Latent Space Engineering
- 77 installs
- 325 repo stars
- Updated August 2, 2026
- athola/claude-night-market
Frame prompts when you dispatch three or more parallel review agents so findings stay evidence-backed and severity-weighted instead of noisy.
About
latent-space-engineering in this catalog entry documents competitive review framing—a small methodology slice for orchestrating multiple agent reviewers on the same artifact. Solo builders running Claude Code or Cursor with parallel subagents often get shallow or duplicated feedback; this skill gives copy-paste dispatch language that nudges rigor when you have three or more reviewers by making comparison explicit, while recommending collaborative framing for only two agents to avoid overhead. It also encodes anti-gaming rules: thoroughness means evidenced, severity-prioritized findings, not inflated issue counts. Use it when you wire up code-reviewer agents, PR review toolkits, or panel-style reviews during ship gates or post-incident code sweeps. It does not replace a single-reviewer checklist; it optimizes the social mechanics of multi-agent review under competitive incentives.
- Competitive framing block for 3+ independent reviewers with comparison of thoroughness
- Collaborative framing for exactly two agents to split scope without redundant competition
- Guards against perverse incentives: citations required, severity weighting, discard findings without code references
- Applies to pensive reviewers, pr-review-toolkit, and war-room-style expert panels
- Parent latent-space-engineering methodology; this slice is competitive-review dispatch (~200 tokens estimated)
Latent Space Engineering by the numbers
- 77 all-time installs (skills.sh)
- Ranked #497 of 1,352 Code Review & Quality skills by installs in the Skillselion catalog
- Security screen: LOW risk (skills.sh audit)
- Data as of Aug 5, 2026 (Skillselion catalog sync)
npx skills add https://github.com/athola/claude-night-market --skill latent-space-engineeringAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| Installs | 77 |
|---|---|
| repo stars | ★ 325 |
| Security audit | 3 / 3 scanners passed |
| Last updated | August 2, 2026 |
| Repository | athola/claude-night-market ↗ |
What it does
Frame prompts when you dispatch three or more parallel review agents so findings stay evidence-backed and severity-weighted instead of noisy.
Files
Latent Space Engineering
Shape agent behavior by framing instructions for optimal performance. Distinct from context engineering (packing the right information), this skill addresses HOW instructions are framed to put agents in productive mental states.
When To Use
- Composing agent dispatch prompts
- Writing skill instructions that guide behavior
- Dispatching 3+ parallel review agents
- Generating code or documentation that must match
an existing style
When NOT To Use
- Packing factual context (use context-optimization)
- Simple single-shot tasks with no behavioral nuance
- Tasks where instruction tone is irrelevant
Core Techniques
1. Emotional Framing
Replace threat-based prompting with calm, confident instructions. Fear-based prompts cause rushing and corner-cutting.
Load module: modules/emotional-framing.md
2. Style Gene Transfer
Inject exemplar code or prose into context before requesting output. Agents reproduce stylistic attributes from pre-loaded samples.
Load module: modules/style-gene-transfer.md
3. Competitive Review
Frame multi-agent review dispatch with competitive incentives to increase rigor and thoroughness.
Load module: modules/competitive-review.md
Quick Reference
| Technique | When | Module |
|---|---|---|
| Emotional framing | Any agent prompt | emotional-framing |
| Style gene transfer | Code/doc generation | style-gene-transfer |
| Competitive review | 3+ parallel reviewers | competitive-review |
Competitive Review Framing
Principle
When multiple agents review the same artifact, framing the dispatch with competitive incentives increases thoroughness and evidence quality.
For 3+ Agents (Competitive)
Add this to each agent's dispatch prompt:
~~~ You are one of N independent reviewers analyzing this code. Each reviewer's findings will be compared. The most thorough and well-evidenced findings will be prioritized for action. Focus on depth over breadth. ~~~
For 2 Agents (Collaborative)
Use collaborative framing instead:
~~~ You and one other reviewer will cover different angles. Your findings will be integrated into a single report. Focus on your assigned scope. ~~~
Competitive framing adds overhead for fewer than 3 agents and can cause redundant coverage.
