identity, or executable instructions. Counters use
-nonnegative integers no larger than `9007199254740991`.
-
-Bound host envelopes require `binding: { environmentId, sessionId, characterId,
-embodimentId }`. The trusted composition root verifies that binding against the
-current host. Model identity is absent from it. UTC timestamps require RFC 3339
-`Z` form with seconds, a four-digit year, and optional 1–9 digit fractional
-seconds. `worldTime`, where present, is `{ timelineId, tick }`, with no assumed
-UTC conversion or fixed tick rate. A reset or incomparable timeline needs a new
-ID. Chat does not require a world clock.
-
-`period` and `horizon` are tagged unions:
-
-- `{ basis: "utc", startAt, endAt }`, using UTC timestamps.
-- `{ basis: "world-ticks", timelineId, fromTick, toTick }`.
-
-Consumers MUST check ordering, available clocks, and agreement with the current
-environment. A host's daily cadence may mean an arbitrary configured game day;
-it is not hardcoded to 24 wall-clock hours or a built-in scheduler.
-
-Optional `extensions` uses lowercase dotted namespaces such as
-`org.example.notes`, preserving arbitrary JSON values. Every optional
-`requiredExtensions` key MUST exist and be understood before processing.
-Unknown optional extensions MUST survive a claimed lossless transfer; inability
-to preserve them requires rejection or a disclosed filtered operation. Unknown
-required extensions reject processing. None can grant authority. Schemas cannot
-verify implementation support or dynamic key membership.
-
-Numeric array/string bounds on profiles, reports, plans, and utterances limit
-one supported exchange or snapshot; they are not lifetime memory quotas. Exceeding
-a bound MUST produce an explicit rejection or a separately agreed filtered
-operation, never silent truncation. Instance memory references are selected
-snapshot/index roots, not an exhaustive enumeration of lifetime history. Archive
-memory streams may be sharded across files; the schema imposes no fixed maximum
-file count or lifetime event count. Implementations still apply explicit resource
-budgets and fail transparently when exceeded.
-
-## Character, model, and embodiment
-
-A **character** is portable authored identity plus permitted lived experience.
-A **model** is a replaceable inference facility, not the character. An
-**embodiment** is the host's present representation and available modalities, not
-a portable inventory, health record, or behavior engine.
-
-| `kind`               | Required fields beyond the envelope fields                                                         |
-| -------------------- | -------------------------------------------------------------------------------------------------- |
-| `character-template` | `templateId`, `revision`, `profile`, `authorship`                                                  |
-| `character-instance` | `characterId`, `participantId`, `createdAt`, `profile`, `memoryRecordIds`, `relationshipMemoryIds` |
-
-One shared `$defs/profile` supplies both documents. It requires `name`,
-`description`, `traits`, `preferences`, `speechStyle`, `goals`, and
-`authoredBiography`. Empty collections and a style description that language is
-unavailable are valid. `speechStyle` contains description, language labels, and
-habits: preferences for expression **when supported**, never an ability or a
-requirement to speak. The same character may be a human, horse, computer, or
-another form; the active embodiment restricts actual output.
-
-Preferences identify a subject, stance (`like`, `dislike`, `avoid`, `seek`),
-description, and optional strength. Goals have IDs, descriptions, priority, and
-status; they are intentions, not host tasks or achieved facts. Authored biography
-is explicitly marked `fictional: true`; it describes this baseline's original
-fictional characters and MUST NOT be presented as lived memory. Templates have
-positive integer revisions, authorship/license claims, and no individual memory.
-Instances may reference an exact template revision and record import/fork lineage;
-template changes do not silently overwrite an instance. Memory and relationship
-references must resolve consistently, subject to declared export filtering.
-
-The [template](examples/character-template.json) defines reusable authored data;
-the [instance](examples/character-instance.json) includes selected references
-into the recorded synthetic trace. Model settings, authentication, host state,
-and host execution bindings do not belong in the portable character document.
-
-## Environment and stable participants
-
-An Environment Profile requires `binding`, `hostId`, `revision`, `medium`,
-`embodiment`, `participantLinks`, `planning`, `communication`, and `hostIntents`.
