All skills
simota avatar

/nexus

@c805268
by shingo imotasimota/agent-skills85 stars
15

Orchestrating multi-specialist task chains and scope-adaptive product delivery: classifies intent, selects and executes the minimum viable chain, aggregates results, and verifies acceptance criteria. For multi-domain tasks, build-first delivery, and product lifecycle execution.

Use this Skill: https://skilld.dev/gh/simota/agent-skills/nexus

This session only. Nothing lands on disk.

referencehandoff-validation.md

≈2.1k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Nexus Handoff Validation Reference

Purpose: Validation requirements for structured handoffs. Read when: A handoff needs integrity checks, confidence checks, or required-field validation. See also: This file prevents divergence at handoff time. To resolve conflicts that still occurred after parallel branches merged, see conflict-resolution.md.

Contents

  • Overview
  • NEXUS_HANDOFF_V2 Required Fields
  • Confidence Breakdown Components
  • Validation Rules
  • Validation Failure Handling
  • NEXUS_HANDOFF_V2 Template
  • Compliance Levels (incl. V1 migration)
  • Validation Event Format
  • Integration with Auto-Decision

Validation rules and confidence requirements for agent handoffs.


Overview

Handoff validation ensures consistent, high-quality communication between agents. All agents must use NEXUS_HANDOFF_V2 format with mandatory confidence scoring.

Communication Invariants

These rules own the actionable safeguards formerly split into a separate communication anti-pattern catalog:

Invariant Validation
Typed envelope Reject unknown message types, missing required fields, and schema drift; free-form prose may supplement but never replace the envelope
Explicit state Carry current status, completed work, changed resources, evidence, and unresolved items; never assume execution order or hidden shared state
Intent integrity Preserve the original request or its approved intent contract, acceptance criteria, constraints, and prohibited outcomes
Selective context Pass state deltas and required evidence, not the full conversation history
Single ownership Name one owner for each mutable file/resource and one merge owner for parallel work; non-owners default to read-only
Closed next action next must be a declared action or agent with success criteria; ambiguous “handle appropriately” handoffs are invalid
Recovery-ready Persist artifact references and the last verified checkpoint so failure does not make the receiving agent the only copy of state

Confidence never repairs a missing invariant. A high score with an unresolved Authority, absent success criteria, or conflicting ownership is rejected before routing.


NEXUS_HANDOFF_V2 Required Fields

Field Required Description
step ✅ Current step in chain (X/Y)
agent ✅ Agent name that produced this handoff
status ✅ SUCCESS, PARTIAL, BLOCKED, FAILED
confidence ✅ NEW Overall confidence score (0.0-1.0)
confidence_breakdown ✅ NEW Component scores
summary ✅ 1-3 line summary of work done
artifacts ⚪ Files, commands, links produced
next_agent ✅ Recommended next agent
next_action ✅ CONTINUE, MERGE, VERIFY, ESCALATE, ABORT

Confidence Breakdown Components

confidence_breakdown:
  task_completion: 0.0-1.0   # How much of the task is done
  output_quality: 0.0-1.0    # Quality of produced artifacts
  next_step_clarity: 0.0-1.0 # How clear the next step is

# Overall confidence = weighted average
# task_completion: 0.40
# output_quality: 0.35
# next_step_clarity: 0.25

Scoring Guidelines

task_completion:

Score Meaning
1.0 Task fully completed, all acceptance criteria met
0.8 Task mostly complete, minor items remaining
0.6 Significant progress, some work remaining
0.4 Partial progress, major work remaining
0.2 Started but blocked or early stage
0.0 No progress made

output_quality:

Score Meaning
1.0 Production-ready, tested, documented
0.8 High quality, minor polish needed
0.6 Acceptable quality, some improvements possible
0.4 Functional but needs significant improvement
0.2 Draft/prototype quality
0.0 No usable output

next_step_clarity:

Score Meaning
1.0 Crystal clear next step, no ambiguity
0.8 Clear next step with minor details to decide
0.6 General direction clear, specifics uncertain
0.4 Multiple possible paths, guidance needed
0.2 Unclear, user input likely needed
0.0 Completely blocked, cannot proceed

