Agent Harness Architect
v1.0.0Diagnose and harden any long-running agent architecture. Describe your agent setup and get a complete harness optimization plan + auto-generated bridge artif...
⭐ 0· 50·1 current·1 all-time
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 SKILL.md: it diagnoses agent harnesses and generates bridge artifacts (agent-progress.json, session protocol, output gate). It does not request unrelated credentials, binaries, or install steps.
Instruction Scope
SKILL.md is an instruction-only advisor that asks the user to describe their agent, then generates JSON/markdown artifacts and example commands (e.g., git commit, openclaw status). It asks users to paste CLAUDE.md / AGENTS.md for context. This is appropriate for the stated purpose, but watch for: (1) users pasting sensitive files/configs—those could expose secrets; (2) the 'Self-Improvement Protocol' language about '自动写入 capability-radar' is ambiguous about where/what is written — the skill gives no clear persistent storage path, so any automated writing would be an agent/platform action, not something inherent to the skill.
Install Mechanism
No install spec and no code files. Instruction-only skills have minimal install risk because nothing is downloaded or written by the skill itself.
Credentials
The skill requires no environment variables, no credentials, and no config paths. Example commands reference git and platform CLIs only as optional platform-specific advice—these are proportional to the task and documented as examples.
Persistence & Privilege
always:false and no declared persistence. However, the SKILL.md discusses an internal 'Self-Improvement Protocol' that describes auto-writing learning logs and updating the skill/template. Because the skill is instruction-only and contains no install/run-time code, such persistence would be carried out by the agent or platform—confirm whether you want the agent to autonomously persist learnings or modify skill files. This is a policy/behavior choice on the agent, not a hidden capability of the skill itself.
Assessment
This skill is an instruction-only advisor and appears coherent for diagnosing and hardening long-running agents. Before using: (1) avoid pasting secrets, API keys, or private tokens into the conversation or into pasted CLAUDE.md/AGENTS.md files; (2) inspect any generated scripts or git commands before running them—especially commits or automated pushes; (3) clarify where any 'learning logs' or capability-radar entries will be stored if you permit the agent to persist them; (4) if you allow the agent to perform automated actions (commits, file writes), limit its permissions and test on a non-production repo first. If you want higher assurance, ask the skill to only produce artifacts for manual review (no auto-commit or auto-write) and confirm all file paths it will use.Like a lobster shell, security has layers — review code before you run it.
latestvk97ejz7an65a0xnyqdw6vhnk3s83y13a
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
Runtime requirements
🏗️ Clawdis