Designs agent teams and generates their skills for domain-specific automation harnesses.
npx skills add https://github.com/bam-bam-2/solo-skills --skill harness-bam-bam-2
Please help me install this Agent Skill. Skill: harness Source: https://github.com/bam-bam-2/solo-skills/tree/main/skills/harness Command: npx skills add https://github.com/bam-bam-2/solo-skills --skill harness-bam-bam-2
.claude/agents/ and skill files in .claude/skills/ with pushy trigger descriptions, progressive disclosure, and team communication protocols. - Orchestration & Evolution: Builds an orchestrator skill with data-flow protocols, error handling, and test scenarios, registers a trigger pointer in CLAUDE.md, and maintains a change log so the harness keeps evolving with feedback. - Use Case: Ask to set up a harness for a content-pipeline project, and it audits the repo, defines researcher/writer/reviewer agents, generates their skills, builds an orchestrator, and validates triggers with should-trigger and near-miss queries. ## Quick Start Ask the agent to configure a harness for your project, for example by saying "이 프로젝트에 하네스 구성해줘" (set up a harness for this project).High-intent search queries and answers about installing and using this skill.
Request harness configuration for your domain and the skill audits existing agents and skills, analyzes the domain, then generates agent definitions in .claude/agents/, skills in .claude/skills/, and an orchestrator that coordinates the team.
Agent teams use TeamCreate, SendMessage, and shared task lists so members communicate directly and self-coordinate, making them the default for two or more collaborating agents. Sub-agents are invoked via the Agent tool and only return results to the main agent, suiting isolated single tasks.
Each skill's description acts as its only trigger mechanism, so it is written aggressively with concrete trigger situations. Validation uses 8-10 should-trigger queries plus 8-10 near-miss should-not-trigger queries to check boundary accuracy and conflicts with existing skills.
Yes. A Phase 0 audit detects the current state and branches into extension or maintenance modes, applying only the needed phases per a change-type matrix, then records every modification in the CLAUDE.md change history table.
Six patterns are supported: pipeline, fan-out/fan-in, expert pool, producer-reviewer, supervisor, and hierarchical delegation. Complex real-world setups typically combine patterns, such as parallel collection followed by consensus-based integration.
Skip the team mode when only one agent is needed or when agents require no intercommunication and only return results, since team coordination adds token overhead. Single isolated tasks are better served by a lightweight sub-agent call.
nnexaiAnalyze codebase domains and generate planning documents in .planning/codebase/.
first-flukeOrchestrate a 5-phase, 17-step development workflow with phase-gate loops.
HerxinsasaAutomate AI skill creation, evaluation, and iterative improvement workflows.
JoernStoehlerCreate custom subagents for Task delegation using a YAML file format.