- Home
- Skills
- Code Quality & Review
- Reason First
Reason First
First-principles reasoning for software requirements, debugging, performance and design tradeoffs. Clarify goals, separate facts from assumptions, test what matters, and verify outcomes with evidence.
Free
Works with the AI Agents you already use
Reason First
Example session with this skill installed
Use Reason First to review this proposed optimization. This is a synthetic exercise, not production telemetry.
An endpoint takes 900 ms in one sequential, non-overlapping request: authentication 70 ms, PostgreSQL queries 80 ms, and a required external service call 750 ms. The target is 500 ms. A teammate proposes replacing PostgreSQL with Redis. PostgreSQL must remain the source of truth, and the response contract must stay the same. Recommend the next step; do not change code.
- Read your context and instructions
- Compiled the reason first
The supplied timings do not support replacing PostgreSQL as the first step toward the 500 ms target. Investigate the external call first and retain PostgreSQL as the source of truth.
| Component | Supplied duration | Share of total |
|---|---|---|
| Authentication | 70 ms | 7.8% |
| PostgreSQL queries | 80 ms | 8.9% |
| Required external service | 750 ms | 83.3% |
| Total | 900 ms | 100% |
What the numbers establish: even eliminating all database latency would leave 820 ms, above the target. The maximum improvement from that change alone is 80 ms, or about 8.9%. A cache would also need a freshness and invalidation policy, so its effect on the response contract cannot be assumed.
Next investigation: confirm that the same non-overlapping timing breakdown holds across representative requests and load. Trace the external call to separate connection setup, server processing, retries and payload transfer. Verify timeout and retry behavior before selecting a change.
If authentication and database time stay at 150 ms, the external-call portion must fit within 350 ms to meet the 500 ms target. Check whether the provider supports a faster operation, a smaller equivalent request, or a different endpoint with the same required semantics. Consider caching only if freshness rules permit it; do not silently relax the contract.
Verification: compare end-to-end latency and the external-call span under comparable load and inputs, including relevant latency percentiles, errors and response equivalence. Reject a candidate if it changes required semantics or fails the agreed latency criterion.
Limits: this conclusion uses the exercise's synthetic timings. No production measurements, benchmarks or code changes were performed, and no speedup is claimed. A single request does not establish tail-latency behavior.
Connects securely to your tools. The creator never sees your data.
About this skill
Reason First
Reason First is a lightweight reasoning workflow for software engineering. It helps an AI coding agent clarify the desired outcome, distinguish evidence from assumptions, test explanations, and choose a solution that fits the actual constraints.
What it helps with
- Clarifying ambiguous requirements and acceptance criteria.
- Investigating bugs with reproduction steps, source, logs and contracts.
- Identifying performance bottlenecks before proposing an optimization.
- Comparing design options by correctness, compatibility and overall cost.
- Verifying changes against the original problem and relevant invariants.
How it works
The agent defines the outcome, classifies the available evidence, builds a useful model, tests consequential assumptions, derives a solution, and verifies the result. Investigation depth scales with uncertainty and impact; routine mechanical changes stay lightweight.
The skill respects explicit technology choices and applicable workspace instructions. An analysis request stays within analysis; an implementation request proceeds through verification within the user's authorized scope. Missing measurements and unverified conclusions are stated explicitly.
What you receive
- A self-contained SKILL.md containing the reasoning workflow.
- agents/openai.yaml with the display name and short description.
No bundled scripts, dependencies, API keys, external accounts or paid Agensi tools are required by the package.
Using it
Ask your compatible agent to use Reason First on a requirement, bug, performance question or design decision. Provide the relevant context, constraints and evidence. The agent should deliver a concise recommendation, key rationale, verification evidence and material limitations.
Boundaries
This is instruction-based guidance. It does not provide a new model, run an independent analyzer, guarantee correctness, or guarantee performance gains. Source inspection, reproduction and benchmarks depend on the host agent's tools and the access you authorize. It does not expose a full internal reasoning transcript.
Instruction-based guidance; output quality depends on the host model and evidence. Reproduction, source inspection and benchmarks require suitable host tools and authorized access. No guaranteed correctness, performance improvement or fully autonomous verification.
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 any payment.
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
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 11 days ago
- Free with an Agensi account
Permissions
The package contains instructions and UI metadata only, with no scripts or network calls. It can analyze context supplied in chat. Repository inspection, command execution or verification use the host agent's existing tools and user-authorized access; no creator-operated service is required.