三省吾身 - 深度自我审视与进化系统
v2.0.0三省吾身 - 深度自我审视与进化系统。基于角色理论、系统思维、认知偏差三大框架，通过五阶段闭环（反思→计划→执行→验证→固化）实现真正的自我进化。适用于任何需要深度反思的场景：项目复盘、工作改进、能力提升、习惯养成。
⭐ 0· 891·8 current·8 all-time
by@demo112
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name and description match the included SKILL.md, templates, and examples. The content is about structured self-reflection, root-cause analysis and improvement; nothing requested or included (no env vars, no binaries, no installs) appears unrelated to that purpose.
Instruction Scope
Runtime instructions are purely conversational/workflow guidance (questions to ask, checklists, templates, step-by-step process). They do not instruct reading system files, contacting external endpoints, or accessing credentials. The only slightly open item is mention of 'automatic trigger' (detecting negative-feedback keywords or periodic reminders) which is a behavioral description rather than a code instruction — it does not itself require extra access.
Install Mechanism
No install spec, no code files that execute, and no archive downloads. This instruction-only skill has the lowest install risk.
Credentials
The skill declares no required environment variables, no primary credential, and no config paths. Templates and prompts will require user-provided facts during use, but the skill does not request secrets or unrelated credentials.
Persistence & Privilege
always:false and no special persistence or modifications to other skills are requested. The skill is user-invocable and allows model invocation (the platform default); that is expected for an agent skill and not flagged by itself.
Assessment
This skill is instruction-only and appears internally consistent with its stated goal of guided self-reflection. Before installing, consider: (1) If your agent platform enables autonomous triggers or periodic reminders, review and control the permissions for scheduling/notifications so the skill can't send unsolicited reminders; (2) The skill will prompt you to paste facts, logs, or feedback into the conversation — avoid pasting secrets or private tokens into its prompts; (3) Templates may contain user-provided data when populated (reports, checklists) — treat those outputs as normal chat content and manage storage/sharing accordingly. If you want the skill to run periodic self-checks, confirm where and how those reminders are stored and that no external endpoints will be called.Like a lobster shell, security has layers — review code before you run it.
latestvk972vej282jj6c5m17p7cpbxg581tb3p
License
MIT-0
Free to use, modify, and redistribute. No attribution required.