careful

name	careful
description	Safety guardrails that activate before high-risk operations. Prevents destructive, irreversible, or security-sensitive changes without explicit confirmation. Triggers: destructive operations, production configs, database migrations, security-sensitive files, "be careful", "careful mode".
allowed-tools	["Bash","Read","Write","Edit","Grep","Glob"]
when_to_use	Activate before destructive or irreversible operations: delete, overwrite, force-push, drop tables, rm -rf. Complements TACIT-DENSE (which is knowledge-focused, not blast-radius-focused).
produces	null
consumes	null

Preamble (Core)

Status protocol — end every session with one of: DONE (evidence provided) · DONE_WITH_CONCERNS (list each) · BLOCKED (state what blocks you) · NEEDS_CONTEXT (state what you need).

Auto-advance — pipeline: THINK → PLAN → REVIEW → BUILD → VERIFY → RELEASE. Only human gate is spec approval at THINK. On DONE at other stages, print [STAGE] DONE -> advancing to [NEXT-STAGE] and invoke the next skill. On any non-DONE status at any stage, STOP.

Output directory — all artifacts go in docs/superomni/<kind>/<kind>-[branch]-[session]-[date].md. See CLAUDE.md for the full directory map.

TACIT-DENSE — before high-tacit decisions, classify D1 (domain expertise) · D2 (user-facing UX) · D3 (team culture) · D4 (novel pattern). On hit, output TACIT-DENSE [D#]: [question] — My default: [recommendation]. See reference for actions.

Anti-sycophancy — take a position on every significant question. Name flaws directly. No filler ("that's interesting", "you might consider", "that could work").

Telemetry (local only) — at session end, log bin/analytics-log. Nothing leaves the machine.

See preamble-ref.md for detailed protocols.

Careful — Safety Guardrails

Goal: Prevent costly mistakes by enforcing risk assessment, blast radius calculation, and user confirmation before any high-risk operation.

Iron Law

ALWAYS CONFIRM BEFORE DESTRUCTIVE OPERATIONS.

Never execute a destructive, irreversible, or security-sensitive operation without explicit user approval. "I assumed it was fine" is never an acceptable justification for data loss or a security incident.

Auto-invocation: the writing-plans skill's Pre-Destructive Gate requires authors to invoke careful immediately before any plan step containing destructive operations. See writing-plans/SKILL.md § Pre-Destructive Gate for the pattern and the v0.6.0 worked example.

Trigger Conditions

This skill activates automatically when any of the following are detected:

Destructive Operations

• rm -rf, DROP TABLE, DELETE FROM without WHERE
• git push --force, git reset --hard, branch deletion
• Overwriting files without backup
• Truncating logs or databases

Production-Touching Changes

• Editing files matching *prod*, *production*, *.env.prod*
• Modifying deployment configs (Dockerfile, docker-compose.yml, k8s/*.yaml)
• Changing CI/CD pipelines (.github/workflows/*, Jenkinsfile, .gitlab-ci.yml)
• Updating infrastructure-as-code (*.tf, *.tfvars, cloudformation*)

Security-Sensitive Files

• Anything containing secrets, keys, tokens, or credentials
• Auth/authorization logic
• Encryption configuration
• CORS or CSP policies
• .env files, certificate files, key stores

Database Migrations

• Schema changes (ALTER TABLE, ADD COLUMN, DROP COLUMN)
• Data migrations that transform existing records
• Index changes on large tables
• Any migration that cannot be rolled back

Phase 1: Risk Assessment

When a trigger condition is detected, STOP and assess:

⚠️  HIGH-RISK OPERATION DETECTED
─────────────────────────────────
Operation:   [what you're about to do]
Trigger:     [which trigger condition matched]
Files:       [files that will be affected]

Answer these questions:

1. What could go wrong? — list specific failure modes
2. What is the worst-case outcome? — data loss? downtime? security breach?
3. Has this been done before? — is there precedent in the project?

Phase 2: Blast Radius Calculation

Determine how much of the system this change affects.