-`medium` is `game` or `chat`. `embodiment.form` is an open opaque ID; the examples
-`human`, `horse`, and `computer` are not a closed species catalog. Modalities are
-an explicit subset of `speech`, `text`, `vocalization`, `gesture`, and `status`,
-and may be empty. No language-capable form is compulsory.
-
-`participantId` is a stable participant reference across sessions. Each
-`participantLinks` entry is an issuer's **claim** connecting that reference to an
-`environmentSubjectId` in this bound session. An entry has a `claimId` and
-`issuerId`; no self-asserted verification flag exists. Trusted identity proof,
-revocation checks, access checks, and the host binding are required before using
-the claim. A display name, repeated ID string, imported relationship, or prior
-session claim proves no identity or consent in a new session.
-
-Planning and communication are independent:
-
-- `planning` declares `cadence: host-requested | daily | event-driven` and a
-  `maxTasks` bound. The host chooses when to issue a planning request; the schema
-  does not schedule it or authorize unsolicited execution.
-- `communication` declares `mode: host-requested | live-channel`,
-  `availability: on-demand | persistent`, `allowUnrequestedMessages`, and known
-  `channelIds`. Persistent availability does not authorize unrequested speech.
-  Host-requested mode structurally requires `allowUnrequestedMessages: false`.
-  Live mode may also forbid unrequested messages while allowing replies.
-
-[Game](examples/environment-game.json), [live chat](examples/environment-chat.json),
-and [non-speaking horse](examples/environment-no-speech.json) profiles illustrate
-independent host cadence, online availability, and output capabilities.
-
-## Host briefing and bounded planning
-
-The host owns NPC state machines, locomotion, steering, pathfinding, combat,
-shooting, interactions, inventory, health, scheduling, and authoritative outcomes.
-The bridge conveys context, a bounded plan proposal, dialogue, and status to
-existing host facilities. It MUST NOT grow an IfSoul locomotion skill executor,
-remote command shell, or replacement game behavior engine. Whether the bridge
-is built by IfSoul or a third party does not change that boundary.
-
-A Context Report requires `reportId`, `contextRevision`, `environmentRevision`,
-`correlationId`, `binding`, `recordedAt`, `period`, `observedState`, `priorNeeds`,
-`priorPlans`, and `changes`. The observed state contains a summary and typed
-string/number/boolean property observations, not authority to write those fields.
-Needs include priority and current/resolved/unknown status. Prior plans carry
-task states; changes/events carry recording times and optional world times.
-Missing observations remain unknown. A bridge receives only scoped selected
-context, not the full archive.
-
-An optional `planningRequest` adds `requestId`, `cadence`, `horizon`, `maxTasks`,
-and `allowedIntentIds`. Every allowed ID must be in the current `hostIntents`
-catalog. The host may expose health or inventory as observed context when
-permitted; these observations MUST NOT become portable state to transplant into
-a new embodiment. Chat context can use UTC periods without `worldTime` and
-without requesting any plan.
-
-Each advertised host intent has `intentId`, description, typed parameter
-descriptors, and a cancellation-support flag. Parameter names are unique; types
-are `entity-ref`, `choice`, `quantity`, and `boolean`. Choice lists, quantity units,
-and numeric bounds must fit that parameter's type. They are a small safe data
-vocabulary, not embedded JSON Schema or executable source.
-
-A Plan Proposal requires the bound `planId`/`correlationId`, planning `requestId`,
-`contextReportId`, `contextRevision`, `environmentRevision`, `createdAt`,
-`expiresAt`, `horizon`, and at most 16 `tasks`. The accepted bound is the minimum
-of protocol, profile, and request bounds. No tasks is a valid no-action proposal.
-Each task has a unique `taskId`, advertised `intentId`, typed `arguments`, priority,
-optional explanation, and explicit `dependsOnTaskIds`.
-
-Argument values are tagged objects: `{ type: "entity-ref", entityId }`,
-`{ type: "choice", value }`, `{ type: "quantity", value, unit }`, or
-`{ type: "boolean", value }`. Consumers MUST validate exact parameter names,
-requiredness, type, current entity visibility, allowed choices, units, and bounds
-against the current advertised descriptor. IDs MUST be advertised and allowed by
-this request. Dependencies must resolve in the proposal and form an acyclic
-graph. Unknown intent IDs, extra arguments, stale revisions, expired proposals,
-and unauthorized operations reject processing. An explanation is untrusted
-non-executable commentary. There is no arbitrary execute-code, command, script,
-filesystem path, raw skill, or model-generated locomotion field.
