guardrails
Skill304 starsChanged 2 months agoMITView on GitHub
Ensure every success metric is paired with guardrail metrics and check for trade-offs before presenting improvements as wins. Guardrails protect against winning the metric game while losing the business game. Apply this skill whenever you define a metric, create a metric spec, document a KPI, report positive findings, present an improvement, analyze metric changes, investigate metric lifts, share good news about metrics, celebrate wins, or present results showing any metric increased. This skill is CRITICAL for preventing false wins where one metric improves at the expense of another (e.g., conversion rate up but average order value down, signup rate up but activation down, resolution time down but customer satisfaction down). Use this skill automatically when you see phrases like "X improved", "X is up", "X increased", "better performance on X", "we're seeing gains in X", or any analysis showing positive metric movement. Also apply when defining metrics alongside the metric-spec skill to ensure guardrails are specified upfront. Never present a metric improvement without running the guardrail check — an improvement with a degraded guardrail is a trade-off or net negative, not a win.
16 other projects use a version of this.
Adopted by
Most recently taken up by
- alexuraxin/retainertheir own version, first seen Nov 15, 2026
- meaningfy-ws/skillerytheir own version, first seen Nov 13, 2026
- the-inconvenience-store/skillstheir own version, first seen Nov 6, 2026
- meyverick/agentictheir own version, first seen Oct 30, 2026
- antrafa/skillstheir own version, first seen Oct 20, 2026
- irahardianto/awesome-agvtheir own version, first seen Oct 14, 2026
Used with
Projects with it also use evaluation, prototype and 2 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
NVIDIASkillevaluation
Evaluates accuracy of quantized or unquantized LLMs using NeMo Evaluator Launcher (NEL). Triggers on "evaluate model", "benchmark accuracy", "run MMLU", "evaluate quantized model", "run nel". Handles deployment, config generation, and evaluation execution. Always load when validating completed runs or reporting evaluation scores, including standalone launching-evals analysis. Not for quantizing models (use ptq), standalone serving (use deployment), or comparing completed baseline-vs-quantized results (use compare-results).
Also in 3 projects that use this
GoogleSkillprototype
Build a throwaway prototype to answer a design question. Use when the user wants to sanity-check whether a state model or logic feels right, or explore what a UI should look like.
Also in 3 projects that use this
GoogleSkillgrill-me
A relentless interview to sharpen a plan or design.
Also in 4 projects that use this
MicrosoftSkillhandoff
Compact the current conversation into a handoff document for another agent to pick up.
Also in 3 projects that use this
Other versions16 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 16
- irahardianto/awesome-agv★ 157same name
- googlicius/obsidian-steward★ 85same name
- bolasblack/claude-skills★ 4same name
- hajekim/agentic-design-patterns-skills★ 4same name
- alexuraxin/retainer★ 3same name
- dfrysinger/skills★ 3same name
- meaningfy-ws/skillery★ 3same name
- adewale/guardrails-skill★ 2same name
- Citadel-Cloud-Management/citadel-saas-factory★ 2same name
- antrafa/skills★ 1same name
6 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository65 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- always-compareSkill
- analysis-designSkill
- analyst-coreSkill
- archaeologySkill
- archive-analysisSkill
- auth-preflightSkill
- businessSkill
- causalSkill
- chart-to-driveSkill
- close-the-loopSkill
- codex-reviewSkill
- compare-datasetsSkill
- connect-dataSkill
- connect-snowflakeSkill
- context-compareSkill
- context-traceSkill
- data-inspectSkill
- data-mapSkill
- data-profilingSkill
- data-quality-checkSkill
- datasetsSkill
- distribution-profilerSkill
- evalSkill
- evaluate-graderSkill
- experiment-briefSkill
- experimentSkill
- exploreSkill
- exportSkill
- forecastSkill
- google-doc-exportSkill
- google-slides-exportSkill
- historySkill
- improve-contextSkill
- knowledge-bootstrapSkill
- log-correctionSkill
- maintain-contextSkill
- metric-specSkill
- metricsSkill
- monitor-evalsSkill
- notion-exportSkill
- paceSkill
- patternsSkill
- question-framingSkill
- question-routerSkill
- reliabilitySkill
- resume-pipelineSkill
- run-pipelineSkill
- runsSkill
- score-analysisSkill
- session-handoffSkill
- setup-dev-contextSkill
- setup-notionSkill
- setupSkill
- setup-slackSkill
- setup-snowflakeSkill
- srm-checkSkill
- stakeholder-communicationSkill
- stress-testSkill
- switch-datasetSkill
- The first 60 shown.
The file
Show its rules and text
We split this skill into the rules it gives an agent
14 rules. 0 are common, 14 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 15 edited versions we read keep these.
