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.

referencedeliver-recipe.md

≈2k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Deliver Recipe — Scope-Adaptive Product Delivery

Purpose: Build and ship a product or MVP through the minimum viable specialist chain, sizing discovery, implementation, verification, and release work to the actual scope.

Read when: deliver matches, the user asks to build a product from scratch, or an MVP/product-lifecycle request is broader than one feature but does not justify the fixed high-investment apex shape.


Recipe Admission Record

Gate Result Evidence
Cross-boundary PASS Product delivery crosses discovery, implementation, verification, and release ownership.
Reusable orchestration value PASS The absorbed delivery workflow carried independent small, medium, and epic request shapes; all need the same scope-classification and anti-stall contract with different chain sizes.
No ownership overlap PASS feature owns one bounded capability; apex owns a high-investment discovery-to-ship run; deliver owns scope-adaptive product/MVP delivery.
Closed execution contract PASS Nexus owns orchestration; specialists own phase artifacts; Radar/Attest/Guardian are independent verification oracles; termination and residual routes are defined below.

Parameters

scope: auto | small | medium | epic
target: <product, MVP, or release objective>
mode: build | plan-and-build | resume
release: true | false
checkpoint: .agents/nexus-delivery-state.md

scope=auto is the default. Classify with evidence from repository shape, requested outcomes, public-interface impact, and dependency breadth:

Scope Typical surface Chain posture
small 1-5 files, one dominant domain, no architecture migration 3-5 specialists; skip optional discovery artifacts when evidence already exists
medium 6-15 files or 2-3 domains, integration or public-interface change 5-9 specialists; explicit specification and impact gates
epic More than 15 files, multiple workstreams, migration/release coordination 8-18 specialists; Sherpa decomposition and parallel hub-spoke branches

File counts are evidence, not automatic permission. Safety and blast radius can raise the scope tier.

Phase Contract

CLASSIFY → PREFLIGHT → SHAPE → BUILD → VERIFY → RELEASE? → DELIVER

Phase Nexus responsibility Specialist ownership Exit evidence
CLASSIFY Freeze target, non-goals, scope tier, and success signals Lens/PDM may inspect current state Delivery Intent + declared tier
PREFLIGHT Reuse existing artifacts and expose blockers before implementation Lens, Ripple, Schema/Gateway/Atlas as needed Reuse Map + Impact Notes
SHAPE Select only the chain required by the tier Scribe/Sherpa for ACs and decomposition Traceable ACs + bounded work packages
BUILD Execute work packages through hub-spoke delegation Builder/Artisan/Native/Scaffold and domain owners Working artifacts; no placeholder residue
VERIFY Keep producer and verifier separate Radar, Attest, Sentinel, Voyager as applicable Build/test/lint evidence + AC disposition
RELEASE Prepare or execute release only when release=true and authorized Guardian, Launch, Gear Release evidence or an explicit release-ready handoff
DELIVER Aggregate evidence and residuals Nexus NEXUS_COMPLETE + Delivery Report

Checkpoint format → reference/delivery-output-formats.md (read when saving/resuming). Checkpoint-resume applies: persist phase outputs to .agents/nexus-delivery-state.md at every boundary. mode=resume continues from the last valid checkpoint after revalidating repository state and permissions.

Chain Selection

  • small: Lens?[reuse] → Builder/Artisan → Radar → Guardian?
  • medium: Lens → Scribe/Sherpa → Ripple? → domain implementer(s) → Radar/Attest → Guardian → Launch?
  • epic: Lens → Sherpa → (domain branches in parallel) → Radar/Attest/Sentinel → Guardian → Launch?

The hub may omit an optional step only with a recorded reason and equivalent evidence. Domain methodology remains in the owning skill; this recipe owns order, handoffs, termination, and verification only.

Anti-Stall and Recovery

Use the bounded recovery ladder from reference/delivery-anti-stall-engine.md: retry only after diagnosis, then narrow the work package, switch to the documented alternate owner/tool, or mark a typed residual. Two identical failures trigger diagnosis rather than a third blind retry. Do not lower acceptance criteria to preserve momentum.

Exit reasons are ACCEPT, target-met, BLOCK, denied, invalid-state, or cancelled. This is a non-loop recipe; Termination bound: N/A — every work package reaches a terminal state exactly once. Any non-ACCEPT exit reports best-so-far, the residual gap, and the named route.

Confirm / Safety Gate

Default tier is announce-and-proceed (no objection window) for reversible in-scope delivery. Use Ask First for L4 security, destructive data work, PUBLIC_API/DATA blast radius, production release, or 10+ files when the repository rules require it. A denied gate yields denied; no side effect is attempted after refusal.

Scale

3-18 specialists; relative cost Low-High by classified scope.

Range derivation: small 3-5; medium 5-9; epic 8-18; global floor/ceiling = min(3,5,8) / max(5,9,18) = 3-18.

Delivery Report

Emit the base ## Nexus Execution Report plus a named Delivery Report:

## Delivery Report
- Target: <frozen target>
- Scope: <small|medium|epic> — <evidence>
- Shipped: <artifacts and behavior>
- Acceptance: <AC-ID → PASS|FAIL|UNVERIFIED + evidence>
- Verification: <commands, tests, review oracles>
- Release: <released|release-ready|not-requested|blocked>
- Exit reason: <canonical exit reason>
- Residual Ledger:
  - RES-<n>: <typed residual> → <owner/recipe> — <why it remains>

Done means the requested product slice works at the frozen acceptance bar, verification evidence is attached, no in-scope placeholder/TODO residue remains, and each residual has an owner and route. Implementation residuals route to feature or the relevant specialist; verification gaps route to Radar/Attest; release gaps route to Guardian/Launch.

Boundaries and Decision Tree

  • One bounded capability in an existing product → feature.
  • A product/MVP build needing a chain that grows with scope → deliver.
  • A high-investment discovery-to-ship run with a fixed deep quality loop → apex.
  • A reusable team/work plan already exists → enact.

Failure Modes Prevented

Failure mode Prevention
Fixed heavyweight chain for a small build Scope classification and minimum-chain selection
Coding before target/AC freeze CLASSIFY and SHAPE exits
Orchestrator absorbing domain work Explicit specialist ownership per phase
False completion after partial progress Independent VERIFY phase and typed Residual Ledger
Retry loops that burn time without new evidence Bounded anti-stall recovery ladder
Release side effects without authority Ask First gate and release parameter

Shared Protocols and References

  • reference/autonomy-quality-protocol.md — completion integrity, producer≠verifier, residual discipline.
  • _common/HANDOFF.md — structured phase handoffs.
  • _common/PARALLEL.md — epic hub-spoke branch ownership and merge.
  • reference/delivery-decision-matrix.md — scope and agent-selection evidence.
  • reference/delivery-exit-criteria-validation.md — phase exit validation.
  • reference/delivery-momentum-system.md — momentum signals without lowering the bar.
  • reference/delivery-product-lifecycle.md — lifecycle phase detail loaded only when the selected scope needs it.

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