creation-feasibility-gate
21
总安装量
21
周安装量
#17354
全站排名
安装命令
npx skills add https://github.com/oimiragieo/agent-studio --skill creation-feasibility-gate
Agent 安装分布
github-copilot
21
gemini-cli
21
cursor
21
amp
20
codex
20
kimi-cli
20
Skill 文档
Creation Feasibility Gate
Overview
Run a fast preflight feasibility check before creating a new agent/skill/workflow/hook/template/schema. This prevents low-value or impossible creator runs.
When to Use
- Phase 0.5 dynamic creation flow
- User asks for net-new capability
- Reflection/evolution recommends artifact creation
Iron Laws
- NEVER create artifacts inside this skill â return PASS/WARN/BLOCK with evidence only; all actual creation happens in the appropriate creator skill downstream.
- ALWAYS run the existence/duplication check first â never proceed toward PASS if a functionally identical artifact already exists in any catalog or registry.
- ALWAYS include concrete file-level evidence for every decision â a bare PASS or BLOCK without referencing specific paths or catalog entries is a spec violation.
- NEVER let WARN silently become PASS â every WARN must list exact caveats that the calling agent must acknowledge before creation proceeds.
- ALWAYS resolve BLOCK status with actionable next steps and recommended target agents â a BLOCK without remediation tasks is an incomplete gate decision.
Anti-Patterns
| Anti-Pattern | Why It Fails | Correct Approach |
|---|---|---|
| Returning PASS without running the duplication check | Creates duplicate artifacts that split agent traffic and inflate catalogs | Always query catalog + registry + filesystem before PASS |
| Returning BLOCK without remediation tasks | Calling agent stalls with no path forward | Include nextActions with specific agents/skills to unblock |
| Skipping the security/creator boundary check | Creator paths may be blocked by governance hooks; silently bypassing them causes runtime failures | Always verify creator skill chain is reachable before PASS |
| Treating WARN as informational only | WARN caveats are not surfaced to the user; creation proceeds with unresolved risks | WARN must be acknowledged explicitly by the caller in its task metadata |
| Running creation steps inside the gate skill | Violates separation of concerns; gate outputs can’t be validated independently | Gate outputs only the decision JSON; delegate creation to creator skills |
Workflow
Step 1: Resolve Target
- Identify proposed artifact type and name
- Identify expected runtime/tool dependencies
- Identify expected owner agents
Step 2: Preflight Checks
Run these checks with concrete evidence:
- Existence/duplication check
- Catalogs + registry + artifact paths
- Stack compatibility check
- Required tooling/runtime present in current project conventions
- Integration readiness check
- Can it be routed/discovered/assigned after creation?
- Security/creator boundary check
- Ensure creator path and governance can be satisfied
Step 3: Decision
Return one status:
PASS: creation is feasible nowWARN: feasible with clear caveatsBLOCK: not feasible; must resolve blockers first
Use this output shape:
{
"status": "PASS|WARN|BLOCK",
"artifactType": "agent|skill|workflow|hook|template|schema",
"artifactName": "example-name",
"evidence": ["..."],
"blockers": [],
"nextActions": ["..."]
}
Output Protocol
If BLOCK, include concrete remediation tasks and recommended target agents.
If PASS or WARN, include exact creator skill chain to run next.
Memory Protocol
Record feasibility patterns and recurring blockers to .claude/context/memory/learnings.md.