- Home
- Skills
- Research & Analysis
- assumption audit
assumption audit
Audit unproven claims and loaded premises to separate correlation from causation in technical decision-making.
Free
Works with the AI Agents you already use
assumption audit
Example session with this skill installed
Audit this claim: "Our new caching layer caused the 20% drop in latency." Latency fell from 100ms to 80ms, but we also upgraded the instance types at the same time.
- Read your context and instructions
- Compiled the assumption audit
I have audited the causal claim regarding the latency reduction. The analysis separates the infrastructure change from the software optimization.
Assumption Audit: Latency Reduction
| Field | Content |
|---|---|
| Original Question | "Our new caching layer caused the 20% drop in latency." |
| Neutral Question | "What contributed to the 20% drop in latency, and to what extent did the caching layer versus the instance upgrade play a role?" |
| Assumption Ledger | 1. Causal Link: The cache is the primary driver of the 20ms delta. 2. Baseline Stability: No other concurrent network or traffic changes occurred. 3. Performance Parity: The new instance types provide identical single-thread performance to the old ones (likely false). |
| Conditional Answer | The claim that the cache caused the drop is currently unproven. While a 20ms reduction is observed, the simultaneous instance upgrade introduces a confounding variable. If the new instances have higher clock speeds or faster I/O, they may account for a significant portion of the gains. |
| Deciding Unknown | Comparative latency metrics from a staging environment where only one variable (cache or instance type) is changed at a time. |
Next steps
- Review the CPU and I/O specifications of the new instance types compared to the previous baseline.
- Run a synthetic benchmark on the new instances with the caching layer disabled to isolate the hardware delta.
- Check CloudWatch or monitoring logs for the exact timestamp of both changes to see if the drop was staggered or instantaneous.
Connects securely to your tools. The creator never sees your data.
About this skill
Technical decisions often fail because they are built on unverified premises, loaded questions, or loose definitions of success. This skill intercepts flawed inquiries by auditing the underlying logic before you commit to a solution. It prevents the common pitfall of solving for the wrong variable by identifying unproven causal claims and undefined metrics like "best" or "guaranteed."
What it does
- Neutralizes bias by reframing loaded premises into objective, testable questions.
- Identifies critical assumptions including definitions, causal claims, and time-bound constraints.
- Evaluates evidence by determining what specific data would support or weaken a specific claim.
- Separates correlation from causation to ensure associations are not mistaken for direct effects.
- Delivers conditional answers that highlight exactly which missing facts prevent a final verdict.
How it works
- Define the decision by quoting the original question and providing a neutral alternative.
- Build the assumption ledger to isolate variables that could materially change the outcome.
- Verify evidence for each high-impact assumption, checking for supportive versus weakening data.
- Generate the conditional response based on which premises passed or failed the audit.
Frameworks & tools
This skill is platform-agnostic and functions within any LLM environment or IDE. It is modeled after deep-research frameworks focused on logic deconstruction and analytical rigor.
Why this beats prompting it yourself
Writing a manual prompt to check for bias often results in the LLM simply agreeing with the user's premise or listing trivial uncertainties. This skill enforces a structured audit ledger that forces the agent to reject implied improvements and identify missing causal links.
Use cases
- Post-mortem analysis to verify if a specific deployment actually caused a performance spike.
- Product strategy to audit claims about "market-leading" features or user behavior.
- Technical troubleshooting to separate observed system associations from actual root causes.
Known limitations
Does not activate for basic arithmetic or well-defined math requests. It will not diagnose human motives or medical conditions. If sources are missing, it provides an audit plan rather than facts.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Research & Analysis.
Skips well-defined arithmetic. Does not diagnose motives or people. Requires source data to provide a factual verdict; otherwise, outputs an audit plan.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 24 days ago
- Free with an Agensi account
Permissions
No special permissions declared or detected