Improvement Orchestrator
v1.1.1当需要一键跑完「生成→评分→评估→执行→门禁」全流程、失败后自动重试、或批量改进多个 skill 时使用。不用于单独评估 skill 质量（用 improvement-learner）或手动打分（用 improvement-discriminator）。
⭐ 0· 59·1 current·1 all-time
by_silhouette@lanyasheng
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name/description match the implementation: the skill coordinates proposer→critic→evaluator→executor→gate. It requires no unusual environment variables or external credentials. The code expects a repo-style layout (other improvement-* skills and lib.* helpers) which is consistent with an orchestrator that dispatches to sibling scripts.
Instruction Scope
SKILL.md and scripts explicitly instruct the agent to run local Python scripts that will read and write filesystem artifacts (candidate/ranking/execution/receipt/state files) and may apply low-risk changes (append_markdown_section) to target skills. This is within scope, but the runtime will modify files under the provided target path and create backups/traces — users should ensure the target path is correct and not a protected or infra-critical path.
Install Mechanism
There is no install spec (instruction-only plus local scripts). Nothing is downloaded or written by an installer. Risk from install-time network downloads is not present.
Credentials
No environment variables, credentials, or external tokens are requested. The orchestrator operates on local file paths and local sibling scripts; it does not request unrelated secrets or remote API keys.
Persistence & Privilege
The skill is user-invocable and can be invoked autonomously by the agent (normal default). It does create persistent local artifacts (state, backups, traces) and modifies target files when the executor accepts a candidate. It does not set always:true. Because it can apply changes to local files, run it with appropriate target paths and review produced artifacts/backups when used.
Assessment
This orchestrator is coherent with its documentation: it runs local improvement scripts and may modify files under the --target path, but only supports conservative, low-risk changes (Markdown appends) in this first version and creates backups and audit artifacts. Before installing or running: (1) point --target at a safe test copy (not infra or gateway configs), (2) inspect the generated state/backups directory after a dry run, (3) prefer using --state-root in an isolated temp directory, and (4) verify the full repo layout (other improvement-* scripts and lib.*) exists in the environment so the orchestrator calls the intended local code. If you want stronger safety, run it without --auto first to inspect each step's outputs or run in a sandbox repository.Like a lobster shell, security has layers — review code before you run it.
latestvk9740qwsf4cq0abnex2ya714ah84b2nz
License
MIT-0
Free to use, modify, and redistribute. No attribution required.