-
-A Plan Decision records host `decisionId`, `requestId`, context revision, time,
-`status: accepted | partially-accepted | rejected`, and per-task decisions.
-The task decisions must uniquely cover all proposed tasks and agree with the
-aggregate status; dependencies of accepted tasks must remain satisfiable. This
-is an acceptance decision, **not achievement**. It does not bypass current host
-restrictions or promise that execution has begun.
-
-Task Reports identify a plan/task, report ID, context revision, correlation, time,
-status, summary, and actual `observedEffects`. Status is `pending`, `running`,
-`succeeded`, `failed`, `cancelled`, or `unknown`. Only the host reports execution
-facts; a plan's predicted effects are not observations. A host can execute
-accepted work independently until it reports back or requests another plan.
-Reports and context revisions can arrive out of order; trusted consumers
-reconcile them rather than overwrite later facts with stale messages.
-
-Cancellation has separate `plan-cancel-request` and `plan-cancel-result` kinds.
-A request names `cancellationId`, plan and task IDs, time, and reason; it does not
-stop the host. The result reports per-task `cancelled`, `already-finished`,
-`rejected`, or `unknown`. A confirmed cancellation prevents further work only as
-reported by the host; it does not roll back observed effects. A cancellation
-request is never a cancellation confirmation.
-
-All decisions, reports, and cancellation messages MUST correlate binding, plan,
-task, request/cancellation IDs, and trusted sender identity. Duplicate delivery
-is not new work. This draft promises no idempotent execution or exactly-once
-transport. Unknown acceptance, execution, cancellation, or delivery requires
-host reconciliation; do not send a replacement plan/task or blindly retry under
-new IDs. If the host cannot reconcile, preserve unknown and request an explicit
-recovery decision. No transport or reconciliation method is implemented here.
-
-The [daily context](examples/context-daily.json), [proposal](examples/plan-daily.json),
-[partial acceptance](examples/plan-partial-decision.json),
-[progress](examples/task-progress.json), [outcome](examples/task-outcome.json),
-[cancel request](examples/plan-cancel-request.json), and
-[cancel result](examples/plan-cancel-result.json) are a synthetic host-owned trace.
-Watering was proposed and rejected; it never becomes an accomplished memory.
-
-## Conversation is a separate exchange
-
-Requests and responses require `binding`, `requestId`, `correlationId`,
-`channelId`, `speakerParticipantId`, and `recipientParticipantIds`.
-A Conversation Request additionally carries the environment/context revisions,
-mode, trigger, requested modality, receipt/deadline timestamps, and bounded
-input messages. Messages identify their author participant, message ID, recording
-time, modality, and content. The requested speaker must be the character's
-verified bound participant; sender/channel identity and recipients require
-trusted verification and current authorization.
-
-`host-requested` mode structurally requires `trigger: "host-request"`. Live mode
-may use `host-request`, `incoming-message`, or `presence-event`. A trusted host
-or authorized runtime issues every correlated request; an always-online character
-cannot manufacture a request to bypass profile policy. Reactive live replies
-can be allowed while unrequested speech is forbidden. Proactive output from a
-presence trigger additionally requires `allowUnrequestedMessages: true`,
-authorized scheduling, and channel permission. If any condition is missing,
-produce no output. The flag alone is not a scheduler or permission grant.
-
-Responses add `responseId`, `createdAt`, `expiresAt`, and an output union:
-`{ type: "utterance", modality, content }` or
-`{ type: "no-output", reason }`. Output is a proposal for host/channel delivery,
-not proof anyone heard it. The host must validate the current modality, binding,
-channel, recipients, request correlation, and expiry before delivery. A missing
-speech modality never causes synthesized spoken language; `unsupported-modality`
-is a valid no-output reason. Style preference cannot override embodiment limits.
-
-Conversation Delivery reports identify the response and channel with a unique
-`deliveryId`, timestamp, and `accepted | delivered | rejected | unknown` status.
