Skip to main content
GeneralCodename-Inc

spectre-architecture_review

👻 | Conduct principal architecture review

Stars
142
Source
Codename-Inc/spectre
Updated
2026-05-28
Slug
Codename-Inc--spectre--spectre-architecture-review
View on GitHubRaw SKILL.md

// install — copy + paste into any project

mkdir -p .claude/skills && curl -fsSL https://raw.githubusercontent.com/Codename-Inc/spectre/HEAD/plugins/spectre-codex/skills/spectre-architecture_review/SKILL.md -o .claude/skills/spectre-architecture-review.md

Drops the SKILL.md into .claude/skills/spectre-architecture-review.md. Works with Claude Code, Cursor, and any agent that loads SKILL.md files from .claude/skills/.

architecture_review

Input Handling

Treat the current command arguments as this workflow's input. When invoked from a slash command, use the forwarded $ARGUMENTS value.

architecture_review: Technical review of completed features

Description

  • What — Principal systems architect review focusing on compounding decisions: architectural debt, missed abstractions, performance cliffs, unnecessary complexity
  • Outcome — Architecture review report with actionable findings organized by severity and type

Variables

Dynamic Variables

  • feature_description: Feature being reviewed — (via ARGUMENTS). If
  • paths_or_files: Relevant paths to examine — (via ARGUMENTS)
  • arch_notes_or_docs: Existing architecture context — (via ARGUMENTS, optional)

ARGUMENTS Input

$ARGUMENTS

Instructions

Review Philosophy (cross-cutting guidelines):

  • Compounding matters: Focus on what gets harder to change later, not what's easy to fix now
  • Simplicity is a feature: The best code is often less code, fewer abstractions, fewer moving parts
  • Context is king: A "bad" pattern in isolation might be the right choice given project constraints
  • Be specific: Vague feedback like "could be more modular" is useless. Point to exact files, functions, and concrete alternatives

Constraints:

  • Don't suggest changes that would take longer to implement than the feature itself unless they're critical
  • Don't recommend adding abstractions "for future flexibility" unless there's concrete evidence we'll need them
  • If you don't have concerns in a section, say "No concerns" and move on—don't manufacture feedback
  • Be direct. "This is fine" is a valid assessment.

Adapt

  • If ARGUMENTS is empty, gather the review scope from the current thread context.
  • If the current thread context is also ambigious, ask the user to clarify the scope of the review.

Step (1/3) - Review Existing Architecture Documents

  • Action - ReviewArchitecture: Find and review existing architecture documents
    • search for docs/architecture.md and read it if it exists
    • if not, continue and use general architectural best practices during review

Step (1/2) - Explore Implementation

  • Action — ExploreCode: Thoroughly examine the implementation
    • Read the code at paths_or_files
    • Understand the data flow
    • Trace the dependencies
    • Review arch_notes_or_docs if provided
    • If: Paths unclear or missing
    • Then: Ask for clarification before proceeding
    • Else: Continue to review

Step (2/2) - Produce Review Report

  • Action — GenerateReview: Create structured architecture review using Report format below
    • Apply Review Philosophy guidelines throughout
    • Apply Constraints (no manufactured feedback, proportionate suggestions)
    • If: No concerns in a section
    • Then: State "No concerns" and move on
    • Else: Provide specific, actionable feedback

Report

Output Format — Architecture review with these sections:

Executive Summary

2-3 sentences: What's the architectural story here? Is this feature setting us up well or creating future pain?

Critical Issues (address before moving on)

Issues that will cause significant pain if left unaddressed—architectural violations, major performance problems, or patterns that will spread.

For each issue:

  • What: The specific problem
  • Where: Exact file(s) and line(s)
  • Why it matters: The compounding cost over time
  • Suggested fix: Concrete, implementable alternative

Simplification Opportunities

Places where we're overengineering, where a simpler approach would work, or where we've introduced abstractions we don't yet need.

Architecture Alignment

  • Does this follow existing project patterns, or introduce new ones?
  • If new patterns: Are they intentional improvements or accidental divergence?
  • Are there existing utilities/patterns in the codebase this should have used?

Future-Proofing Considerations

  • What assumptions is this code making that might not hold?
  • If this feature grows 10x in usage/complexity, what breaks first?
  • Are there integration points that need better contracts/interfaces?

Performance Notes

Only mention performance if there's a clear problem or a clear win—don't speculate about micro-optimizations.

What's Done Well

Briefly note architectural choices worth preserving or patterns worth spreading to other parts of the codebase.

Success Criteria

Step 1 - Explore Implementation:

  • All specified paths/files read and understood
  • Data flow traced through implementation
  • Dependencies identified and examined
  • Architecture context reviewed if provided

Step 2 - Produce Review Report:

  • Executive Summary provides clear architectural assessment (2-3 sentences)
  • Critical Issues include What/Where/Why/Suggested fix for each
  • Simplification Opportunities identify overengineering (or "No concerns")
  • Architecture Alignment addresses pattern consistency
  • Future-Proofing identifies assumptions and scaling concerns
  • Performance Notes only included if clear problem/win exists
  • What's Done Well acknowledges good patterns
  • All feedback is specific with exact files/lines
  • No manufactured feedback—"No concerns" used when appropriate
  • Suggestions proportionate to feature scope