菜单
Use when an agent is interacting with UNITARES governance for the first time, needs to onboard, check in, or recover from a pause/reject verdict. Covers the full agent lifecycle from session start through check-ins to recovery.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | governance-lifecycle |
| description | Use when an agent is interacting with UNITARES governance for the first time, needs to onboard, check in, or recover from a pause/reject verdict. Covers the full agent lifecycle from session start through check-ins to recovery. |
| last_verified | 2026-06-29 |
| freshness_days | 14 |
| source_files | ["unitares/src/mcp_handlers/core.py","unitares/src/mcp_handlers/identity/handlers.py","unitares/src/mcp_handlers/admin/handlers.py","unitares/src/mcp_handlers/tool_stability.py","unitares/src/mcp_handlers/middleware/envelope_step.py"] |
Last Updated: 2026-06-29
The core lifecycle should use primary task-verb tools. Each is implemented by a raw tool with the same parameters and identity rules, and returns a normalized envelope: the operationally useful fields first (next_action, state_summary, risk_summary, memory_suggestions, recovery_hint), with the full raw payload preserved under raw_governance.
| Task | Primary workflow tool | Raw implementation tool |
|---|---|---|
| Start a fresh process identity | start_session(force_new=true, ...) | onboard |
| Check in after meaningful work | sync_state(response_text=..., complexity=...) | process_agent_update |
| Check your working state | check_working_state() | get_governance_metrics |
| Avoid duplicate work | search_shared_memory(query=...) | knowledge(action="search") |
| Record what actually happened | record_result(...) | outcome_event |
| Ask for a structured review | request_review(issue_description=...) | dialectic(action="request") |
Use the primary workflow tools by default. Use raw implementation names only for older servers, compatibility code, or when you explicitly need the unwrapped handler response. start_session(force_new=true) is a process-start operation, not a per-turn continuation primitive.
Choose creation, lineage, or proof-owned resume explicitly:
start_session(force_new=true) # one fresh process identity — the default; co-location is not lineage
start_session(force_new=true, parent_agent_id="<dispatcher-uuid>",
spawn_reason="subagent") # dispatched subagent (usually set automatically by the dispatcher)
start_session(force_new=true, parent_agent_id="<prior-uuid>",
spawn_reason="new_session") # handoff from a finished prior session
identity(agent_uuid="<uuid>", continuity_token="<token>", resume=true) # same live owner / proof-owned rebind
Declaring a currently-live agent as parent is rejected (lineage_coincidental_rejected): a live agent is a concurrent sibling, not a predecessor. subagent and compaction are exempt — their parent is legitimately live. A genuine handoff to an exited predecessor stays provisional until R1 confirms it. Continuing the same still-running process means reusing the active binding or client_session_id, not minting another child.
Use raw onboard(...) instead when targeting older servers or when you
need the unwrapped raw response.
Returns:
session_resolution_source, identity_assurance, and deprecation warnings when relevantname= is a cosmetic label, not a resume key. Passing the same name on a later session does not prove identity.
Default rules:
start_session(force_new=true) with no parent. Co-location in a workspace is not lineage.parent_agent_id="<dispatcher-uuid>", spawn_reason="subagent", usually set automatically by the dispatcher) or a handoff from a finished prior session (parent_agent_id="<prior-uuid>", spawn_reason="new_session"). Declaring a currently-live agent as parent is rejected.identity(agent_uuid="<uuid>", continuity_token="<token>", resume=true).client_session_id; reserve continuity_token for explicit proof-owned rebinds.Avoid these patterns:
identity(agent_uuid=X, resume=true): UUID alone is an unsigned claim. It currently logs/emits hijack-suspected telemetry and is strict-mode rejected when UNITARES_IDENTITY_STRICT=strict.onboard(continuity_token=...) as cross-process resume: S1-a accepts this only during the deprecation window and returns a warning. Declare lineage with parent_agent_id instead.onboard(): older code may still pin-resume by weak session/IP:UA evidence. Use force_new=true when creating a new process identity.continuity_token is now intentionally narrow: 1-hour TTL, rolling, and retained as possession proof for anti-hijack gates. It does not establish process-instance continuity by itself.
