compensating-complexity-auditor
Skill0 starsChanged 7 months agoMITView on GitHub
Audit system prompts and agent pipelines for compensating complexity — scaffolding, procedural hacks, and duct tape built around previous model limitations that should be tested for deletion before a model upgrade. Use when upgrading models, reviewing system prompts, or preparing for model migration. Trigger on "audit prompt", "compensating complexity", "model upgrade prep", "prompt scaffolding", "what can I delete".
In this repository24 other files it gives its agents.
- aarSkill
- agent-cost-modelSkill
- banana-makerSkill
- bitter-lesson-scorecardSkill
- boot-tax-monitorSkill
- context-fork-guideSkill
- context-hygieneSkill
- counterargument-stress-testSkill
- dark-code-auditSkill
- executive-briefingSkill
- failure-asymmetrySkill
- failure-postmortemSkill
- file-intelSkill
- geopolitical-signal-enricherSkill
- get-api-docsSkill
- gh-reviewSkill
- model-routerSkill
- optimize-descriptionSkill
- prompt-rewriterSkill
- skill-auditSkill
- skill-maintenanceSkill
- spec-gap-detectorSkill
- token-burn-auditorSkill
- what-am-i-forgettingSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Ask the user to provide ONE of:
- Distinctive: line 17
If the user says "audit this project", read the project's CLAUDE.md and any system prompts found in the codebase.
- Distinctive: line 69
Every component in the input must appear in the classification table (no items skipped)
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Compensating Complexity Auditor
Analyze a system prompt or agent pipeline configuration line by line. Categorize every component and recommend deletion tests.
Phase 1: Intake
Ask the user to provide ONE of:
- A system prompt (paste or file path)
- A pipeline/agent configuration
- A CLAUDE.md or similar instruction file
If the user says "audit this project", read the project's CLAUDE.md and any system prompts found in the codebase.
Phase 2: Line-by-Line Classification
Walk through every instruction, constraint, or configuration block. Classify each as:
| Category | Definition | Action |
| Outcome Logic | Defines WHAT the system should achieve | KEEP — this is the core |
| Constraint | Business rule, compliance, safety, or quality gate | KEEP — these survive any model |
| Scaffolding | Step-by-step procedures that encode HOW to do something the model can figure out | TEST FOR DELETION |
| Duct Tape | Workarounds for specific model failures (hallucination patches, format hacks, retry logic for known weaknesses) | TEST FOR DELETION |
| Coordination | Multi-agent routing, handoff logic, tool selection rules | EVALUATE — some may be scaffolding |
Output a table with: line/section reference, classification, confidence (high/medium/low), and rationale.
Phase 3: Deletion Test Plan
For every SCAFFOLDING and DUCT TAPE item:
- State what the item compensates for
- Propose a deletion test: remove it, run against the target model, measure specific output quality
- Rate deletion risk: LOW (safe to try), MEDIUM (test carefully), HIGH (has downstream dependencies)
- Suggest what to REPLACE it with (often: nothing — just the outcome spec)
Phase 4: Summary Report
Output a structured report:
## Audit Summary
- Total components analyzed: N
- Outcome Logic: N (keep)
- Constraints: N (keep)
- Scaffolding: N (test for deletion)
- Duct Tape: N (test for deletion)
- Coordination: N (evaluate)
## Complexity Ratio
- Essential complexity: X% (outcome + constraints)
- Compensating complexity: Y% (scaffolding + duct tape)
## Top 3 Quick Wins (30-min deletion tests)
1. ...
2. ...
3. ...
## Deletion Test Backlog (prioritized)
...Verification
- Every component in the input must appear in the classification table (no items skipped)
- Classifications must include rationale, not just labels
- Deletion tests must be specific and measurable, not generic "test it"
- Quick wins must be genuinely achievable in 30 minutes
Source
Nate's Newsletter (2026-04-01): "Every workaround you built for the last model is now breaking the next one" — compensating complexity framework for model step-changes.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.