Learns when to stop and review. Self-critiques before showing you, fewer revision rounds.
Security Analysis
high confidenceThis is an instruction-only local “self-reflection” helper that is consistent with its description: it performs local pre-delivery checks, logs reflections, and stores state under ~/reflection/ without requesting credentials or installing remote code.
The skill's name/description (self-critique, fewer revision rounds) matches the files and runtime instructions. All declared metadata and required resources (none) align with a local reflection/memory assistant. There are no unrelated requirements (no cloud keys, no external binaries).
SKILL.md and supporting files instruct the agent to pause before delivering output, evaluate 7 dimensions, and log reflections to ~/reflection/ (memory.md, reflections.md, patterns.md). This stays within the stated purpose. Note: the instructions explicitly tell the agent to read/write files in the user's home directory and to capture user corrections verbatim — those logs may contain sensitive user-provided content unless sanitized.
No install spec, no code files, no downloads. Instruction-only skill — lowest installation risk; nothing will be written to disk except the user-visible ~/reflection/ state the skill asks to create.
The skill requests no environment variables or credentials, which is proportionate. However, because it logs user corrections and lessons to local files, those files could contain sensitive or private data. The skill does not mention any sanitization or redaction steps, so stored content could be a privacy concern if not managed.
The skill persists state in ~/reflection/ (creates files, archives, and updates a MEMORY.md). It does not have always:true and does not request elevated system privileges. Persistent local storage is expected for this functionality, but users should be aware that the skill will create and update files in their home directory.
Guidance
This skill is coherent with its stated purpose and does not request credentials or external installs. Before enabling it, consider: (1) the skill will create and update files under ~/reflection/ — inspect those files and their permissions, and move or remove them if you don't want persistent logs; (2) user corrections and agent reasoning may include sensitive data (passwords, API keys, PII) — ensure you or the agent sanitize logs before storing or avoid logging sensitive exchanges; (3) review your agent configuration to confirm these files are not being uploaded or shared to external services; (4) if you need stricter privacy, run the skill in a restricted workspace or change the configured path to an encrypted location. If you want, I can produce a short checklist of commands to inspect and set safe permissions for ~/reflection/.
Latest Release
v1.1.0
Major rewrite. Visual workflow, pattern detection system, proactive lesson injection, and multi-trigger architecture.
More by @ivangdavila
Published by @ivangdavila on ClawHub
