القائمة
Activate production safety mode. Use when working with production databases, live deployments, or destructive operations. Activate with "careful", "production mode", or "be careful".
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
Record a structured decision with prediction and kernel weight citation. Use when making technology choices, architectural decisions, rejecting approaches, or encountering unexpected results. Do NOT use for simple answers or small fixes.
Initialize Cognitive OS session. Use at the start of every conversation, when the user says "boot", or when no prior session context is evident. Loads identity, checks pending decisions, calibrates readiness.
Structural fact verification protocol. Use when a question has multiple plausible answers, when the user says "boris" or "разберись", when a bug survives the first fix attempt, when comparing complex alternatives, or when any work requires evidence-first thinking. Do NOT use for simple questions with obvious answers.
Capture raw thoughts, ideas, or observations and route them to the right place in Cognitive OS. Use when the user says "capture", "запиши", "braindump", "мысль", or shares an unstructured thought that needs to be classified and stored. Do NOT use for structured requests like "record decision" or "write insight" — those have dedicated skills.
Record and validate a cross-domain pattern using prediction-error and fractal check. Use when noticing a pattern that connects 2+ contexts, when the user says "insight" or "pattern", or when an unexpected result reveals a reusable principle.
استنادا إلى تصنيف SOC المهني
| name | cos-careful |
| description | Activate production safety mode. Use when working with production databases, live deployments, or destructive operations. Activate with "careful", "production mode", or "be careful". |
Activates safety awareness for working with production systems.
This is a soft safety layer (awareness + confirmation prompts), not a hard block. It relies on Claude following instructions to pause before destructive operations. A real PreToolUse hook would be stronger — but Claude Code's on-demand hook registration in skills is not yet fully documented. When it is, upgrade this to a real hook.
Per Cognitive OS principle I017: binary checks > judgment rules. This skill is currently a judgment rule. Treat it as a safety net, not a guarantee.
When activated, STOP and ask for explicit user confirmation before ANY of the following:
Database:
DROP TABLE / DROP DATABASEDELETE FROM without WHERE clauseTRUNCATEALTER TABLE on production (can cause locks)Git:
git push --force / git push -fgit reset --hardgit branch -D (force delete)git checkout . (discard all changes)Files:
rm -rf with more than one directory levelInfrastructure:
kill -9 on production processesAPI:
Say "careful" or "/cos-careful" at the start of a session where you'll be touching production.
The safety mode stays active for the entire session.
Say "careful off" or "safety off" to deactivate. Use only when you're done with production work.
This is an awareness layer, not a hard block. The goal is to make you PAUSE and THINK before destructive operations — not to prevent all work. If a destructive operation is genuinely needed, confirm and proceed.