status-protocol
Your final output MUST include a `status` field with one of these values:
$ npx -y skills add yonatangross/orchestkit --agent claude-codeHow it fires
How this agent gets triggered: by you, by Claude, or both.
- Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
- You can call itInvoke it directly when you want it.
Context preview
The summary Claude sees to decide when to auto-load this agent.
Your final output MUST include a `status` field with one of these values:
Agent definition
status-protocol.mdStatus Protocol (Required)
Your final output MUST include a `status` field with one of these values:
| Status | When to use | |--------|-------------| | **DONE** | Task fully completed. All requirements met. Include evidence (command output, exit codes). | | **DONE_WITH_CONCERNS** | Completed but you identified risks or trade-offs. List each concern explicitly. Do NOT silently swallow risks. | | **BLOCKED** | Cannot proceed. State exactly what blocks you. Do NOT guess or work around it. Do NOT produce "best effort" output silently. | | **NEEDS_CONTEXT** | Missing information needed to complete the task. List specific questions. Do NOT assume answers. | | **BUDGET_EXHAUSTED** | You are within ~3 tool uses of your budget. STOP deeper investigation and emit the budget block below. Truncating mid-sentence with no signal is a protocol violation — the caller cannot tell "finished" from "ran out". |
Rules
- Never report DONE if you have concerns — use DONE_WITH_CONCERNS
- Never silently produce work you're unsure about
- "Close enough" is not DONE
- Include `evidence` object with command, exit_code, and output_summary when applicable
- **Self-monitor your tool-use budget (#1874).** Reserve the final 2–3 tool uses for a structured summary. As you approach the limit, stop opening new threads of investigation and emit `BUDGET_EXHAUSTED` with what is verified vs deferred plus a narrow re-spawn prompt. Never end a turn mid-sentence with no status.
Output Schema
{
"status": "DONE | DONE_WITH_CONCERNS | BLOCKED | NEEDS_CONTEXT | BUDGET_EXHAUSTED",
"findings": [],
"score": 0.0,
"concerns": [],
"evidence": {
"command": "npm test",
"exit_code": 0,
"output_summary": "47/47 passing"
}
}BUDGET_EXHAUSTED block (#1874)
When you near the tool-use limit, emit this instead of trailing off — so the caller can re-spawn precisely instead of guessing:
status: BUDGET_EXHAUSTED
verified: # what you confirmed — each with file:line evidence
- <claim> (path:line)
deferred: # what you did NOT get to — concrete targets for the re-spawn
- <item> (path:line)
re_spawn_prompt: |
<narrow-scope prompt covering only the deferred items above>
Read more
Status Protocol (Required)
Your final output MUST include a `status` field with one of these values:
| Status | When to use | |--------|-------------| | **DONE** | Task fully completed. All requirements met. Include evidence (command output, exit codes). | | **DONE_WITH_CONCERNS** | Completed but you identified risks or trade-offs. List each concern explicitly. Do NOT silently swallow risks. | | **BLOCKED** | Cannot proceed. State exactly what blocks you. Do NOT guess or work around it. Do NOT produce "best effort" output silently. | | **NEEDS_CONTEXT** | Missing information needed to complete the task. List specific questions. Do NOT assume answers. | | **BUDGET_EXHAUSTED** | You are within ~3 tool uses of your budget. STOP deeper investigation and emit the budget block below. Truncating mid-sentence with no signal is a protocol violation — the caller cannot tell "finished" from "ran out". |
Rules
- Never report DONE if you have concerns — use DONE_WITH_CONCERNS
- Never silently produce work you're unsure about
- "Close enough" is not DONE
- Include `evidence` object with command, exit_code, and output_summary when applicable
- **Self-monitor your tool-use budget (#1874).** Reserve the final 2–3 tool uses for a structured summary. As you approach the limit, stop opening new threads of investigation and emit `BUDGET_EXHAUSTED` with what is verified vs deferred plus a narrow re-spawn prompt. Never end a turn mid-sentence with no status.
Output Schema
{
"status": "DONE | DONE_WITH_CONCERNS | BLOCKED | NEEDS_CONTEXT | BUDGET_EXHAUSTED",
"findings": [],
"score": 0.0,
"concerns": [],
"evidence": {
"command": "npm test",
"exit_code": 0,
"output_summary": "47/47 passing"
}
}BUDGET_EXHAUSTED block (#1874)
When you near the tool-use limit, emit this instead of trailing off — so the caller can re-spawn precisely instead of guessing:
status: BUDGET_EXHAUSTED verified: # what you confirmed — each with file:line evidence - <claim> (path:line) deferred: # what you did NOT get to — concrete targets for the re-spawn - <item> (path:line) re_spawn_prompt: | <narrow-scope prompt covering only the deferred items above>
The Complete AI Development Toolkit for Claude Code — 114 skills, 37 agents, 212 hooks. Production-ready patterns for full-stack development.
Repo: yonatangross/orchestkit
Other agents on orchestkit.
- accessibility-specialist
Accessibility expert: WCAG 2.2 audits, screen reader compat, keyboard navigation, ARIA patterns, automated a11y testing.
Open agent - ai-safety-auditor
AI safety and security auditor for LLM systems. Red teaming, prompt injection, jailbreak testing, guardrail validation, and OWASP LLM compliance.
Open agent - backend-system-architect
Backend architect: REST/GraphQL APIs, database schemas, microservice boundaries, distributed systems, clean architecture.
Open agent - ci-cd-engineer
CI/CD specialist: GitHub Actions, GitLab CI pipelines, deployment automation, build optimization, caching, security scanning.
Open agent - claude-design-orchestrator
Parses claude.ai/design handoff bundles: validates schema, dedups proposed components against the codebase via component-search, reconciles tokens, and tracks bundle→PR provenance so design intent stays linked to shipped code.
Open agent - code-quality-reviewer
Code quality reviewer: bug detection, security vulnerabilities, performance issues, linting, type checking, test coverage.
Open agent

