# Review Approach

> Research modern solutions and approaches for issues or features proposed in a report or plan. Use when user says "review approach", "review approaches", "research solutions", or wants external validation of a proposed direction.

- Skill: `talont-org/review-approach` (Agent Skill)
- Install (CLI): `npx skillmds@latest add talont-org/review-approach`
- Raw SKILL.md: https://api.skillmd.com/api/skills/talont-org/review-approach/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Docs & Writing
- Author: TalonT-Org (https://skillmd.com/u/talont-org)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/talont-org/review-approach

---


# Review Approach Skill

Research modern solutions, approaches, and strategies relevant to the issues or features proposed in a report or plan. Uses web search subagents to gather external perspective and surface options the team may not have considered.

## When to Use

- User says "review approach", "review approaches", or "research solutions"
- User wants to validate a proposed direction against current industry practice
- User has a plan or report and wants to explore what modern solutions exist
- After an investigation or plan, before committing to an approach

## Critical Constraints

**NEVER:**
- Modify any source code files
- Create files outside `{{AUTOSKILLIT_TEMP}}/review-approach/` directory
- Run subagents in the background (`run_in_background: true` is prohibited)

**ALWAYS:**
- Use subagents with web search for parallel research
- Use `model: "sonnet"` when spawning all subagents via the Task tool
- Keep findings concise and actionable
- Present options with trade-offs
- Make recommendations based on technical merit and project fit
- Tie research back to the specific problem context
- Include source URLs for all referenced material
- After writing the review file, emit the **absolute path** as a structured output
  token as your final output. Resolve the relative `temp/review-approach/...`
  save path to absolute by prepending the full CWD:
  ```
  review_path = /absolute/cwd/temp/review-approach/{filename}.md
  ```
  This token is MANDATORY — the pipeline cannot proceed without it.

## Workflow

### Step 1: Extract Research Targets

From the report, plan, or conversation context, identify the core problems and proposed features that need external research. Break them into distinct research topics.

### Step 2: Launch Parallel Web Search Subagents

Spawn general-purpose subagents (with web search) for each research topic. Each subagent should investigate:

- What modern solutions exist for this problem class
- How mature projects and frameworks approach it
- Recent developments, libraries, or patterns worth considering
- Known pitfalls and trade-offs of common approaches

Tailor the search queries to the specific technologies and constraints of the project.

### Step 3: Synthesize

Consolidate subagent findings into a concise review. For each research topic:

- **What exists**: The relevant modern approaches found
- **Trade-offs**: Strengths and weaknesses in the context of this project
- **Relevance**: How each option relates to the proposed direction

Drop anything that doesn't meaningfully inform the decision.

### Step 4: Write Review

Save to: `{{AUTOSKILLIT_TEMP}}/review-approach/review_approach_{topic}_{YYYY-MM-DD_HHMMSS}.md` (relative to the current working directory)

```markdown
# Approach Review: {Topic}

**Date:** {YYYY-MM-DD}
**Source:** {Name of the report/plan being reviewed}

## Context
{Brief statement of the problem and what was proposed}

## Research Findings

### {Research Topic 1}
{What modern solutions exist, trade-offs, relevance to this project}

**Sources:**
- [{title}]({url})

### {Research Topic 2}
{...}

## Recommendations
{What approaches to pursue and why, based on the research}

## Key Takeaways
{Concise bullets — what matters most for the decision at hand}
```

After saving the review file, emit the structured output token as the very last line
of your text output:

> **IMPORTANT:** Emit the structured output tokens as **literal plain text with no
> markdown formatting on the token names**. Do not wrap token names in `**bold**`,
> `*italic*`, or any other markdown. The adjudicator performs a regex match on the
> exact token name — decorators cause match failure.

```
review_path = {absolute_path_to_review_file}
```