-Accepted is not delivered; unknown must be reconciled before retrying to avoid
-duplicate speech. Host delivery controls remain authoritative in both modes.
-Model output is not directly broadcast or treated as a world action.
-
-Examples: [host request](examples/conversation-requested-request.json) and
-[utterance](examples/conversation-requested-response.json),
-[live incoming message](examples/conversation-live-request.json) and
-[text reply](examples/conversation-live-response.json),
-[unsupported speech request](examples/conversation-no-speech-request.json) and
-[non-speaking response](examples/conversation-no-speech-response.json).
-The chat profile is persistent but explicitly disallows autonomous unrequested
-messages. No conversation or game connection is actually running.
-
-## Portable memory and intent checkpoints
-
-All memory variants require `memoryId`, `characterId`, `recordedAt`, `scope`, and
-`provenance`. Scope uses stable `participantIds`, audience, and export policy.
-Private means exactly one participant; participants means a nonempty explicit
-list; public means an empty list. These cardinalities and identity verification
-are semantic checks. Public world visibility does not grant public publication
-or export consent. `export` is allowed, consent-required, or forbidden; actual
-access/consent must still be checked. Derived records cannot broaden their
-sources' scope or export rights.
-
-Provenance sources have explicit `sourceType` and `sourceId`. Types distinguish
-authored material, host reports, conversations, memory, derivation, import,
-plan proposals, and plan decisions. References preserve their original meaning:
-
-- A `plan-proposal` proves only that an intention was proposed.
-- A `plan-decision` proves only what the host accepted or rejected.
-- A trusted `host-report` can support a claim about observed execution, subject
-  to its actual reported status and access scope.
-- Authored biography remains authored material, never generated lived history.
-
-| Memory `kind`         | Additional content                                                                                              |
-| --------------------- | --------------------------------------------------------------------------------------------------------------- |
-| `memory-event`        | `eventType`, required original `recordedPayload`, participants; optional summary/environment/session/world time |
-| `working-context`     | Summary, active topics, selected relevant memory IDs, expiry                                                    |
-| `memory-episode`      | Summary, nonempty source memory IDs, period                                                                     |
-| `memory-fact`         | Claim, nonempty source memory IDs, optional uncalibrated confidence                                             |
-| `memory-relationship` | Stable subject/object participants, relation, description, source memory IDs                                    |
-| `intent-checkpoint`   | Bounded portable intents, optional next review time                                                             |
-
-Working context is a bounded current selection, not the full archive. Episodes,
-facts, and relationships are interpretations with provenance; they do not replace
-raw event records or become trusted instructions. Every raw Memory Event requires
-`recordedPayload`, an open JSON object preserving the original authorized content
-that IfSoul actually received, including unknown fields. Its optional `summary`
-is an index annotation and MUST NOT replace the received content. Raw means the
-received observation or message, not omniscient world history. When privacy or
-consent requires redaction, the filtered export must explicitly disclose the
-omission; it MUST NOT claim the remaining summary is the original record. Confidence is a source score,
-not a calibrated probability. Missing or deleted source data requires explicit
-reference/filter handling and deletion or recomputation of affected derivatives.
-No store, retention engine, summarizer, or deletion propagation is implemented.
-
-Each checkpoint intent contains a portable ID, description, desired outcome,
-status, and source memory references. It contains **no host plan/task execution
-handle**, environment binding, inventory, health, combat state, path, skill
-executor continuation, or running game state machine. Text and extensions MUST
-NOT smuggle executable commands or host continuation credentials. Restoring an
-active intention means re-evaluating a desire under a freshly verified profile
-and context, not restarting a host operation. A new host request/decision is
-required before any execution. Portable status cannot prove current host state.
-
-The [proposal memory](examples/memory-event-plan-proposal.json),
-[decision memory](examples/memory-event-proposed.json), and
-[observed outcome memory](examples/memory-event-observed.json) deliberately
-separate intentions, acceptance, and reported facts. The
-[initial checkpoint](examples/intent-checkpoint.json) contains a general learning
-intention, not the daily plan's host task handles.
