Convert session corrections and factual errors into guardrail SKILL.md entries.
npx skills add https://github.com/hve4638/hve-cc-marketplace --skill reflect-hve4638
Please help me install this Agent Skill. Skill: reflect Source: https://github.com/hve4638/hve-cc-marketplace/tree/main/common/skills/reflect Command: npx skills add https://github.com/hve4638/hve-cc-marketplace --skill reflect-hve4638
Extract user corrections and factual errors from the session, and turn each item into a child guardrail skill (reflect-<slug>/SKILL.md). In future sessions the children auto-trigger via description matching and prevent the same mistake.
Run reflect after a session to generate child reflect-<slug>/SKILL.md entries from user corrections and factual errors.
High-intent search queries and answers about installing and using this skill.
The reflect skill identifies user corrections and factual errors from a session and converts each item into a child SKILL.md file with a Rule body and a Not a violation section. These child skills then auto-trigger in future sessions via description matching to prevent the same mistakes.
Retrospection in AI safety is the process of extracting points corrected and misknown or outdated facts from a session's context. The reflect skill uses this retrospective thinking to generate future-facing child skills that automatically prevent repeated errors without relying on external lookups.
Run the reflect skill after a session to generate child reflect-<slug>/SKILL.md entries from user corrections and factual errors. Each entry requires a frontmatter with name and description, enforces kebab-case slugs, and is stored under .claude/skills/reflect-<slug>/SKILL.md.
No, the reflect skill follows a self-exploration principle where retrospective extraction relies only on session context. External lookups are avoided, ensuring that guardrail rules are generated strictly from the interactions within the current session.
A limitation is that child skills only carry future-facing instructions and must not include the four internal retrospection slots. Additionally, the generated catalog acts as the single source of truth, requiring proper kebab-case slug formatting and frontmatter to function correctly.
harkersLog learnings, errors, and corrections to standardized markdown files.
sunLeeeExtract metadata from SKILL.md files and update the central skill catalog.
ianyimiEnforce coding standards for the global tech stack in Claude Code.
rodacatoCreate or update AGENTS.md to codify AI agent startup behavior and governance.
