nexus_skill
- Shell
8
GitHub Stars
1
Bundled Files
2 months ago
Catalog Refreshed
4 months ago
First Indexed
Readme & install
Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.
Installation
Preview and clipboard use veilstrat where the catalogue uses aiagentskills.
npx veilstrat add skill simota/agent-skills --skill nexus- SKILL.md14.5 KB
Overview
This skill is an orchestrator that coordinates specialized AI agents to solve software development tasks end-to-end. It decomposes requests, selects a minimal agent chain, and runs agents in AUTORUN modes or emits guided prompts for manual steps. Nexus enforces hub-and-spoke handoffs, guardrails, and automatic recovery to produce reliable, auditable outcomes.
How this skill works
Nexus classifies the request, scores context confidence, and designs a minimal chain using predefined orchestration patterns (sequential, parallel, conditional, recovery, verification, escalation). In AUTORUN/AUTORUN_FULL modes it executes agents internally through phased pipelines (plan, select, execute, aggregate, verify, deliver) with layered guardrails and error-recovery logic. In GUIDED/INTERACTIVE modes it emits routing prompts and NEXUS_HANDOFF formatted artifacts for manual or downstream agent invocation.
When to use it
- End-to-end bug fixes that require investigation, implementation, and testing.
- Complex changes spanning frontend, backend, and infra where parallel branches help.
- Security audits or fixes that need coordinated Sentinel and Builder workflows.
- Investigations (regressions, root cause) that require Lens/Scout/Rewind analysis.
- Automated runs for repeatable tasks where immediate execution is desired (AUTORUN_FULL).
Best practices
- Always provide a 1–3 line goal and acceptance criteria before orchestration.
- Prefer minimum viable chains; add agents only when scope or failures demand them.
- Use NEXUS_AUTORUN for simple tasks and GUIDED for any destructive or multi-step changes.
- Log orchestration insights to .agents/nexus.md and append actions to .agents/PROJECT.md.
- Declare technical_domain and scope_indicators to improve routing accuracy.
Example use cases
- Fix a failing test suite: Scout → Builder → Radar with auto-retry and rollback on major failures.
- Implement a new API endpoint: Gateway → Builder → Radar, plus Quill for docs and Schema for DB.
- Security patch rollout: Sentinel → Builder → Radar with Probe for dynamic checks and L4 abort on critical findings.
- Refactor a module: Zen → Radar with Atlas added for architectural scope and Grove for structure.
- Investigate regression: Rewind → Lens → Scout to identify commit, reproduce, and propose a fix.
FAQ
Nexus computes a context confidence score and compares chain/approach thresholds. AUTORUN_FULL auto-executes when confidence and decision metrics meet safety thresholds; GUIDED is used for complex or risky flows.
What happens on test failures or security alerts during execution?
Guardrails trigger layered recovery: auto-retry for lint/syntax, inject Builder for moderate failures, rollback and Sherpa for major failures, and abort with L4 escalation for critical security issues.