14
Distinctive
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 10
Ensure that every success metric is paired with at least one guardrail metric, and that positive findings are checked for trade-offs before …
- Distinctive: line 28
The rule: Never celebrate an improvement on a success metric without checking its guardrail(s).
- Distinctive: line 47
After specifying a metric using the Metric Spec skill, add a guardrail section:
- Distinctive: line 61
If no obvious guardrail exists, use customer satisfaction or support ticket volume as defaults
- Distinctive: line 63
Guardrail selection framework: When choosing guardrails, think about what could go wrong if you optimize the success metric aggressively.
- Distinctive: line 65
Quality vs. Quantity: If the success metric measures volume (signups, orders, sessions), check quality metrics (conversion rate, AOV, …
- Distinctive: line 66
Speed vs. Accuracy: If the success metric measures speed (time to resolution, checkout time), check accuracy/quality (error rate, reopen …
- Distinctive: line 67
Revenue vs. Experience: If the success metric measures revenue (RPU, monetization rate), check experience metrics (NPS, support tickets, …
- Distinctive: line 68
Short-term vs. Long-term: If the success metric measures immediate outcomes, check long-term health (retention, LTV, repeat purchase rate)
- Distinctive: line 71
First priority: Check if the guardrail can be proxied (e.g., if CSAT data doesn't exist, use support ticket volume or product return rate …
- Distinctive: line 72
Second priority: Document the missing guardrail as a known risk and recommend tracking it for future iterations
- Distinctive: line 78
Before presenting any statement like "[metric] improved by X%," run this check:
- Distinctive: line 104
When a guardrail is degraded:
- Distinctive: line 113
When guardrails are checked, add this section to the analysis report:
“Common” is measured: kept by at least 6 in 10 of the 15 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Skill: Guardrails Awareness
Purpose
Ensure that every success metric is paired with at least one guardrail metric, and that positive findings are checked for trade-offs before being presented as wins.
When to Use
Apply this skill in two situations:
- When defining metrics — after using the Metric Spec skill, check whether the metric has a guardrail pair
- When reporting positive findings — before presenting any improvement, check whether a related guardrail metric degraded
Instructions
What Are Guardrails?
A guardrail metric is a metric you don't want to degrade while optimizing a success metric. Guardrails protect against winning the metric game while losing the business game.
SUCCESS METRIC: The metric you're trying to improve
GUARDRAIL: The metric that must not get worseThe rule: Never celebrate an improvement on a success metric without checking its guardrail(s). An improvement with a degraded guardrail is a trade-off, not a win.
Common Guardrail Pairs
| Success Metric | Guardrail(s) | Why |
| Conversion rate | Average order value, Return rate | Aggressive discounts inflate conversion but erode margin and invite returns |
| Signup rate | Activation rate, 7-day retention | Lowering the signup bar brings in unqualified users who churn immediately |
| Revenue per user | User satisfaction (NPS/CSAT), Support ticket volume | Monetization pressure degrades experience |
| Feature adoption | Core workflow completion, Session duration | Forcing feature usage may disrupt existing workflows |
| Time to complete (speed) | Error rate, Quality score | Rushing degrades accuracy |
| Cost reduction | Quality, Customer satisfaction | Cutting costs can degrade service |
| Engagement (DAU, sessions) | Revenue per user, Churn rate | Engagement tricks (notifications, dark patterns) don't translate to value |
| Support resolution time | Customer satisfaction, Reopen rate | Fast close ≠ good close if tickets reopen |
How to Apply
When Defining Metrics
After specifying a metric using the Metric Spec skill, add a guardrail section:
### Guardrails
| Guardrail Metric | Acceptable Range | Check Frequency |
|-----------------|-----------------|-----------------|
| [guardrail 1] | [must stay above X / must not increase by >Y%] | [same cadence as success metric] |
| [guardrail 2] | [threshold] | [cadence] |Rules for selecting guardrails:
- At least one guardrail per success metric
- The guardrail should measure a different dimension of value (e.g., if success is quantity, guardrail is quality)
- The guardrail must be measurable with available data
- If no obvious guardrail exists, use customer satisfaction or support ticket volume as defaults
Guardrail selection framework: When choosing guardrails, think about what could go wrong if you optimize the success metric aggressively. Ask yourself:
- Quality vs. Quantity: If the success metric measures volume (signups, orders, sessions), check quality metrics (conversion rate, AOV, retention)
- Speed vs. Accuracy: If the success metric measures speed (time to resolution, checkout time), check accuracy/quality (error rate, reopen rate, customer satisfaction)
- Revenue vs. Experience: If the success metric measures revenue (RPU, monetization rate), check experience metrics (NPS, support tickets, churn)
- Short-term vs. Long-term: If the success metric measures immediate outcomes, check long-term health (retention, LTV, repeat purchase rate)
When guardrail data is unavailable:
- First priority: Check if the guardrail can be proxied (e.g., if CSAT data doesn't exist, use support ticket volume or product return rate as a quality signal)
- Second priority: Document the missing guardrail as a known risk and recommend tracking it for future iterations
- Do NOT skip the guardrail check entirely — use whatever quality signals are available, even if imperfect
- Example: "Conversion rate improved 20%, but we lack return rate data to confirm quality. As a partial guardrail check, AOV increased 5% (suggesting legitimate demand, not discount-driven conversions). Recommend adding return rate tracking."
