multi-agent-testing skill
Validates multi-agent systems by analyzing delegation, coordination, shared state, conflicts, ownership, termination, and traceability.
- GitHub stars
- 306
- Files
- 11
- Last commit
- Nov 10, 2026
- License
- PolyForm-Noncommercial-1.0.0
- Allowed tools
- —
What does the multi-agent-testing skill do?
multi-agent-testing validates multi-agent systems by analyzing delegation, coordination, shared state, conflicts, ownership, termination, and traceability.
Summary generated from the skill's SKILL.md.
About the multi-agent-testing skill
Validating multi-agent systems by analyzing delegation, coordination, and shared state is what this skill enables for developers building complex LLM or RAG architectures. It provides a structured way to review agent plans and results to produce actionable improvements based on evidence. An agent should load this skill when it needs to perform an audit of coordination protocols, conflict handling, or ownership boundaries within a multi-agent environment.
The skill relies on specific execution instructions found in prompts/multi-agent-testing.md and can use domain-decision-example.md to calibrate outcomes. It produces findings organized by domain risk, evidence state, and priority, which can be delivered in Markdown, CSV, JSON, or ticket formats. It distinguishes itself by strictly separating confirmed facts from inferences and ensuring that static designs or dry runs are never mistaken for real execution evidence. It focuses on tracing correlation IDs, message ownership, and termination rules to ensure system reliability.
What an agent can do with it
- 1Audit agent coordination protocolsA developer can use this skill to review how multiple agents manage shared state and resolve conflicts during a task.
- 2Improve LLM orchestration plansAn AI client can review an existing agent plan to identify gaps in delegation or traceability before execution.
- 3Generate structured testing reportsDevelopers can request findings in specific formats like JSON or CSV to integrate testing results into their existing workflows.
- 4Analyze partial system failuresThe skill helps in reconciling unfinished effects and ensuring that child agent failures do not silently discard tasks.
Written from the skill's SKILL.md. Every use case is checked against a quote from it.
How to install the multi-agent-testing skill
Each command copies the skill folder into the client's skills directory. Restart the client or start a new session to load it.Install the multi-agent-testing skill in Claude Code
~/.claude/skills/multi-agent-testing/SKILL.md
$ npx degit naodeng/awesome-qa-skills/skills/en/testing-types/multi-agent-testing ~/.claude/skills/multi-agent-testingInstall the multi-agent-testing skill in Claude Code (project)
.claude/skills/multi-agent-testing/SKILL.md
$ npx degit naodeng/awesome-qa-skills/skills/en/testing-types/multi-agent-testing .claude/skills/multi-agent-testingInstall the multi-agent-testing skill in Codex
~/.agents/skills/multi-agent-testing/SKILL.md
$ npx degit naodeng/awesome-qa-skills/skills/en/testing-types/multi-agent-testing ~/.agents/skills/multi-agent-testingInstall the multi-agent-testing skill in Gemini CLI
~/.gemini/skills/multi-agent-testing/SKILL.md
$ npx degit naodeng/awesome-qa-skills/skills/en/testing-types/multi-agent-testing ~/.gemini/skills/multi-agent-testingInstall the multi-agent-testing skill in Claude.ai
npx degit naodeng/awesome-qa-skills/skills/en/testing-types/multi-agent-testing multi-agent-testing && zip -r multi-agent-testing.zip multi-agent-testing # claude.ai → Customize → Skills → + → Create skill → Upload a skill → multi-agent-testing.zip
Automated checks
Ran Nov 15, 2026Full report →Not verified yet. A skill needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablenaodeng/awesome-qa-skills/skills/en/testing-types/multi-agent-testing is public and reachablegithub.com/naodeng/awesome-qa-skills/tree/HEAD/skills/en/tes ↗Pass
- Open source license (OSI)PolyForm-Noncommercial-1.0.0 is not OSI-approvedspdx.org/licenses/PolyForm-Noncommercial-1.0.0.html ↗Finding
- Commit or release in the last yearLast commit 2026-11-10github.com/naodeng/awesome-qa-skills/commits ↗Pass
- No download-and-run or credential theft in scriptsNo scripts in the skill folderPass
- No hidden instructions in SKILL.mdNo override phrases, safeguard switches, credential requests or hidden charactersgithub.com/naodeng/awesome-qa-skills/blob/HEAD/skills/en/tes ↗Pass
- Pre-approved tools are scopedNo pre-approved tools; the agent asks before using onegithub.com/naodeng/awesome-qa-skills/blob/HEAD/skills/en/tes ↗Pass
- SKILL.md follows the Agent Skills formatValid name and description; says when to use it; 80 linesgithub.com/naodeng/awesome-qa-skills/blob/HEAD/skills/en/tes ↗Pass
SKILL.md
View on GitHubMulti-Agent Testing
When to Use
- Use this skill when you need evidence-bounded analysis, design, or validation preparation for delegation, coordination, shared state, conflicts, ownership, termination, and traceability.
