agentleFS
bakhod1rbakhod1r/awesome-agents/.claude/skills/cost-review/SKILL.md

cost-review

Skill0 starsChanged 3 months agoMITView on GitHub

Review a design or a running system for cost. Use before committing to an architecture, when a bill jumps, or during quarterly optimisation.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    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
  1. Distinctive:

    Every row: a number, a mechanism, an honest risk.

    line 58
  2. Distinctive:

    No recommendation without a measured saving.

    line 75
  3. Distinctive:

    Never trade away a reliability guarantee without saying so explicitly and loudly.

    line 76
  4. Distinctive:

    Verify in the following billing period.

    line 77
  5. Distinctive:

    Check for a hard limit before downsizing anything — quotas, connection pools, licences.

    line 79

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Cost Review

Procedure

  1. Attribute. Break spend by team, service, and environment. Untagged spend is finding number one — you cannot optimise what you cannot attribute.
  2. Find the drivers. Rank by absolute spend, then look for the usual suspects:

    DriverTypical cause
    Cross-zone / egress trafficChatty services split across zones; no locality awareness
    Storage classEverything on hot storage forever; no lifecycle policy
    Idle capacityNon-production running nights and weekends
    Over-provisioningSized for a peak that never arrived, never revisited
    Managed service premiumPaying for an operator you do not need at this scale
    Log and metric volumeDebug logging left on; high-cardinality labels
    LLM tokensNo caching, oversized model for the task, unbounded context
  3. Compute unit economics. Cost per request, per tenant, per model call. Totals hide everything — a bill that doubled while traffic tripled is a win.
  4. Model at scale. What does this cost at 10x? A design that is cheap today and ruinous at scale is rejected now, while changing it is still cheap.
  5. Recommend with numbers. Each item: measured monthly saving, effort, and the performance or reliability risk it introduces.

Output

| Finding | Monthly cost | Proposed change | Saving | Risk | Effort |

Worked example

Weak — unactionable, and quietly dangerous:

- NAT Gateway costs are high, consider reducing traffic.
- Instances look oversized, recommend downsizing.
- Consider reserved instances for savings.

Nobody can act on these, and "downsize" with no risk statement is how a cost review causes an outage.

Strong:

FindingMonthlyChangeSavingRiskEffort
NAT Gateway data processing: 41 TB/mo. 78% is S3 traffic routed through NAT.$4,920Add S3 Gateway VPC endpoint$3,840 (verified: gateway endpoints are free, traffic bypasses NAT)None. Route table change, reversible in minutes.1h
14 non-prod RDS instances running 24/7$6,200Stop 19:00–07:00 and weekends via scheduler$4,100 (66% of hours)Overnight CI jobs on 2 of the 14 — exclude those.4h
api service: p99 CPU 12%, sized m5.4xlarge × 9$3,740→ m5.xlarge × 9, keep replica count$2,800Real. 4x less burst headroom. Verified against the last Black Friday peak: 34% CPU at 4.1x normal traffic. Still 3x headroom after the change. Requires a load test before rollout.1d + load test
CloudWatch: 2.1 TB logs/mo, 61% is DEBUG from one service$1,480Set that service to INFO; 7-day retention on debug streams$900Lose debug detail on that service; it has tracing, so diagnosis is unaffected.2h
Total$11,640/mo

Every row: a number, a mechanism, an honest risk. The third row is the important one — it is the biggest behavioural change and it names the load test as a precondition instead of pretending the risk is zero.

Anti-patterns

Anti-patternWhy it fails
Recommendation with no measured savingUnprioritisable. The 5% item gets done, the 40% item does not.
Trading reliability for cost silentlyThe outage costs more than the year of savings
Optimising the total instead of the unit costPunishes growth; rewards shrinking the business
Reserved commitments without utilisation dataLocks in today's mistake for one to three years
Findings sent to financeFinance cannot change a route table. Send to the owning team.
Savings never verifiedHalf of claimed savings do not appear on the next bill

Rules

  • No recommendation without a measured saving. Estimates are labelled as estimates.
  • Never trade away a reliability guarantee without saying so explicitly and loudly.
  • Verify in the following billing period. Unverified savings do not count.
  • Route findings to the team that can act, not to finance.
  • Check for a hard limit before downsizing anything — quotas, connection pools, licences.

Done when

  • Attribution coverage stated; untagged spend quantified.
  • Findings ranked by absolute saving, not by ease.
  • Every row has a saving, a risk, and an effort estimate.
  • Unit economics reported alongside totals.
  • Cost at 10x scale modelled for design reviews.
  • A verification date is set for the following billing period.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.