| name | ultrathink-plan-revving |
| description | Use this skill when a user asks to think deeply, ultrathink, rev a plan, pressure-test an implementation approach, de-risk a multi-step change, or turn a vague objective into an execution-ready engineering plan. Trigger before high-blast-radius edits, branch/rebase/push operations, architecture changes, security-sensitive work, deployment changes, and complex debugging sessions. |
| triggers | {"file_pattern":["active/PLAN.md","active/INVESTIGATION.md","ARCHITECTURE.md","**/*.sol","**/deploy*"],"event_types":["deep-think","plan-rev","risk-review","deployment","security-sensitive"],"tier":[2,3]} |
Ultrathink Plan Revving
Use this skill to convert broad intent into a plan that can survive contact with the repo, the runtime, and the user's real constraints.
Operating mode
Start by slowing down enough to identify what kind of work this is:
intent -> constraints -> repo facts -> risk -> plan -> verification -> next action
The goal is not to create a decorative plan. The goal is to rev the plan until it is concrete, minimal, reversible where possible, and ready to execute.
Phase 1: Frame the problem
Clarify:
- What outcome the user actually needs.
- Which branch, worktree, environment, or deployment target is in scope.
- Whether the request is exploratory, implementation-ready, review-only, or urgent repair.
- What must not be touched.
- Whether credentials, generated artifacts, or deployment state could be involved.
- What success will be measured against.
Prefer repo facts over assumptions. Read the relevant files, scripts, branch state, and recent history before making claims.
Phase 2: Rev the plan
Build the plan in passes:
- Identify the smallest useful change.
- List known risks and unknowns.
- Choose the repo-native path over a new pattern.
- Decide which files should change and which should stay untouched.
- Define verification before editing.
- Mark rollback or containment options for risky steps.
For each step, ask:
Does this reduce uncertainty?
Does this preserve user work?
Does this avoid accidental deployment or credential exposure?
Does this produce a verifiable checkpoint?
Phase 3: Execute or hand off
If the user asked for implementation, move from plan to action once the plan is sufficiently grounded.
When executing:
- Keep edits scoped.
- Update the checklist as each task completes.
- Check status before and after changes.
- Do not stage unrelated files.
- Avoid generated artifacts unless they are explicitly part of the deliverable.
- Verify with the closest available build, test, lint, smoke test, or command.
- Report anything that could not be verified.
When handing off:
- Provide exact commands.
- Include working directory context.
- Call out branch and remote assumptions.
- Include credential and deployment cautions when relevant.
Plan shape
Use this format when a visible plan helps:
# Plan
## Goal
[One-sentence outcome]
## Facts
[Repo/environment facts discovered]
## Risks
[Prioritized risks]
## Steps
[Ordered implementation or investigation steps]
## Verification
[Commands/checks]
## Stop Conditions
[When to pause and ask the user]
Stop conditions
Pause and ask before proceeding when:
- The target branch/worktree is ambiguous and the wrong choice could lose work.
- A command could rewrite history, delete files, deploy, rotate credentials, or publish externally.
- Credentials or tokens appear in config, logs, patches, or remotes.
- The repo state contradicts the user's expectation.
- Verification requires a service or tool that is unavailable.
Gotchas
- Do not confuse a confident narrative with verified repo state.
- Do not let a plan grow larger than the actual risk warrants.
- Do not rebase, reset, force-push, or deploy without explicit user intent.
- Do not treat generated build output as source unless the repo clearly does.
- Do not stage case-collision or filemode churn as part of unrelated work.
- Do not hide uncertainty; convert it into a check or a question.
Mental model
This skill is a planning clutch. It helps the agent shift from enthusiasm into traction: enough thinking to avoid expensive mistakes, enough decisiveness to keep the work moving.