Agent Introspection
v1.0.0Agent自省机制。在执行任务的关键节点触发结构化自省，防止盲目执行、惯性偏离、过度自信。Use when: 任务开始前、方案选择时、执行受阻时、完成交付前。Triggers on: 自省, 反思, 审视, 回顾, introspect, reflect, 我做得对吗, 检查思路, 方向对吗。
⭐ 0· 65·1 current·1 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 (agent introspection/self-reflection) line up with the content of SKILL.md; the skill declares no binaries, env vars, or installs and does not ask for capabilities unrelated to introspection.
Instruction Scope
SKILL.md stays within introspection scope (structured prompts, checklists, and a required output format). One area to note: the timed/self-scheduled rules say to "read past self-reflection files" but the skill does not declare any file paths or explain where those files live; this is reasonable for an introspection skill but could lead to ambiguous file access if the agent is given broad filesystem permissions. Also the skill emphasizes doing actions (not just writing plans) — which is a behavioral requirement, not a technical one, but it may cause the agent to invoke other capabilities (other skills or side effects) when integrated.
Install Mechanism
Instruction-only skill with no install spec and no code files — lowest install risk (nothing is written or downloaded by the skill itself).
Credentials
No environment variables, credentials, or config paths are requested; this is proportionate to an introspection/operational-guidance skill.
Persistence & Privilege
always:false and default model invocation settings are used. The skill does not request persistent presence or modifications to other skills or system-wide settings.
Assessment
This skill is internally coherent and low-risk: it is a set of instructions/prompts for structured self-reflection and requests no credentials or installs. Before installing, confirm two operational details with whoever will run the agent: (1) if you plan to run timed/scheduled introspection, where are past self-reflection files stored and does the agent have permission only to those locations (avoid giving broad filesystem access), and (2) because the skill emphasizes taking immediate "actions" rather than just plans, review what actions the agent is permitted to perform (invoking other skills, modifying resources) so the introspection outputs can't cause unexpected side effects. If you want stricter limits, require the skill to specify allowed file paths and restrict agent privileges in a sandboxed test run first.Like a lobster shell, security has layers — review code before you run it.
latestvk97f2hznxxqxqjcr467tcsaecd83k94v
License
MIT-0
Free to use, modify, and redistribute. No attribution required.