Call sync_state() after meaningful work:
sync_state(
response_text: "Brief summary of what you did",
complexity: 0.0-1.0, # Task difficulty estimate
confidence: 0.0-1.0 # How confident you are (be honest)
)
Use raw process_agent_update(...) when you need the raw handler payload;
primary workflow responses preserve it under raw_governance.
A verdict plus current EISV metrics. Read the verdict and act on it.
| Verdict | What to Do |
|---|---|
| proceed | Continue normally |
| guide + guidance text | Read the guidance, adjust your approach, keep going |
| pause | Stop your current task. Reflect on what is flagged. Consider requesting a dialectic review |
| reject | Significant concern. Requires dialectic review or human intervention |
| margin: tight | You are near a basin edge. Be more careful with next steps |
A guide verdict is an early warning. Ignoring it makes pause more likely.
client_session_id, or short-lived continuity tokenidentity() when continuity seems unclearidentity_statusbound_identitysession_resolution_sourcecontinuity_token_supportedidentity_assurancedeprecationsStrong ownership proof is better than implicit continuity. If the runtime falls back to weak signals such as fingerprinting, mint a fresh process identity and declare lineage.
When you are paused, stuck, or need intervention:
| Situation | Tool | Notes |
|---|---|---|
| Stuck or paused, want automatic recovery | self_recovery() | Attempts to restore healthy state |
| Disagree with verdict, want structured review | request_dialectic_review() | Starts thesis/antithesis/synthesis process |
| Manual override needed | operator_resume_agent() | Requires human/operator action |
Recovery is not a shortcut — self_recovery() examines your EISV state and determines if resumption is safe. If your metrics are genuinely degraded, it will not force a resume.
start_session(force_new=true, parent_agent_id=...) — Create a fresh process identity once, optionally declaring lineagesync_state() — Check in with work summary, complexity, confidencecheck_working_state() — Read your current EISV stateidentity() — Confirm who the runtime thinks you are and how continuity was resolved; include continuity_token for proof-owned UUID rebindshealth_check() — Check operator-facing server health when behavior seems oddsearch_shared_memory(query=...) — Find existing knowledge before creating new entriesknowledge(action="note", ...) — Quick contribution to the knowledge graphknowledge() — Full knowledge graph CRUD (store, update, details, cleanup)agent() — Agent lifecycle (list, archive, get details)calibration() — Check or update calibration datarequest_dialectic_review() — Start a dialectic sessionexport() — Export session historycall_model() — Delegate to a secondary LLM for analysisdetect_stuck_agents() — Find unresponsive agentsself_recovery() — Resume from stuck or paused statesubmit_thesis() / submit_antithesis() / submit_synthesis() — Dialectic participationUse when an agent needs to understand UNITARES governance concepts — EISV state vectors, basins, policy actions, coherence, calibration. Reference material for interpreting governance metrics as proprioceptive state estimation, not outcome judgment.
Use when an agent is participating in a UNITARES dialectic session — paused and needs to submit a thesis, reviewing another agent's thesis, or synthesizing conditions for resolution. Covers structured argumentation and convergence.
Use when adding, editing, or reviewing sections on the unitares dashboard (dashboard/redesign/). Captures the redesign's conventions: the section-module pattern (window.X = { load }), the live-or-snapshot data seam, theme-aware charts via design tokens, and the app.html wiring (nav / pane / lazyLoad / RELOAD / retheme). A repo-specific reference — not general dashboard advice.
Compatibility umbrella skill for the UNITARES governance framework. Use this as the entrypoint when you need the overall model and route into the split governance skills.
Use when setting up or operating the UNITARES Discord bridge — a standalone bot that surfaces governance events, agent presence, Lumen's state, and autonomous governance actions as a living Discord server.
Use when an agent needs to search the shared knowledge graph, contribute a discovery, or update existing entries. Covers search, tagging, discovery types, and status lifecycle.