Prompt Design Tuning
v1.0.0Collaboratively design, evaluate, iterate on, and recommend a final launch candidate for a target prompt under the principle of “human-gated, agent-executed”...
⭐ 1· 65·0 current·0 all-time
byyuanjie@abysscat-yj
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
The name and description (prompt design & tuning) match the SKILL.md: it describes a human-gated, agent-executed workflow for drafting prompts, judge prompts, evaluation plans, and producing script skeletons and reports. No unrelated capabilities, binaries, or credentials are requested.
Instruction Scope
The instructions stay within prompt-design and evaluation workflow boundaries: they ask the agent to gather task specs, produce drafts, prepare evaluation plans, and (if a runnable environment is provided) run batch generation and evaluation. Human approval is explicitly required for high-cost loops. The SKILL.md does not instruct the agent to read arbitrary system files or to exfiltrate data. It does assume that a model invocation method (curl/SDK/API) may be supplied by the user before execution — which is reasonable for an execution-capable mode.
Install Mechanism
There is no install spec and no code files — this is instruction-only, so nothing is written to disk by an installer. Recommended deliverables include scripts and docs the agent may draft, but no automatic downloads or installs are specified.
Credentials
The skill declares no required environment variables, credentials, or config paths. The instructions reference using a provided invocation method (which would require credentials supplied by the user in Execution Mode), but the skill itself does not request or demand any secrets — this is proportionate to its purpose.
Persistence & Privilege
always:false (default) and autonomous invocation is not disabled (also normal). The skill does not request permanent presence or system-wide configuration changes. Human gates are required for costly operations, limiting autonomous high-impact actions.
Assessment
This skill is an instruction-only workflow for designing and iterating prompts and appears coherent. Before using it in Execution Mode, remember: (1) only provide API keys or credentials if you expect the agent to actually call models — do not paste secrets unless you trust the human-approved run; (2) review any scripts or code the agent generates before executing them in your environment; (3) approve high-cost evaluation loops explicitly as the skill requests; and (4) if you need the agent to access private data or systems, confirm that it’s necessary and scope the data tightly. If you want extra assurance, ask the skill to produce a dry-run plan (Design-Only Mode) before giving it permission to run evaluations.Like a lobster shell, security has layers — review code before you run it.
latestvk9790ftxck1bkxm0yp55gnrp0583pebb
License
MIT-0
Free to use, modify, and redistribute. No attribution required.