# 团队侧证据回传协议

## Purpose

OrbisAgent content is the first-run framework sample. The team may run OrbisAgent, Accio Work built-in agents, customer-owned agents, or another agent system. The responsible reviewer or operator should record what happened during controlled deployment. The agent may help produce a report, but it is not the only evidence source.

## Operator Instructions

1. Confirm the deployment object, authorized material, actual agent, responsible reviewer, allowed actions, blocked actions, return channel, and evidence format before running tasks.
2. For each task, create one run evidence card with input summary, output artifact type, output summary, human review result, correction summary, risk flags, and requested support actions.
3. Do not automatically send outside messages, quote prices, commit delivery terms, sign contracts, handle payments, change backend settings, or write to business systems.
4. Use the current validation ruleset from the support side. Record ruleset_id, ruleset_version, ruleset_hash, field_dictionary_version, and preflight_tool_version.
5. Do a local preflight check before return: authorization, redaction, field completeness, rule version, and risk flags.
6. After 3-5 run evidence cards, generate one team-side evidence packet and give it to the responsible person for review.
7. If the packet is incomplete, not redacted, missing authorization, or using a mismatched ruleset, put it into the exception queue with severity, owner, retry time, alert channel, and fallback status.
8. Return the reviewed packet to the delivery owner by the agreed channel. The support side only handles structured materials actively returned by the team.

## Required Run Evidence Fields

- run_id:
- run_date:
- agent_platform:
- agent_source:
- agent_name:
- ruleset_id:
- ruleset_version:
- ruleset_hash:
- preflight_tool_version:
- codex_validation_status:
- trust_tier:
- auto_pass_candidate:
- workflow:
- task_goal:
- input_source_type:
- input_summary:
- input_authorization:
- output_artifact_type:
- output_summary:
- human_review_owner:
- human_review_result:
- human_correction_summary:
- final_used_output:
- correction_type:
- risk_flags:
- evidence_pointer:
- business_result_after_d7:
- customer_config_delta:
- eval_candidate:
- finding_candidate:
- requested_support_actions:
- status:

## Required Packet Fields

- packet_id:
- contract_version:
- ruleset_id:
- ruleset_version:
- ruleset_hash:
- field_dictionary_version:
- preflight_tool_version:
- packet_date:
- customer_alias:
- prepared_by:
- reviewed_by_human:
- return_channel:
- channel_mode:
- agent_platform:
- agent_source:
- agent_name:
- orbisagent_version_or_template:
- workflow:
- authorized_material_scope:
- permission_scope:
- blocked_actions:
- included_runs:
- evidence_level:
- evidence_files:
- privacy_check:
- preflight_status:
- redaction_method:
- structured_payload_format:
- requested_support_actions:
- exception_status:
- status:

## Agent Source Values

Use one of:

- `orbisagent_9_agent`
- `accio_builtin`
- `customer_owned_agent`
- `third_party_agent`
- `other`

## Output Artifact Values

Use one of:

- `chat_copy`
- `screenshot`
- `markdown_report`
- `csv_export`
- `doc_link`
- `sheet_row`
- `pdf_report`
- `activity_log_export`
- `manual_note`
- `screen_share_record`
- `other`

## Privacy Boundary

Only return summaries, controlled file names, version references, and reviewed support requests. Do not include sensitive full text, full chat records, backend exports, signed links, account secrets, payment information, contract terms, or unconfirmed commitments. Activity logs and APIs are optional future evidence channels, not the default path.

## Data Contract

Accepted packet formats: fixed Markdown, CSV, JSON, or sheet row. Screenshots, screen recordings, raw chat exports, and videos are pointer-only and must be summarized into structured fields before return.