-
-## Archive: resume or fork explicitly
-
-An Archive Manifest requires `archiveVersion`, `archiveId`, `characterId`,
-`createdAt`, `transferIntent`, `schemaInventory`, `files`, `filter`, `provenance`,
-and `portableCheckpointIds`, in addition to the envelope fields. The archive
-version is independent from wire and provider versions even though its current
-value is also `draft-0.2.0`.
-
-`transferIntent` is exactly one of:
-
-- `{ mode: "resume", expectedCharacterId }`: continue the same portable identity
-  after trusted authorization; the expected ID must equal the snapshot identity.
-- `{ mode: "fork", newCharacterId, newParticipantId }`: request a distinct branch
-  with newly allocated, different character and participant IDs. Preserve source
-  lineage and provenance; the new identity inherits no access rights or consent.
-
-A transfer intent is not itself authorization. Neither mode resumes host
-execution. Existing world bindings, sessions, execution handles, inventory,
-health, provider keys/settings, account credentials, or other secrets MUST NOT
-be exported as portable state. Resume requires fresh host mapping and
-reconciliation. Fork needs an explicit identity/provenance mapping; original
-source authors and participants are not silently relabelled as the new identity.
-No fork implementation or conflict-merge policy is claimed.
-
-`schemaInventory` explicitly lists supported character/memory schema IDs and
-versions. `files` entries have safe relative path, role, schema ID, record count,
-SHA-256 over exact uncompressed bytes, and byte count. Character files contain
-one UTF-8 JSON document; memory files contain UTF-8 JSONL with one complete record
-per nonempty line, and can be sharded. Roles must match the actual tagged kind;
-all referenced schemas must appear in the inventory. Only v0.2 character/memory
-files are admitted. Direct host execution/profile files and model/provider
-configuration are excluded. Authorized original reports and conversation content
-are retained inside memory-event recordedPayload as historical evidence, not
-imported as live execution messages. Historical plan/task/report IDs support
-provenance; they are not active execution handles. A recorded health or inventory
-observation may be remembered, but MUST NOT initialize or overwrite the next
-host embodiment's health or inventory. Secrets remain excluded even from raw
-payloads; any required redaction is explicit filtered export.
-
-A future importer MUST verify unique confined normalized paths, reject traversal,
-symlinks, unlisted entries and platform collisions, enforce decompression/size
-budgets, verify counts/checksums, validate records and versions, check reference
-closure/identity/scope/consent, and operate safely before mutating a destination.
-These checks exceed JSON Schema and are not an implemented import pipeline.
-Checksums prove byte equality, not a trusted author. Missing data is not invented.
-Portable reference closure covers the declared character/memory records and
-their provenance sources, which may retain an original source object inside
-recordedPayload. Historical world/session/entity/revision identifiers inside that
-original content are evidence, not promises to include or reconnect the entire
-host world. They MUST NOT be resolved into live host authority during import.
-
-`filter` explicitly declares complete-authorized or filtered mode, omitted
-categories, reasons, and whether external references remain. Complete-authorized
-means all authorized portable data in the declared source snapshot, including
-original received memory events rather than summaries alone, with empty
-omissions/reasons and no unresolved external references; it does not bypass consent. A filtered export
-records actual exclusions and nonempty reasons without revealing withheld
-content or private IDs. Unknown optional extensions must be preserved or their
-loss disclosed. Unsupported required extensions or versions reject import.
-There is no automatic v0.1-to-v0.2 migration, behavioral equivalence guarantee,
-round-trip implementation, or silent down-conversion.
-
-[Resume](examples/archive-resume.json) and [fork](examples/archive-fork.json)
-manifests describe the same synthetic snapshot after the interaction trace. Their
-hashes cover the two sibling character files and the nine-record
-[memory stream](examples/portable-memory.jsonl), including original host-report,
-plan/decision and chat content, derived selections, relationships, and the
-portable checkpoint. The other envelope examples are not claimed as executable
-archive contents.
-
-## Structural versus semantic validation
-
-Schemas enforce exact versions, tags, closed shapes, typed values, bounded
-exchanges, and the two requested-only communication constraints. A schema pass
-cannot prove identity, authorization, truthful provenance, scope/consent,
-reference closure, intent-catalog membership, argument-descriptor agreement,
-unique selected IDs, acyclic dependencies, valid period ordering, fresh revisions,
-clock accuracy, plan-decision aggregation, actual task effects, checkpoint safety,
-or archive integrity. Future consumers MUST perform the applicable checks before
-acting. Unknown outcomes stay unknown until authoritative reconciliation.