- Use it to review an Agent, RAG, or LLM plan, result, or evidence set and produce actionable improvements.
- Use it when context is incomplete but a bounded first pass with assumptions, gaps, and human-decision boundaries is still useful.
Output Format Options
- Default to Markdown organized by domain risk, evidence state, priority, and boundary.
- When the user requests tables, CSV, JSON, or ticket fields, preserve the same finding fields, evidence, and decision boundaries.
- Before machine consumption, confirm the schema, enums, required fields, and evidence sources.
How to Use
- Read and follow
prompts/multi-agent-testing.md, including its input audit, domain coverage, and output order. - Extract scope, environment, version, time window, constraints, success criteria, and available evidence, with attention to delegation, coordination protocol, shared state, conflict handling, termination and traceability.
- Separate confirmed facts, evidence-backed inferences, candidate recommendations, and Human decisions before ranking by risk and evidence strength.
- Turn high-risk items into preconditions, steps, expected behavior or decision criteria, required evidence, and a validation method.
- When input is incomplete, deliver a bounded first pass, state unsupported conclusions, and never present static material as execution evidence.
Reference Files
-
When calibrating domain outcomes, valid controls, or evidence gaps, read references/domain-decision-example.md for sanitized input, per-item oracles, and reproducible calculations.
-
Always read
prompts/multi-agent-testing.md; it is the complete execution specification for this skill. -
For evaluation, read
evals/eval.yamland the matching cases underevals/cases/. -
Load
references/,examples/,scripts/, oroutput-formats.mdonly when those directories exist and the task needs them.
Domain Execution Essentials
- Trace task/message/correlation IDs, sender/receiver, parent delegation, scope, owner, shared-state version, and ACK using actual acceptance/rejection/termination rules.
- Claim/commit follows the project's atomic ownership or fencing protocol. Stale/duplicate messages must not regain ownership or duplicate writes; never assume exactly-once delivery.
- Reconcile completed, unfinished, and indeterminate effects on partial failures before reassignment. One child's failure/timeout cannot silently discard unresolved tasks.
- Check stop propagation, ACKs, terminal states, and remaining owners; expired leases cannot commit. Include valid single claims/handoffs so message redelivery alone is not a false defect.
Core Constraints
- Keep the analysis focused on delegation, coordination, shared state, conflicts, ownership, termination, and traceability; do not replace business owners or Human risk acceptance, exception approval, or safety decisions.
- Never invent system behavior, fields, model outputs, data, thresholds, root causes, execution records, or pass claims.
- Static design, plans, file presence, or a dry run retain their evidence state and cannot become proof of real execution.
- When evidence is insufficient, use pending confirmation, blocked, unassessed, or NOT_SCORED and give the smallest validation method.
- For user data, production, or safety work, use least privilege, masked data, mocks, dry runs, or isolation.
Delivery Checklist
- Covered delegation, coordination protocol, shared state, conflict handling, termination and traceability, with source, evidence state, and validation method for each.
- Separated facts, inferences, candidate recommendations, gaps, and Human decisions.
- Gave high-risk items P0/P1/P2/P3 or an equivalent priority, owner role, and close condition.
- Did not turn plans, static checks, or dry runs into test execution, all-passed, or safety-approved claims.
- Stated residual risk, stop/escalation conditions, and next actions.
Common Pitfalls
- Listing checks without triggers, expected concerns, owner roles, close conditions, and evidence.
- Treating adjacent tests or model tools as a complete multi-agent coordination judgment.
- Using unexplained numbers for false precision or writing correlation as causation.
- Refusing incomplete input, or pretending that incomplete evidence is conclusive.
Best Practices
- Start with paths most likely to cause user harm, business loss, or decision blockage.
- Use the smallest verifiable experiment to reduce uncertainty and record conditions, versions, sources, and evidence.
- Make the Skill independently installable, executable, and reviewable by another engineer.
Files in this skill
11multi-agent-testing skill questions
How do I install the multi-agent-testing skill in Claude Code?
Run npx degit naodeng/awesome-qa-skills/skills/en/testing-types/multi-agent-testing ~/.claude/skills/multi-agent-testing. Claude Code loads skills from ~/.claude/skills (all projects) or .claude/skills (one project) and uses a skill when its description matches the task.
Which clients can use the multi-agent-testing skill?
Any client that reads the Agent Skills format: Claude Code, Codex, Gemini CLI and claude.ai (upload as a ZIP). Scripts in the skill run only where the client allows code execution.
When should I use the multi-agent testing skill?
Use it for evidence-bounded analysis, design, or validation of delegation, coordination, shared state, conflicts, ownership, termination, and traceability, or to review Agent, RAG, or LLM plans.
What are the core constraints of this skill?
The analysis must focus on coordination and ownership without replacing human decision-making, and it must never invent system behavior or treat static plans as proof of execution.
How should the skill handle incomplete input?
When input is incomplete, provide a bounded first pass, state any unsupported conclusions, and use statuses like pending confirmation or blocked instead of pretending evidence is conclusive.