Avoiding Perverse Incentives
Thoroughness means evidence-backed and prioritized by severity, not volume. To prevent inflated issue counts:
- Require evidence citations for each finding
- Weight findings by severity, not count
- Discard findings that lack specific code references
Where To Apply
pensivereview agents (code-reviewer, etc.)pr-review-toolkitmulti-agent reviewsattune:war-roomexpert panels- Any parallel dispatch of 3+ review subagents
Anti-Pattern
Do NOT use competitive framing for:
- Implementation agents (cooperation > competition)
- Planning agents (synthesis > competition)
- Single-agent dispatch (no comparison possible)
Emotional Framing
Principle
Calm, confident instructions produce better results than threats or fear-based prompting. Agents under stress (threat-heavy prompts) rush, cut corners, and produce lower-quality output.
Anti-Patterns (Replace These)
| Threat Pattern | Problem |
|---|---|
| "You MUST do X or the system will fail" | Creates urgency that bypasses careful reasoning |
| "CRITICAL: failure to comply will..." | Frames task as punishment avoidance |
| "WARNING: do NOT deviate from..." | Implies deviation is the default |
| "NEVER do X under ANY circumstances" | Absolute prohibitions invite edge-case failures |
| "This is your LAST CHANCE to..." | Artificial scarcity degrades quality |
Preferred Patterns (Use These)
| Confident Pattern | Why It Works |
|---|---|
| "You've got this. Take your time with X." | Encourages careful reasoning |
| "Focus on getting X right. The details matter here." | Directs attention without threat |
| "This is important work. Here's what good looks like..." | Sets positive exemplar |
| "Take a careful look at X before proceeding." | Promotes deliberation |
| "Your goal is to produce Y. Here's the approach..." | Outcome-focused, not fear-focused |
Checklist for Skill Authors
1. Read your prompt aloud. Does it sound threatening? 2. Count urgency markers (MUST, NEVER, CRITICAL, WARNING). Are they justified by genuine safety constraints? 3. Replace threat language with confidence language. 4. Keep urgency markers for genuine safety constraints only (e.g., "NEVER commit secrets to git" is valid).
When Urgency IS Appropriate
Some constraints are genuinely critical:
- Security boundaries (secret handling, auth)
- Data loss prevention (destructive operations)
- Constitutional rules (human approval requirements)
For these, urgency markers are appropriate. The test: would violating this instruction cause real harm? If yes, keep the strong language. If no, soften it.
Style Gene Transfer
Principle
Agents reproduce stylistic attributes from pre-loaded samples. By injecting a representative exemplar before requesting output, you transfer naming conventions, comment style, error handling patterns, and prose voice.
Template
~~~ Review this prior work for style and conventions: --- [exemplar snippet, 50-200 lines] --- Now apply the same style to your output for: [task] ~~~
When To Use
- Generating code that must match codebase conventions
- Writing documentation in an established voice
- Creating tests that follow existing test patterns
- Producing configuration that matches project style
When NOT To Use
- Greenfield projects with no style precedent
- Exemplar exceeds 200 lines (diminishing returns, wasted tokens)
- Task is purely algorithmic (style is irrelevant)
- Output format is rigidly specified (template-driven)
Size Guidelines
| Exemplar Size | Effectiveness | Token Cost |
|---|---|---|
| 20-50 lines | Basic style transfer | Low |
| 50-100 lines | Good pattern coverage | Medium |
| 100-200 lines | Excellent fidelity | High |
| 200+ lines | Diminishing returns | Wasteful |
Selection Criteria
Choose exemplar code that:
1. Is from the same codebase and language 2. Represents the BEST style (not legacy code) 3. Contains the patterns you want reproduced 4. Is recent (reflects current conventions)
Related skills
Improve Codebase ArchitectureSafely deepen clusters of shallow modules into cohesive, testable units while respecting their external dependencies.566k196k
Caveman ReviewGet ultra-compressed, one-line code review comments that cut noise while keeping every actionable fix.270k94.7kCodebase DesignShared vocabulary for designing deep modules: improve a module's interface, find deepening opportunities, decide where a seam goes, make code more testable.266k196kCavecrewDelegate coding tasks to specialized subagents that return compressed output, keeping the main context window usable for much longer sessions.220k94.7kCode ReviewReviews a branch or PR diff on two axes at once: conformance to coding standards plus a code-smell baseline, and whether it actually implements the original spec.201k196k
Requesting Code ReviewDispatch a consistent, high-signal code reviewer subagent that catches plan deviations and quality issues before merging or continuing development.183k264kFAQ
Is Latent Space Engineering safe to install?
skills.sh reports 3 of 3 security scanners passed. Review the Security Audits panel on this page before installing in production.