Memory to Skill Crystallizer
ReviewAudited by ClawScan on May 10, 2026.
Overview
This skill has a coherent purpose, but it can automatically turn memory entries into persistent local skills, which may copy private or untrusted text into future agent behavior without review.
Use only if you are comfortable with the agent reading local memory files and creating persistent local skills. Review every generated skill before use, remove any private or project-specific details, and consider adding an explicit approval and cleanup workflow.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Sensitive or misleading text recorded in memory could be copied into reusable skills and later influence future agent sessions.
The workflow reads local memory files, captures the text after error/blocker labels, and writes that captured memory text directly into generated skill content.
$content = Get-Content $file.FullName -Raw ... $key = $m.Groups[2].Value ... description: Auto-recovery for: $($pattern.Key)
Require a user review/redaction step before creating skills, and avoid copying raw memory text into skill descriptions, triggers, or instructions.
The agent may accumulate new local skills based on past logs, potentially making future behavior harder to predict or audit.
The skill instructs the agent to create persistent local skill entries automatically, which can change the agent's future behavior without a required confirmation gate.
Convert memory lessons into reusable skills automatically ... New-Item -ItemType Directory -Path $skillPath -Force ... $skillContent | Out-File "$skillPath/SKILL.md" ... Reuse forever.
Make generated skills drafts by default, require explicit user approval before registration/use, and provide a clear cleanup or rollback procedure.
A user may trust that no sensitive memory content is copied, even though the artifact does not enforce that guarantee.
The privacy statement is stronger than the workflow supports, because the captured pattern string is reused directly in generated skill descriptions, headings, and triggers.
## Privacy/Safety - No sensitive data in extracted patterns - Pattern names only, no specific content
Replace the blanket privacy claim with accurate limitations and implement concrete filtering for secrets, personal data, and project-specific details.