Autoloop Controller
v1.0.0When continuous automated improvement of a Skill is needed. Wraps improvement-orchestrator in a persistent loop with convergence detection (plateau/oscillati...
⭐ 0· 35·0 current·0 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: autoloop.py runs repeated improvement-orchestrator subprocesses, persists state, detects plateau/oscillation, and enforces cost/iteration caps. No unrelated binaries or secrets are requested.
Instruction Scope
SKILL.md and autoloop.py limit actions to launching the orchestrator as a subprocess, reading/writing local state/log files, computing scores, and writing handoff artifacts. There are no instructions to read unrelated system paths or to transmit state to external endpoints. The code does modify sys.path to locate sibling modules and expects a repo layout that includes lib.common and the improvement-orchestrator skill.
Install Mechanism
No install spec provided (instruction-only). Files are included in the skill bundle; nothing is downloaded from external URLs and no archive extraction/install steps are present.
Credentials
The skill itself declares no required environment variables or credentials, which is appropriate. Note: it invokes improvement-orchestrator (a separate skill) as a subprocess — that orchestrator or the LLM/evaluator it runs may require API keys or other credentials outside this bundle. Users should review the orchestrator/learner skills for any credential needs and ensure autoloop is run with a non-sensitive state-root path.
Persistence & Privilege
No always:true privilege; the skill writes state and logs to the user-provided state-root (expected behavior). It does not modify other skills' configs or global agent settings beyond creating files under the specified state directory.
Assessment
This skill appears to do what it claims: repeatedly run the improvement-orchestrator, persist state, and stop on convergence/cost limits. Before installing or running: 1) Verify the improvement-orchestrator and improvement-learner skills (invoked by this loop) and any LLM/evaluator they call — those components may require API keys or network access. 2) Provide a dedicated state-root (not a sensitive system directory) because autoloop writes logs, JSONL iteration logs, handoffs, and a persistent autoloop_state.json. 3) Use --dry-run to test loop logic and verify outputs, and set conservative --max-cost before running continuous/overnight jobs. 4) If you will schedule this via cron, ensure the cron user's environment has necessary credentials and that you accept the cost implications of automated runs.Like a lobster shell, security has layers — review code before you run it.
latestvk97f6avmnghpmww60a2san6rzn8491a1
License
MIT-0
Free to use, modify, and redistribute. No attribution required.