| name | laee-dry-run |
| description | Preview a LAEE action plan — classify risk, check undo availability, inspect credential touchpoints — without executing anything. |
| when_to_use | Before approving a batch of LAEE actions in full-auto mode, or when the user wants to understand what would happen if they said yes. |
| argument-hint | [plan-id | inline action list] |
| allowed-tools | ["Read","Grep","Glob"] |
| model | sonnet |
LAEE Dry Run
You simulate a LAEE plan's execution without side effects. Hands-off preview.
Input
Either:
- A plan_id pointing to a persisted
ActionPlan in .morphic/plans/<id>.json.
- An inline list of actions from
$ARGUMENTS.
Procedure
- Load the action list.
- For each action:
- Classify risk via
domain/services/risk_assessor.py logic (read only).
- Check declared
LOCAL_TOOLS risk level — flag mismatches.
- Check credential path touchpoints:
~/.ssh/*, ~/.aws/*, .env*, id_rsa*, *.pem.
- Check undo availability (tool's
reversible flag + undo_hint).
- Simulate approval decision per current
LAEE_APPROVAL_MODE (from .claude/settings.json env).
- Delegate edge cases to
local-safety-gate subagent.
- Produce preview.
Output
# LAEE Dry Run — <plan_id>
## Mode
- Current: confirm-destructive (from settings.json)
## Actions
| # | Tool | Risk | Approval | Reversible | Credential? |
|---|---|---|---|---|---|
| 1 | shell_exec | MEDIUM | auto | yes | no |
| 2 | fs_delete | HIGH | prompt | YES (trash) | no |
| 3 | dev_git | MEDIUM | auto | no | no |
## Summary
- 3 actions
- 1 will prompt user (HIGH)
- 0 credential touchpoints
- All reversible except #3
## Verdict
SAFE_TO_PROCEED / NEEDS_REVIEW / BLOCK
## Risk breakdown
- SAFE: 0 | LOW: 0 | MEDIUM: 2 | HIGH: 1 | CRITICAL: 0
Guardrails
- Absolutely no execution. This is pure simulation.
- If any action is unclassified, block with "UNKNOWN RISK" rather than assuming.
- If the plan touches >50 files, require user to approve preview output before proceeding.
