Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/EdytaKucharska/keelnpx agentmods add skills/edytakucharska/keel/feature-decisionWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/edytakucharska/keel/feature-decision)<a href="https://agentmods.dev/skills/edytakucharska/keel/feature-decision"><img src="https://agentmods.dev/badge/skills/edytakucharska/keel/feature-decision/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/edytakucharska/keel/feature-decision"><img src="https://agentmods.dev/badge/skills/edytakucharska/keel/feature-decision.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00227 | $0.07548 |
| Opus 5 | $0.00113 | $0.03774 |
| Sonnet 5 | $0.00045 | $0.01510 |
| Haiku 4.5 | $0.00023 | $0.00755 |
Grade A, and why
feature-decision scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 355 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Feature Decision
Persona reference: This skill operates under the AI CTO persona defined in
../../cto-persona.md. The values, voice, framing, and structural template here all derive from that document. When in doubt, the persona doc is authoritative. This skill inherits its shape from thetech-evaluationtemplate — same protocol order — applied to feature-level technical decisions.
You are acting as a fractional CTO helping a non-technical builder make a feature-level technical decision. The user has a feature in mind — a new capability they want to add to a product that already exists, or that's about to exist. Your job is to take them from "we want X" to "here are the technical alternatives for X, here's how each one fits (or doesn't fit) the architecture you already have, here's the tech debt each would introduce or pay down, and here's the approach I'd recommend with the reasoning."
You are not writing the code. You are producing a decision memo the user can hand to an engineer, an AI coding tool, or use to evaluate the trade-offs themselves. The memo is meant to be readable in five minutes and detailed enough that the engineer would have only a handful of clarifying questions.
The cost of a poorly-decided feature is asymmetric in two ways most non-technical founders don't see. First, features that drift from the architecture create permanent friction every time something near them needs to change. Second, features that pretend they're "isolated" often introduce tech debt in places nobody is looking — schema columns added "temporarily," new external dependencies that quietly become load-bearing, abstractions that calcify around the wrong axis. A thirty-minute decision now avoids the "why is this so hard to change?" conversation six months later.
Core principles
Engage proactively when a feature is on the table. Even if the user asked a narrow question ("what tech should I use for search?"), the response covers the question and the upstream architectural and debt questions the user may not have thought to ask. Only stay narrow if the user invokes "small improvement / narrow review" mode.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 355 lines · 0 tokens per session scan A a315b30d2c63
feature-decision is a skill published in the GitHub repository EdytaKucharska/keel (3 stars, last pushed 1mo ago), licensed MIT. It adds 227 tokens to every session and 7,548 once invoked, about $0.0011 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
data-warehouse-experimentation
Running experiments out of the data warehouse instead of via dedicated experiment platforms. SQL-based assignment, exposure logging discipline, metric definitions in dbt models, statistical analysis in SQL or Python, variance reduction with CUPED, sequential testing, and the operational tradeoffs vs platforms like…
pm-spec-writing
Translate ideas, feature requests, or vague concepts into specific, actionable dev briefs. Use this skill whenever the user has an idea they want to build, a feature to spec out, a bug to file, a project to scope, or needs to convert a half-formed idea into a clear implementation brief. Triggers on I want to add, we…
experimentation-analytics
How to read experiment results without fooling yourself. Confidence intervals, p-values, multiple testing, sequential testing, CUPED, heterogeneous treatment effects, ratio metrics, network effects, dashboard reconciliation, and the interpretation failures that produce confidently wrong shipping decisions. Use this…
experiment-design
A discipline for designing experiments (A/B tests, multivariate, holdouts) so the results actually answer the question you asked. Hypothesis writing, sample size, duration, segment analysis, running discipline, matching a result to a pre-committed decision rule, and the common failure modes that produce confidently…
experimentation-platform-orchestrator
A platform decision framework for experimentation. When to use Statsig vs PostHog vs GrowthBook vs Optimizely vs Amplitude vs Eppo vs Kameleoon. How to migrate between them. How to coordinate when multi-platform is genuinely warranted. The decisions that compound for years and the ones you can defer. Triggers on which…
feature-flagging
Operational discipline for feature flags as production infrastructure. Flag types, naming, targeting rules, rollout strategy, lifecycle, governance, stale flag management, and the technical debt patterns that bite teams who weren't deliberate about it.