careful

name	careful
description	Adds a protective decision layer before risky operations so destructive or irreversible actions require explicit evidence and intent. Use when the user asks to be careful with this, add a safety guard, protect the risky action, or pause before destructive work — even when they only say "don't break anything". Adds an intent/confirmation check on actions; defers locking a path from edits to `safety-guardrails/freeze`, the combined intent-plus-boundary guard to `safety-guardrails/guard`, and lifting a boundary to `safety-guardrails/unfreeze`.
version	1.0.0

Careful

Purpose

Adds a protective decision layer before risky operations so destructive or irreversible actions require explicit evidence and intent.

Use This Skill When

Use this skill to add a confirmation checkpoint before a destructive or irreversible action runs:

• "be careful with this" / "pause before destructive work" — require explicit intent before the action proceeds
• "add a safety guard" — insert an evidence-and-intent check ahead of the risky step
• "protect the risky action" — make the consequences explicit before committing

Route elsewhere to lock a path from any edits (safety-guardrails/freeze), combine intent checks with a write boundary (safety-guardrails/guard), or lift an existing boundary (safety-guardrails/unfreeze).

Inputs

• Description of the risky or destructive action, the affected boundary, and the current environment state.
• Evidence required to authorize the action, such as backup confirmation, rollback availability, or owner approval.
• Known constraints including protected environments, irreversibility thresholds, and side-effect blast radius.

Outputs

• Safety guard decision record with go, no-go, or escalate verdict and the exact evidence used.
• Blocked-action list identifying what cannot proceed and what proof would unlock it.
• Escalation notes when the requested action exceeds the available evidence or authority.

Workflow

1. Identify the risky action, affected boundary, and the evidence required before any destructive, irreversible, or hard-to-undo step is allowed.
2. Check that the current environment, target surface, and requested intent all match; if any one is ambiguous, stop before execution.
3. Record a go, no-go, or escalate decision with the exact evidence used, the remaining uncertainty, and the safest next action available.
4. Return a careful record that states what can proceed now, what stays blocked, and what proof would unlock the next step safely.

Required Contracts

• Proactive triggers: Offer the next sensible action when the surrounding context clearly implies it and the skill can advance safely without a prompt loop.
• Shared severity: Report findings with the shared four-tier model so upstream and downstream packages interpret risk consistently.
• Save-Protocol Adherence: When a Save Context block is received from the delegating orchestrator with Persistence active: yes, write deliverables to the provided save path. Saving is mandatory when persistence is active.

Collaboration Surface

• None required beyond the active task surface.

Review Expectations

• Base every safety verdict on explicit evidence — backup confirmation, rollback test, or owner approval — not on assumed safety.
• Surface missing prerequisites and irreversibility risks before the action reaches execution.
• Shape the guard record so the requesting workflow knows exactly what is safe to proceed with and what remains blocked.

Skip Rule

Skip only when the requested surface, tool, or environment does not exist and a safe fallback is unavailable.

Failure Modes

Scenario	Response
The risky action is described too broadly to tell what would actually be affected	Narrow the boundary first and refuse to bless a vague request that could hide destructive scope.
Intent, evidence, and environment point at different targets	Freeze the decision, name the mismatch, and require one coherent target before any action proceeds.
The requested step depends on a prerequisite that has not been verified	Keep the action blocked and state the minimum evidence needed to prove the precondition.
Another contributor could misread the outcome as unconditional approval	Write the residual risk and release conditions directly into the careful record so the next step is bounded.

Save Protocol

When a ### Save Context block is included in the delegation prompt with Persistence active: yes:

1. Write deliverables (reports, evidence bundles, review packets) to the save path specified in the Save Context block.
2. Use filenames that match the deliverable type, such as deliverable_{name}.md, report_{name}.md, or review-packet.md.
3. Do not create or manage _phase-state.md — that is the delegating orchestrator's responsibility.

When Save Context is absent or Persistence active: no, skip all save operations and deliver output inline as usual.

References

• references/workflow.md for the detailed operating sequence and decision rules.
• references/examples.md for concrete request patterns and response shapes.

Packaging Notes

Package SKILL.md, references/workflow.md, and references/examples.md together. Keep generated reports and archives outside the skill directory.