Skip to content
Back to skills

Review Wave Plan

ASecurity

Adversarial review of a WavePlan before forging. Checks agent validity, path existence, desire alignment, and structural completeness. Returns ADVANCE or REVISE with specifics.

  • 4 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 9, 2026
ai-agentstypescriptpythonbashrefactoringgit

Security analysis

A100/100

Scanned October 9, 2026

npx -y skills add jgwill/miadi-orchestration-kit --skill review-wave-plan --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Review Wave Plan?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Review Wave Plan
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/jgwill-review-wave-plan/badge)](https://www.skillsdirectory.com/skills/jgwill-review-wave-plan)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: review-wave-plan
description: 'Adversarial review of a WavePlan before forging. Checks agent validity, path existence, desire alignment, and structural completeness. Returns ADVANCE or REVISE with specifics.'
---

Use this skill to validate a WavePlan before committing to script forging. Catch problems early rather than producing a broken orchestration script.

## When to use

Use this skill after `scan-and-plan` produces a WavePlan and before `forge-wave-script` writes the script.

## Review checklist

Work through each check below in order. For each check, record PASS or FAIL with a one-line finding.

### Check 1 — Agent name validity

For every agent named in the WavePlan's Phase Sequence:
- Verify the agent name appears in the "Available Agents" table of the Orchestration Scan Report (or in the agents directory of a selected kit)
- If an agent name does not match any known agent, mark FAIL and list the unmatched name

### Check 2 — Plugin dir path existence

For every path in the WavePlan's "Selected Plugins" table:
- Run `ls <path>/.github/plugin/plugin.json` to confirm the kit exists at that path
- If a path does not exist or has no `plugin.json`, mark FAIL and list the missing path

### Check 3 — Add-dir path existence

For every path in the WavePlan's "Selected Add-Dirs" table:
- Run `ls <path>` to confirm the directory exists
- If a path does not exist, mark FAIL and list the missing path

### Check 4 — Desire alignment

Re-read the original user desire. Re-read the Phase Sequence.
Ask: does each phase's tasks connect to the stated desire?

Flag FAIL if:
- The phases describe generic scaffolding work not related to the desire
- The desire involves a specific technology (e.g. TypeScript, Python, monorepo) but no agents with relevant expertise are selected
- The desire implies more than analysis (i.e. code changes are expected) but no implementation phase exists

### Check 5 — Issue creation presence

If the desire implies code changes (refactoring, new features, bug fixes):
- Check whether the target repo uses GitHub Issues (look for an `AGENTS.md`, `.github/copilot-instructions.md`, or CI workflows that reference `gh issue` or `github.com`).
- If the repo uses GitHub Issues: verify an Issue Creation phase is present in the Phase Sequence. If absent, mark FAIL: "Issue Creation phase missing for a desire that implies code changes in a GitHub-hosted repo."
- If the repo does not use GitHub Issues (e.g. private/local-only, Jira-based, or no evidence of issue workflow): mark N/A and note the reason.

### Check 6 — Adversarial review phase presence

- Verify Phase Sequence includes an adversarial review phase (Phase 5 or equivalent)
- The phase must name a specific agent responsible for adversarial review
- If absent or unnamed, mark FAIL

### Check 7 — Synthesis/closure phase presence

- Verify Phase Sequence ends with a synthesis or closure phase
- The phase must specify that it produces a session summary and relaunch instructions
- If absent, mark FAIL

## Output format

Produce the review result in this format:

```markdown
# WavePlan Review

## Checks

| # | Check | Result | Finding |
|---|-------|--------|---------|
| 1 | Agent name validity | PASS/FAIL | <finding or "all agents verified"> |
| 2 | Plugin dir paths | PASS/FAIL | <finding or "all paths exist"> |
| 3 | Add-dir paths | PASS/FAIL | <finding or "all paths exist"> |
| 4 | Desire alignment | PASS/FAIL | <finding or "phases align with desire"> |
| 5 | Issue creation | PASS/FAIL/N-A | <finding or "present" or "N/A: desire is analysis-only"> |
| 6 | Adversarial review phase | PASS/FAIL | <finding or "present and named"> |
| 7 | Synthesis phase | PASS/FAIL | <finding or "present with relaunch instructions"> |

## Verdict

**ADVANCE** — All checks passed. Proceed to forge-wave-script.

— or —

**REVISE** — The following items must be fixed before forging:
1. <specific item to fix>
2. <specific item to fix>
(list only failing checks with actionable fix descriptions)
```

If verdict is REVISE, do not proceed to forging. Return the review to the user or the orchestrating skill for plan correction.

If verdict is ADVANCE, state:
```
✓ WavePlan is valid. Use 'forge-wave-script' to produce the bash script.
```

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.