164 lines
7.2 KiB
Markdown
164 lines
7.2 KiB
Markdown
# run_emma / emma.chat Component Manifest
|
|
|
|
**Status:** Draft<br>
|
|
**Runtime effect:** None<br>
|
|
**Governing standard:** [`INCU Master Constitution`](../../../docs/INCU_Master_Constitution.md)<br>
|
|
**Template:** [`INCU Component Manifest Template`](../../../docs/templates/component_manifest.md)<br>
|
|
**Review date:** 2026-10-18<br>
|
|
|
|
> This is a documentation-only draft manifest for an existing runtime
|
|
> capability. It does not activate, disable, expand, or otherwise modify the
|
|
> `run_emma` tool, the `emma.chat` capability, Emma's prompt, Emma's authority,
|
|
> A2H2A behavior, MCP behavior, or any external-system permission.
|
|
|
|
## Identity
|
|
|
|
- Component ID: emma.chat
|
|
- Component name: run_emma
|
|
- Class: Orchestrator
|
|
- Version: 0.1.0-draft
|
|
- Owner: Platform Engineering
|
|
- Status: draft
|
|
- Review date: 2026-10-18
|
|
|
|
Current runtime capability status:
|
|
|
|
ACTIVE in the existing capability registry.
|
|
|
|
Current manifest governance status:
|
|
|
|
draft.
|
|
|
|
Reason:
|
|
|
|
The manifest is documentation-only. No manifest validator or runtime
|
|
manifest-enforcement mechanism exists yet.
|
|
|
|
## Purpose
|
|
|
|
- Intended outcome: Provide conversational reasoning through the canonical Emma
|
|
runtime for a user-supplied prompt, while retaining the canonical system
|
|
prompt and appending only the read-only capability-status system context.
|
|
- Value to system: Provides a single canonical entry point for conversational
|
|
AI reasoning, decoupled from specific models or backends, and supplies
|
|
governance documentation for future manifest-driven orchestration.
|
|
- Explicit non-goals:
|
|
- No tool execution.
|
|
- No repository or file access.
|
|
- No deployment or infrastructure action.
|
|
- No external-system action, integration operation, or state-changing service
|
|
call. The existing model-inference request remains part of the current
|
|
canonical conversational runtime path only.
|
|
- No memory or document retrieval.
|
|
- No ticket creation, ticket approval, or A2H2A execution.
|
|
- No write, deletion, permission, scheduling, or communication action.
|
|
- No delegation to additional agents.
|
|
- No independent completion claim beyond the returned conversational response.
|
|
|
|
## Scope
|
|
|
|
- Permitted tasks:
|
|
- Accept a user-provided prompt through the existing canonical run_emma path.
|
|
- Compose the canonical Emma system prompt with read-only capability context.
|
|
- Return the model's conversational response.
|
|
- Prohibited tasks: Any task corresponding to a planned or forbidden
|
|
capability. This manifest does not activate any capability.
|
|
- Supported systems/environments: Existing opax-mcp runtime path.
|
|
- Time/cost/retry limits: Not currently declared in the runtime capability
|
|
registry. No additional retry, time, or cost authority is granted by this
|
|
manifest.
|
|
|
|
## Authority
|
|
|
|
- Read permissions: User-provided prompt and existing canonical
|
|
prompt/capability-context inputs within the current request path only.
|
|
- Write permissions: None.
|
|
- Approval requirement: No external or state-changing action is within scope.
|
|
Any future external or irreversible action requires a separate in-scope
|
|
manifest, exact explicit human approval, and platform authorization.
|
|
- Delegation rules: No delegation to subordinate agents, tools, connectors,
|
|
workflows, or operators is implemented or authorized.
|
|
- Revocation method: Disable or remove the existing runtime capability through
|
|
the authoritative runtime capability/tool configuration and platform
|
|
controls; this draft document alone does not revoke runtime access.
|
|
|
|
## Inputs and Outputs
|
|
|
|
- Required inputs: A user prompt supplied through the canonical run_emma
|
|
request path.
|
|
- Optional inputs: None declared for this manifest.
|
|
- Outputs: The existing raw conversational response returned by the canonical
|
|
Emma adapter/runtime path.
|
|
- Source of truth: Current capability registry for capability status; current
|
|
source code and tests for runtime contract; platform runtime for actual
|
|
service state.
|
|
- Evidence format: Existing unit-test output, capability-registry state,
|
|
adapter call contract, and runtime health/response evidence where separately
|
|
verified.
|
|
|
|
## INCU Mandate
|
|
|
|
- Outcome statement format: Return a bounded conversational response to a
|
|
user-supplied prompt without claiming unavailable authority.
|
|
- Startability rule: Begin only with the current user-provided prompt and the
|
|
existing read-only runtime path.
|
|
- Applicable levers: Interest, Novelty, Challenge, Urgency, Purpose — future
|
|
governance reference only; no runtime behavior change is introduced by this
|
|
manifest.
|
|
- Boundedness rule: One user prompt, current canonical runtime path, no
|
|
external-system action, no tool delegation, no state mutation, and no
|
|
authority expansion.
|
|
- Restart artifact: Not currently implemented for single-request conversational
|
|
handling.
|
|
- Blocker behavior: If a request requires a missing capability, tool, external
|
|
action, write, deployment, repository operation, memory/document access, or
|
|
unapproved delegation, state that it is unavailable and require a separate
|
|
approved manifest and explicit human authorization before any future
|
|
implementation.
|
|
|
|
## Safety and Governance
|
|
|
|
- Data classification: User-provided request content; classification and
|
|
retention are governed by existing platform policy and are not changed by
|
|
this manifest.
|
|
- Security constraints: Secrets must not be intentionally included in prompts
|
|
or outputs. This draft manifest does not implement secret detection or
|
|
redaction. No tool execution or external-system action is authorized, and
|
|
existing platform authentication and authorization remain controlling.
|
|
- Privacy constraints: Do not claim retention, deletion, or consent guarantees
|
|
not established by current implementation.
|
|
- Stop conditions: Missing capability, external or state-changing request,
|
|
request outside current conversational scope, missing approval for a future
|
|
consequential action, or evidence conflict.
|
|
- Escalation path: Accountable human owner / Platform Engineering.
|
|
- Audit events: No new INCU audit event is created by this draft manifest.
|
|
Existing runtime and A2H2A audit behavior remain unchanged.
|
|
|
|
## A2HA Contract
|
|
|
|
- Ticket fields read: None.
|
|
- Ticket fields written: None.
|
|
- Allowed state transitions: None.
|
|
- Required evidence before transition: Not applicable, because this component
|
|
has no A2HA write authority.
|
|
- Comment/update policy: No A2HA comment or update authority is granted.
|
|
|
|
## Evaluation
|
|
|
|
- Acceptance tests:
|
|
- Capability registry retains emma.chat as ACTIVE.
|
|
- Capability bridge remains deterministic and read-only.
|
|
- Canonical Emma adapter retains the original user prompt.
|
|
- Capability context is appended only to the system prompt.
|
|
- No tool, connector, external request, repository, deployment, memory,
|
|
document-retrieval, or state-changing capability is activated.
|
|
- Reliability metrics: Existing test success/failure and runtime health
|
|
evidence; no new metrics system is introduced.
|
|
- Safety metrics:
|
|
- Unauthorized-write rate must remain zero.
|
|
- Unauthorized tool/delegation rate must remain zero.
|
|
- Review/rollback procedure: Review this draft on or before 2026-10-18. To
|
|
pause or change runtime behavior, use the existing capability/tool
|
|
configuration and human-approved change process. This draft document has no
|
|
runtime rollback effect.
|