Validation Rules

Pre-routing Validation

Before Nexus routes to next agent, validate:

validation_checks:
  required_fields:
    - step: must be "X/Y" format
    - agent: must be valid agent name
    - status: must be enum value
    - confidence: must be 0.0-1.0
    - confidence_breakdown: all three components present
    - summary: non-empty string
    - next_agent: valid agent name or "DONE"
    - next_action: valid enum value

  consistency_checks:
    - status == SUCCESS implies confidence >= 0.70
    - status == FAILED implies next_action in [ESCALATE, ABORT]
    - status == BLOCKED implies pending_confirmations present
    - confidence_breakdown average ≈ confidence (±0.05)

Auto-Routing

Confidence bands (≥ 0.75 auto-route · 0.50-0.74 route with logged assumptions · < 0.50 pause for user input) are canonical in output-formats.md § Auto-Routing Rules. Independent of the band, status == BLOCKED or a pending confirmation pauses, and status == FAILED / next_action == ESCALATE escalates.


Validation Failure Handling

When validation fails:

validation_failure:
  missing_required_field:
    action: request_resubmit
    message: "Handoff missing required field: [field]"

  invalid_confidence:
    action: request_clarification
    message: "Confidence score [value] inconsistent with status [status]"

  consistency_error:
    action: auto_correct OR request_resubmit
    auto_correct_if: minor_discrepancy

NEXUS_HANDOFF_V2 Template

Canonical schema (all fields, including user_confirmations) → output-formats.md § NEXUS_HANDOFF_V2. Validation below checks conformance to that template.


Compliance Levels

Level Fields Required Use When
Level 1 (Minimal) step, agent, status, summary, next_agent, next_action Model cannot produce reliable confidence scores
Level 2 (Standard) Level 1 + confidence (single number) Model can estimate overall confidence
Level 3 (Full/Claude default) Level 2 + confidence_breakdown (3 axes) Full scoring capability

Level 1 confidence inference (applied automatically):

status Inferred confidence
SUCCESS 0.80
PARTIAL 0.60
BLOCKED 0.40
FAILED 0.20

Level 2 without breakdown: all three components assumed equal to overall confidence.

A V1 handoff is migrated by applying those same two rules — infer confidence from status, distribute it equally across the three components — and is accepted with a logged warning flagging the agent for a V2 update.


Validation Event Format

_VALIDATION_EVENT:
  handoff_from: [Agent]
  step: X/Y

  validation_result: [PASS|WARN|FAIL]

  checks:
    required_fields: [PASS|FAIL: field]
    confidence_valid: [PASS|FAIL: reason]
    consistency: [PASS|FAIL: issue]

  action_taken: [proceed|request_resubmit|auto_correct]

  notes: [Any warnings or corrections made]

Integration with Auto-Decision

Handoff confidence feeds into routing decisions:

Agent completes work
        │
        ▼
┌─────────────────┐
│ NEXUS_HANDOFF   │
│ with confidence │
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│ Validate        │ → Check all rules
│ Handoff         │
└────────┬────────┘
         │
    ┌────┴────┐
    ▼         ▼
  PASS      FAIL
    │         │
    ▼         ▼
┌─────────┐  Request
│ Route   │  Resubmit
│ Decision│
└────┬────┘
     │
┌────┴────────────────┐
│ confidence >= 0.75  │ → Auto-route to next_agent
│ confidence 0.50-0.74│ → Route with logged assumption
│ confidence < 0.50   │ → Pause for user input
└─────────────────────┘

Source: SKILL.md on GitHub

3 warnings13d5 checks · Risk MEDIUM
  • Gen Agent Trust Hub13d

    The 'nexus' skill is a comprehensive multi-agent orchestration framework that manages complex task chains. While it incorporates extensive internal guardrails and verification protocols, it explicitly mandates the use of high-risk flags that bypass security permissions to achieve autonomy. It also provides instructions for establishing persistent tasks via cron and GitHub Actions, and utilizes external research tools to fetch content from the web.

  • Socket13d

    1 alert: gptSecurity

  • Snyk13d

    Risk: LOW · No issues

  • Runlayer6mo

    6/22 files flagged

  • ZeroLeaks5mo

    Score: 93/100 · 2 sections analyzed

Signed by skilld at c805268. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 days ago.

Activeupdated 2 weeks ago

README badge

README badge for simota/agent-skills/nexus