Skill Creator
v1.0.0Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize a...
⭐ 0· 65·0 current·0 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
The name/description (create/modify/evaluate skills) align with the included artifacts: SKILL.md, evaluator agents (analyzer/comparator/grader), and multiple scripts (run_eval, run_loop, generate_review, package_skill, etc.). There are no unrelated environment variables, external credentials, or surprising binaries required.
Instruction Scope
The SKILL.md instructs the agent to run the bundled evaluation scripts (e.g., eval-viewer/generate_review.py) and to read skill files, transcripts, and workspace outputs to produce analyses and HTML review pages. This is expected for an eval/skill-authoring tool, but those scripts will read arbitrary files in the provided workspace and embed them into a served HTML page — so confidential files in the workspace could be exposed if you run the viewer without checking contents.
Install Mechanism
No install spec; the skill is instruction-first and ships Python scripts. No network downloads or archives are fetched during install. The code uses only standard-library Python modules and bundled files.
Credentials
The package declares no required environment variables, credentials, or config paths. The scripts access the local workspace and rely on standard tools (e.g., lsof if present) but do not request unrelated secrets or cloud credentials.
Persistence & Privilege
always:false and no persistent system-wide changes are requested. However, generate_review.py contains logic to locate and SIGTERM processes listening on a port (via lsof + os.kill) to free the port for its local HTTP server — this can terminate other local services and is somewhat intrusive. The script also serves workspace files over HTTP and auto-saves feedback.json in the workspace.
Scan Findings in Context
[subprocess.run_lsof] expected: The generate_review script uses subprocess.run to call lsof to find processes on a port before starting its server. This is consistent with trying to open a local preview server but is intrusive because it may identify and terminate unrelated processes.
[os.kill_sigterm] expected: The script sends SIGTERM to PIDs occupying the chosen port to free it. For an eval viewer that wants to bind a local port this is plausible, but it can disrupt other services if run without care.
[read_and_embed_files_base64] expected: The viewer embeds arbitrary output files (including binaries) as base64/data URIs in the generated HTML. This is expected for a self-contained review UI, but it means any file in the workspace (including secrets) could be exposed in the generated page.
[webbrowser_open] expected: The script uses webbrowser.open to open the generated page in the system browser. This is typical for local preview tools.
[http_server_serving_workspace] expected: A tiny stdlib HTTP server is used to serve the embedded review page. Expected for an eval viewer; review access controls and the workspace contents before running.
Assessment
This skill is coherent for building and evaluating skills and includes useful scripts to run local evals and generate a self-contained review page. Before running anything or pointing the tool at a workspace: 1) inspect the workspace directory — the viewer will read and embed any file it finds (so remove or move sensitive files first); 2) be aware generate_review.py tries to free a port by finding and SIGTERM-ing processes (avoid running it on machines hosting important services); 3) run the scripts in an isolated environment or VM if you are unsure about the workspace contents; 4) on Windows or systems without lsof the port-kill code may behave differently — review/modify the script if needed; 5) if you plan to allow the agent to run the skill autonomously, limit access to workspaces that don't contain credentials or private data. If you want, I can point out the exact lines in the scripts that perform port-killing, file embedding, and HTTP serving so you can review or modify them before execution.Like a lobster shell, security has layers — review code before you run it.
agentvk979262mp6rq29jnc1j473y1fd83h38nchinesevk979262mp6rq29jnc1j473y1fd83h38nlatestvk979262mp6rq29jnc1j473y1fd83h38ntoolvk979262mp6rq29jnc1j473y1fd83h38n
License
MIT-0
Free to use, modify, and redistribute. No attribution required.