-
-All examples are original synthetic fixtures under the project's normative
-example license. No real conversations, credentials, installed game content,
-third-party characters, or private project code are included.

package/specs/v0.2/examples/archive-fork.json

@@ -1,64 +0,0 @@
-{
-  "schemaVersion": "draft-0.2.0",
-  "kind": "archive-manifest",
-  "archiveVersion": "draft-0.2.0",
-  "archiveId": "archive-mira-fork",
-  "characterId": "character-mira",
-  "createdAt": "2026-01-01T08:03:01Z",
-  "transferIntent": {
-    "mode": "fork",
-    "newCharacterId": "character-mira-fork",
-    "newParticipantId": "participant-mira-fork"
-  },
-  "schemaInventory": [
-    {
-      "schemaId": "urn:ifsoul:protocol:draft-0.2.0:character",
-      "schemaVersion": "draft-0.2.0"
-    },
-    {
-      "schemaId": "urn:ifsoul:protocol:draft-0.2.0:memory",
-      "schemaVersion": "draft-0.2.0"
-    }
-  ],
-  "files": [
-    {
-      "path": "character-template.json",
-      "role": "character-template",
-      "schemaId": "urn:ifsoul:protocol:draft-0.2.0:character",
-      "recordCount": 1,
-      "sha256": "bf726a3b4c4f69a28f41fc83f2b563818da279270bffd8985e9de0dcd12a3644",
-      "sizeBytes": 1269
-    },
-    {
-      "path": "character-instance.json",
-      "role": "character-instance",
-      "schemaId": "urn:ifsoul:protocol:draft-0.2.0:character",
-      "recordCount": 1,
-      "sha256": "8b803f886b4749e38af239e9be66809556333bccd61644d9b6bc314b20d2c4ee",
-      "sizeBytes": 1580
-    },
-    {
-      "path": "portable-memory.jsonl",
-      "role": "memory",
-      "schemaId": "urn:ifsoul:protocol:draft-0.2.0:memory",
-      "recordCount": 9,
-      "sha256": "a2c58c69e5901fc35625149da2d30e063bce3939a241a412ac1e17f5aafbef2f",
-      "sizeBytes": 7325
-    }
-  ],
-  "filter": {
-    "mode": "complete-authorized",
-    "omittedCategories": [],
-    "reasons": [],
-    "externalReferences": false
-  },
-  "provenance": {
-    "sources": [
-      {
-        "sourceType": "authored",
-        "sourceId": "template-mira-v2"
-      }
-    ]
-  },
-  "portableCheckpointIds": ["checkpoint-initial"]
-}

package/specs/v0.2/examples/archive-resume.json

@@ -1,63 +0,0 @@
-{
-  "schemaVersion": "draft-0.2.0",
-  "kind": "archive-manifest",
-  "archiveVersion": "draft-0.2.0",
-  "archiveId": "archive-mira-resume",
-  "characterId": "character-mira",
-  "createdAt": "2026-01-01T08:03:01Z",
-  "transferIntent": {
-    "mode": "resume",
-    "expectedCharacterId": "character-mira"
-  },
-  "schemaInventory": [
-    {
-      "schemaId": "urn:ifsoul:protocol:draft-0.2.0:character",
-      "schemaVersion": "draft-0.2.0"
-    },
-    {
-      "schemaId": "urn:ifsoul:protocol:draft-0.2.0:memory",
-      "schemaVersion": "draft-0.2.0"
-    }
-  ],
-  "files": [
-    {
-      "path": "character-template.json",
-      "role": "character-template",
-      "schemaId": "urn:ifsoul:protocol:draft-0.2.0:character",
-      "recordCount": 1,
-      "sha256": "bf726a3b4c4f69a28f41fc83f2b563818da279270bffd8985e9de0dcd12a3644",
-      "sizeBytes": 1269
-    },
-    {
-      "path": "character-instance.json",
-      "role": "character-instance",
-      "schemaId": "urn:ifsoul:protocol:draft-0.2.0:character",
-      "recordCount": 1,
-      "sha256": "8b803f886b4749e38af239e9be66809556333bccd61644d9b6bc314b20d2c4ee",
-      "sizeBytes": 1580
-    },
-    {
-      "path": "portable-memory.jsonl",
-      "role": "memory",
-      "schemaId": "urn:ifsoul:protocol:draft-0.2.0:memory",
-      "recordCount": 9,
