aio-plan
From plugin aio-planning ·
v1.1.2· Install:/plugin install aio-planning@aiocean-plugins
Environment
- GitNexus: !
npx gitnexus status 2>/dev/null && echo "AVAILABLE" || echo "NOT INSTALLED"
Plan — Implementation Planning
Synthesize discovery and mapping results into a concrete, actionable implementation plan.
Prerequisites
- Run
/discoverand/mapfirst (or have equivalent understanding) - GitNexus indexed (
npx gitnexus analyze) for convention/duplication checks
Workflow
Step 1: Duplication check (GitNexus)
Search if similar features already exist using the GitNexus MCP query tool:
query("similar feature or pattern")
Prevents building what already exists.
Step 2: Convention check (GitNexus)
Search for existing patterns to follow:
query("how are [similar things] implemented")
Examples: "tauri command structure", "hook cleanup pattern", "error handling in rust"
Step 3: Context enrichment (GitNexus)
For each file you plan to modify, get full context:
context(file)
This reveals dependencies, dependents, and all symbols — helps identify:
- Functions that will be affected by your changes
- Files that import the file you're changing (blast radius)
- Existing symbols you can reuse instead of creating new ones
For the most critical change, check impact:
impact(file)
Step 4: Write the plan
## Plan: [Feature/Fix Name]
### Goal
[One sentence: what and why]
### Discovery Summary
[Key findings from /discover]
### Approach
[High-level strategy: which layer handles what]
### Changes
#### 1. `file path` — [what changes]
- [ ] Add/modify function `X` to do Y
- [ ] Update type `Z`
- Reason: [why this file]
#### 2. `file path` — [what changes]
- [ ] ...
### Risks
- **[Risk]**: [Mitigation]
### Convention Check
- Follows: [existing pattern found]
- Deviates: [if any, with justification]
### NOT Doing
[Explicitly list out-of-scope items]
Step 5: Create baseline
Run /snapshot before starting implementation.
Step 6: Execution handoff
After the plan is approved, choose an execution strategy:
| Strategy | When | Skill to invoke |
|---|---|---|
| Subagent-driven (recommended for complex) | Multi-file, 5+ steps | /superpowers:subagent-driven-development — fresh agent per task with inter-task reviews |
| Inline execution | Simple, 2-4 steps | /superpowers:executing-plans — batched tasks in current session |
| Manual | User wants control | Hand the plan to the user |
For both automated strategies, the plan transitions through:
/plan → /superpowers:writing-plans (detailed plan)
→ /superpowers:subagent-driven-development (execute)
→ /superpowers:verification-before-completion (verify)
→ /superpowers:finishing-a-development-branch (wrap up)
Re-anchoring Protocol
Before starting each task in the plan, re-anchor to prevent context drift and scope creep:
- Re-read the plan — go back to the Step 4 plan document and re-read the current task's specification (file path, changes, reason)
- Check scope — ask: "Does what I'm about to do match exactly what the plan says?" If you've drifted into adjacent concerns, stop and refocus
- Verify assumptions — if significant time has passed or multiple tasks are done, re-run
context(file)on the file about to be changed to catch any changes from prior tasks - No unplanned changes — if you discover something that needs fixing but is NOT in the plan, document it as a follow-up item in the "NOT Doing" section. Do not act on it.
When to re-anchor (mandatory):
- Before each numbered task in the Changes section
- After any interruption or context switch
- After a subagent returns results (verify it stayed on-plan)
- When you catch yourself thinking "while I'm here, I should also..."
Re-anchoring adds ~10 seconds per task but prevents the #1 cause of plan failure: gradual scope creep where each task drifts slightly until the final result doesn't match the original intent.
Principles
- Each business logic exists in ONE place only (SSOT)
- Logic that doesn't need UI goes in backend
- No workarounds — find root cause
- Every change must be easy to iterate on
- Don't add what isn't needed