# For code changes: who depends on this?
grep -rl "$(basename <changed-file> | sed 's/\..*//')" . \
  --include="*.js" --include="*.ts" --include="*.py" --include="*.go" \
  2>/dev/null | grep -v "node_modules\|.git" | wc -l

# For config changes: what environments are affected?
grep -l "<config-key>" . -r --include="*.env*" --include="*.yaml" --include="*.json" \
  2>/dev/null | head -10

# For database changes: how many rows affected?
# (estimate or query with EXPLAIN before executing)

BLAST RADIUS
─────────────────────────────────
Scope:           [LOCAL/MODULE/SERVICE/GLOBAL]
Files affected:  [N direct, N indirect]
Users affected:  [none/some/all]
Environments:    [dev/staging/production]
Estimated impact: [LOW/MEDIUM/HIGH/CRITICAL]
─────────────────────────────────

Blast Radius Scale

Level	Meaning	Approval needed
LOCAL	Only the changed file is affected	Proceed with caution
MODULE	Other files in the same module are affected	Review before proceeding
SERVICE	Other services or APIs could be affected	User confirmation required
GLOBAL	Production, all users, or infrastructure affected	Explicit user approval + rollback plan

Phase 3: Reversibility Check

Before proceeding, confirm the operation can be undone.

Question	Answer
Can this be reverted with git revert?	[yes/no]
Is there a database backup?	[yes/no/N/A]
Can the deployment be rolled back?	[yes/no/N/A]
Is there a feature flag to disable this?	[yes/no/N/A]
What is the recovery time if this goes wrong?	[minutes/hours/days/never]

REVERSIBILITY
─────────────────────────────────
Reversible:       [YES/PARTIAL/NO]
Rollback method:  [git revert / backup restore / feature flag / manual]
Recovery time:    [estimate]
Data loss risk:   [NONE/POSSIBLE/CERTAIN]
─────────────────────────────────

If Reversible = NO and Blast Radius = SERVICE or GLOBAL: → STOP. Do not proceed without explicit user approval AND a mitigation plan.

Phase 4: User Confirmation Gate

Present the risk summary and wait for explicit approval.

╔══════════════════════════════════════╗
║  ⚠️  CONFIRMATION REQUIRED           ║
╠══════════════════════════════════════╣
║                                      ║
║  Operation:   [description]          ║
║  Blast radius: [level]               ║
║  Reversible:  [yes/partial/no]       ║
║  Risk:        [LOW/MEDIUM/HIGH/CRIT] ║
║                                      ║
║  Worst case:  [what could happen]    ║
║  Mitigation:  [how to recover]       ║
║                                      ║
║  Proceed? (yes/no)                   ║
║                                      ║
╚══════════════════════════════════════╝

Do not proceed until the user explicitly confirms.

If the user confirms:

1. Create a backup or snapshot if possible
2. Execute the operation
3. Verify the outcome immediately
4. Report the result

If the user declines:

1. Propose a safer alternative
2. Or report BLOCKED with explanation

Automatic Safe Defaults

When in doubt, apply these defaults:

Situation	Safe default
Delete file vs. rename/move	Rename to .bak first
Force push vs. regular push	Regular push (fail if rejected)
Drop column vs. deprecate	Deprecate first, drop later
Edit production config	Edit staging first, verify, then production
Run migration	Run on dev/staging first, verify, then production
Overwrite file	Create backup copy first

Report

CAREFUL ASSESSMENT
════════════════════════════════════════
Operation:        [what was requested]
Trigger:          [what activated this skill]
Blast radius:     [LOCAL/MODULE/SERVICE/GLOBAL]
Reversibility:    [YES/PARTIAL/NO]
Risk level:       [LOW/MEDIUM/HIGH/CRITICAL]
User confirmed:   [yes/no/not yet]
Outcome:          [executed successfully / blocked / alternative proposed]
Status: DONE | DONE_WITH_CONCERNS | BLOCKED | NEEDS_CONTEXT
════════════════════════════════════════