-      "sha256": "a2c58c69e5901fc35625149da2d30e063bce3939a241a412ac1e17f5aafbef2f",
-      "sizeBytes": 7325
-    }
-  ],
-  "filter": {
-    "mode": "complete-authorized",
-    "omittedCategories": [],
-    "reasons": [],
-    "externalReferences": false
-  },
-  "provenance": {
-    "sources": [
-      {
-        "sourceType": "authored",
-        "sourceId": "template-mira-v2"
-      }
-    ]
-  },
-  "portableCheckpointIds": ["checkpoint-initial"]
-}

package/specs/v0.2/examples/character-instance.json

@@ -1,57 +0,0 @@
-{
-  "schemaVersion": "draft-0.2.0",
-  "kind": "character-instance",
-  "characterId": "character-mira",
-  "participantId": "participant-mira",
-  "createdAt": "2026-01-01T08:00:00Z",
-  "template": {
-    "templateId": "template-mira-v2",
-    "revision": 1
-  },
-  "profile": {
-    "name": "Mira Vale",
-    "description": "An original synthetic character who enjoys careful gardening.",
-    "traits": {
-      "patience": "high",
-      "curiosity": 0.7
-    },
-    "preferences": [
-      {
-        "preferenceId": "quiet",
-        "subject": "conversation setting",
-        "stance": "seek",
-        "description": "Prefers calm settings when available.",
-        "strength": 0.8
-      }
-    ],
-    "speechStyle": {
-      "description": "Uses concise, warm language when the current embodiment supports language.",
-      "languages": ["en"],
-      "habits": ["Asks one question at a time."]
-    },
-    "goals": [
-      {
-        "goalId": "grow-herbs",
-        "description": "Learn to care for a small herb garden.",
-        "priority": "normal",
-        "status": "desired"
-      }
-    ],
-    "authoredBiography": {
-      "text": "Mira grew up beside an imagined hillside garden. This is authored fiction, not a record of lived sessions.",
-      "fictional": true
-    }
-  },
-  "memoryRecordIds": [
-    "checkpoint-initial",
-    "event-inspection",
-    "event-visitor-question",
-    "event-plan-proposal",
-    "event-plan-proposed",
-    "working-garden",
-    "episode-garden",
-    "fact-inspection",
-    "relationship-visitor"
-  ],
-  "relationshipMemoryIds": ["relationship-visitor"]
-}

package/specs/v0.2/examples/character-template.json

@@ -1,45 +0,0 @@
-{
-  "schemaVersion": "draft-0.2.0",
-  "kind": "character-template",
-  "templateId": "template-mira-v2",
-  "revision": 1,
-  "profile": {
-    "name": "Mira Vale",
-    "description": "An original synthetic character who enjoys careful gardening.",
-    "traits": {
-      "patience": "high",
-      "curiosity": 0.7
-    },
-    "preferences": [
-      {
-        "preferenceId": "quiet",
-        "subject": "conversation setting",
-        "stance": "seek",
-        "description": "Prefers calm settings when available.",
-        "strength": 0.8
-      }
-    ],
-    "speechStyle": {
-      "description": "Uses concise, warm language when the current embodiment supports language.",
-      "languages": ["en"],
-      "habits": ["Asks one question at a time."]
-    },
-    "goals": [
-      {
-        "goalId": "grow-herbs",
-        "description": "Learn to care for a small herb garden.",
-        "priority": "normal",
-        "status": "desired"
-      }
-    ],
-    "authoredBiography": {
-      "text": "Mira grew up beside an imagined hillside garden. This is authored fiction, not a record of lived sessions.",
-      "fictional": true
-    }
-  },
-  "authorship": {
-    "authors": ["IfSoul contributors"],
-    "license": "Apache-2.0",
-    "source": "Original synthetic example"
-  }
-}