When Reporting Positive Findings
Before presenting any statement like "[metric] improved by X%," run this check:
GUARDRAIL CHECK
□ Identified guardrail metric(s) for [success metric]
□ Computed guardrail metric(s) over the same time period
□ Compared guardrail to baseline / acceptable range
□ Result: [CLEAR / TRADE-OFF / DEGRADED]Verdicts:
| Verdict | Guardrail Status | How to Present |
| CLEAR | Guardrail stable or improved | Present the improvement as a win |
| TRADE-OFF | Guardrail slightly degraded (<10% relative) | Present the improvement AND the trade-off: "Conversion improved 15%, but AOV decreased 5%. Net revenue impact is +8%." |
| DEGRADED | Guardrail significantly degraded (>10% relative) | Do NOT present as a win. Present as: "Conversion improved 15%, but return rate doubled. The net impact may be negative — further investigation needed." |
Note on thresholds: The 10% threshold is a starting point, not a universal rule. Adjust based on context:
- High-impact guardrails (e.g., customer churn, security incidents): Even 5% degradation may be DEGRADED
- Volatile guardrails (e.g., daily engagement metrics): 10-15% degradation may be normal variance, use TRADE-OFF
- Revenue-critical guardrails (e.g., AOV, margin): Compute absolute dollar impact to determine severity, not just percentages
- When in doubt: Err on the side of caution — flag smaller degradations and let stakeholders decide if the trade-off is acceptable
Guardrail Escalation
When a guardrail is degraded:
- Quantify both sides — compute the success metric gain AND the guardrail loss in the same units (usually dollars or users)
- Compute net impact — is the gain larger than the loss?
- Flag uncertainty — guardrail degradation often has delayed effects (e.g., returns take weeks to materialize, churn shows up months later). Note this.
- Recommend investigation — "The conversion improvement looks positive, but the return rate increase warrants investigation before concluding this is a net win."
Output Format
When guardrails are checked, add this section to the analysis report:
## Guardrail Check
| Success Metric | Change | Guardrail | Change | Verdict |
|---------------|--------|-----------|--------|---------|
| [metric] | +X% | [guardrail 1] | [no change / +Y% / -Z%] | CLEAR / TRADE-OFF / DEGRADED |
| | | [guardrail 2] | [change] | [verdict] |
**Net assessment:** [The improvement is real / The improvement comes with a trade-off / The improvement may be net negative]Examples
Example 1: Clear Win
## Guardrail Check
| Success Metric | Change | Guardrail | Change | Verdict |
|---------------|--------|-----------|--------|---------|
| Checkout conversion | +12% | Avg order value | +2% (stable) | CLEAR |
| | | Return rate | -1% (improved) | CLEAR |
**Net assessment:** The conversion improvement is a genuine win. Both guardrails are stable or improving.Example 2: Trade-Off
## Guardrail Check
| Success Metric | Change | Guardrail | Change | Verdict |
|---------------|--------|-----------|--------|---------|
| Signup rate | +25% | 7-day activation | -8% | TRADE-OFF |
| | | 30-day retention | -3% (within normal range) | CLEAR |
**Net assessment:** The signup rate improvement is partially offset by lower activation. The new signups are less qualified. Recommend segmenting the new signups to identify which acquisition channel is bringing lower-quality users.Example 3: Degraded Guardrail
## Guardrail Check
| Success Metric | Change | Guardrail | Change | Verdict |
|---------------|--------|-----------|--------|---------|
| Resolution time | -40% (faster) | Reopen rate | +85% | DEGRADED |
| | | CSAT score | -22% | DEGRADED |
**Net assessment:** The faster resolution time is coming at the cost of quality. Tickets are being closed prematurely and reopened, and customer satisfaction has dropped significantly. This is NOT a net improvement. Recommend reverting the process change and investigating sustainable ways to reduce resolution time.Anti-Patterns
- Never report a success metric improvement without checking guardrails — a 20% conversion lift with a 30% return rate increase is not a win
- Never define a metric without at least one guardrail — every metric can be gamed; guardrails prevent it
- Never dismiss a small guardrail degradation — small degradations compound and may have delayed effects (churn shows up months later)
- Never use the same metric as both success and guardrail — they must measure different dimensions of value
- Never skip the net impact calculation — "conversion up, returns up" is not actionable without knowing which effect is larger
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.

