Skip to content
Skills Library
DevOps & cloud · Analysis

ci-improve

/ci-improve

Reduce CI wait or compute by measuring the critical path, removing repeated work, selecting affected tests, and balancing shards. Use for a CI performance improvement with evidence before and after.

Repository
ci-skills ↗
Licence
MIT
Last repository change
11.11.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
1
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 11.11.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

Reduce CI wait or compute by measuring the critical path, removing repeated work, selecting affected tests, and balancing shards. Use for a CI performance improvement with evidence before and after.

Original author description

Reduce CI wait or compute by measuring the critical path, removing repeated work, selecting affected tests, and balancing shards. Use for a CI performance improvement with evidence before and after.

How it works

  1. Establish the baseline
  2. Remove the cause
  3. Prove the change
  4. Deliver
Original SKILL.md headings · 73 lines, 594 words
SKILL.md preview
---
name: ci-improve
description: Reduce CI wait or compute by measuring the critical path, removing repeated work, selecting affected tests, and balancing shards. Use for a CI performance improvement with evidence before and after.
---

# Improve CI

Reduce the user's stated metric. Preserve test coverage, release proof, and trust
boundaries. Read the repository's rules and current workflows first.

## Establish the baseline

Trace a representative run from its event to its final required check. Record
source SHA, event, runner class, cache state, attempts, job IDs, and timestamps.
Separate queue wait, setup, test execution, job duration, critical-path elapsed
time, and summed runner time. A fast final gate can still wait on slow upstream
jobs. Worker phase totals can overlap; do not add them as wall time.

Pick the bottleneck from that trace. Read its callers, inputs, outputs, and
consumers before changing it. Use existing reports and runners before adding
instrumentation. Save the baseline and raw samples without secrets or user data.

## Remove the cause

Choose the smallest change supported by the baseline:

- Reuse an existing output instead of producing it twice. Prove identical input
  identity and preserve scan or build contents. For example, pass a current-job
  SBOM to the scanner that would otherwise catalog the same image again.
- Select work from its actual dependency and fixture inputs. Unknown imports,
  missing history, deleted files, and parse failures should run the full affected
  suite. Include workflow, toolchain, lockfile, and shared configuration inputs.
- Avoid allocating a runner or installing tools for work the plan omits. Reuse
  the saved plan so setup and execution agree.
- Parallelize independent work on the critical path. Preserve database isolation,
  serial suite constraints, and resource limits. Balance shards using measured
  file timings. Check that their union includes every intended test once.
- Reuse caches only when keys bind all relevant inputs. Separate trusted cache
  writers from untrusted readers. Measure restore and save overhead as well as
  hits. Keep cold and warm results separate.
- Keep short coordination jobs from waiting behind long builds when the runner
  pool supports it. Measure queue delay separately from execution.

Moving or deleting a required suite changes coverage policy. Obtain the owner's
authorization when the task does not already grant it. A raised performance
budget is a tradeoff to record, not a speed improvement. A retry is not a fix for
a reproducible test or infrastructure defect.

## Prove the change

Compare the same source, runner class, event, and cache state when possible.
List every difference when they cannot match. Keep individual samples and report
sample counts. For noisy trials, choose an acceptance threshold before measuring,
repeat baselines, alternate baseline and candidate runs, and compare full outputs.
Reject a result that falls inside baseline noise.

Verify the behavior at risk: test selection on shared inputs and deletes, shard
coverage, output identity, failure propagation, cache invalidation, or exact-source
release proof. Run the narrow repository checks that cover the change.

Report elapsed time and summed runner time together for parallelism changes.
Faster feedback can use more compute. Do not price runner-seconds as dollars
without the actual billing model.

## Deliver

Save the change, a short result, and an evidence table with source links, units,
before/after values, sample size, and limits. Label each result as a hosted run,
local measurement, historical replay, or model. Keep inconclusive trials and
reliability fixes separate from speed results. Never add overlapping savings or
generalize one lane's result to the whole pipeline.

If asked for a post or chart, use measured comparisons with direct labels and a
zero baseline. Put the sample limit in the post text, not on the image. Keep estimates visibly separate.
Draft copy and portable instructions without internal names, keys, or private
evidence links. Publishing requires the user's requested destination and scope.

Versions

The database has no other skills named ci-improve.

Database snapshot: 13.11.2026 · package version f7571d3854e8 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100optimise-ci
    Reduce local testing and continuous integration duration or compute use from measured bottlenecks, while preserving required verification and release gates. Use when designing or improving test execution, CI pipelines or deployment verification.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.