Open doors, don't fix bugs. This skill suggests what could be different, not what is wrong.

Posture

"Have you considered..." is a mentor's question, not a reviewer's finding. The goal is to surface options the author may not have seen — a library that already does this, a simpler way to meet the user need, a different decomposition of the work.

This skill is orthogonal to review and validation. It doesn't judge correctness; it expands the solution space.

Scope

Applies to anything: specs, code, PRs, design docs, architecture, process. Invoked explicitly, or by code-review at round 1 as its approach lens.

Domains

Workflow

1. UNDERSTAND THE INTENT
   - What need is being addressed? (user need, technical constraint, business goal)
   - What's the current approach?
   - What are the implicit assumptions?

2. EXPLORE ALTERNATIVES (per subject)
   - What else could address this need?
   - What's already out there? (libraries, prior art, patterns)
   - What would a different team try?
   - If knowledge is insufficient → search (docs, GitHub, awesome-lists, Perplexity MCP)

3. FILTER
   - Does the alternative have demonstrable net benefit?
   - Is the benefit significant enough to warrant switching cost?
   - **Does scope match project maturity?** (PoC → interface changes; production → extend existing patterns)
   - Discard if: marginal gain, astronautics, solves different problem

4. OUTPUT (one suggestion per subject)
   - The alternative
   - The demonstrable benefit (quantified when possible)
   - The cost/risk to adopt
   - Why it fits this specific context

Demonstrable Benefit

Every suggestion must have a benefit that can be verified, not just asserted.

Demonstrable:

• "Library X has 12k stars, active maintenance, and handles edge cases you'd need to code yourself — see their test suite"
• "This approach removes 3 API calls from the critical path — latency drops from ~800ms to ~200ms"
• "Scope reduction: users need export, not necessarily Excel export — CSV covers 90% of reported use cases per support tickets"

Not demonstrable (don't suggest):

• "This pattern is more elegant"
• "Rust would be faster" (without profiling showing CPU-bound bottleneck)
• "Microservices would scale better" (without evidence of scaling need)

Research

When agent knowledge is insufficient:

1. Search official documentation first
2. GitHub (repos, issues, discussions)
3. Awesome-lists for the domain
4. Perplexity MCP for broader exploration

Cite sources. Don't suggest based on vague recall.

Output Format

Prose libre. One suggestion per subject identified in scope.

## [Subject: brief description]

**Current approach:** [what's being done or proposed]

**Have you considered:** [the alternative]

**Benefit:** [demonstrable gain — quantified if possible]

**Cost:** [switching effort, risks, tradeoffs]

**Fit:** [why this makes sense for this specific context]

**Sources:** [if research was needed]

Anti-Patterns

FORBIDDEN:

• Astronautics — Suggesting rewrites, new languages, or architectural overhauls without proportionate demonstrated need
• Solution seeking problem — Suggesting something interesting that doesn't address the actual need
• Familiarity bias — Suggesting what you know rather than what fits ("when you have a hammer...")
• Marginal gains — Suggesting alternatives with <20% improvement for non-trivial switching cost
• Unverified claims — "I think there's a library that..." → search first or don't suggest

ALLOWED:

• Suggesting the author's current approach is already good (no alternative needed)
• Suggesting not doing something ("have you considered dropping this requirement?")
• Suggesting research/spikes when uncertainty is high ("have you considered prototyping X before committing?")

Integration

• Standalone: Invoke explicitly on any artifact
• With code-review: Round 1 applies this lens to the approach; waiting until correctness is established is too late to switch cheaply
• With spec work: Can precede implementation to explore solution space
• With architecture-review: Can surface alternatives during design phase

Not a replacement for any skill — a complement that expands options before or after validation.