JAIPilot — Ship better Java with your coding agent.

JAIPilot helps Codex and Claude Code research maintainer intent, clean, test, review, safely upgrade dependencies and JDKs, and optimize real Java repositories without drifting into unnecessary code or unproved changes.

By default, your agent runs Java builds, tests, analysis, profiling, and benchmarks on a ready remote Java machine whenever the task does not require laptop-only access or state.

JAIPilot does not replace your coding agent or add another AI. It gives your agent focused Java workflows, remote compute, and one rule: show evidence, not confidence.

Further reading: Preventing agent drift: A guide to shipping serious code via vibe-coding explains the engineering principles behind JAIPilot's bounded, evidence-first workflows.

Install and run

Codex

JAIPilot is packaged for the official OpenAI plugin directory. Once the listing is approved, install it there to get all eight local skills and the optional OAuth remote MCP together. The same public MCP remains available as a direct connection:

codex mcp add jaipilot --url https://api.jaipilot.com/functions/v1/jaipilot/mcp

Claude Code

/plugin marketplace add JAIPilot/jaipilot
/plugin install jaipilot@jaipilot

Any MCP client

If your client supports remote HTTP MCP servers and OAuth, connect directly to JAIPilot without installing a plugin or running a local process:

https://api.jaipilot.com/functions/v1/jaipilot/mcp

Clients that use the common mcpServers configuration shape can add:

{
  "mcpServers": {
    "jaipilot": {
      "type": "http",
      "url": "https://api.jaipilot.com/functions/v1/jaipilot/mcp"
    }
  }
}

Complete the OAuth sign-in when the client first connects. Clients with MCP Skills support can discover the JAIPilot skills directly; tool-only clients can load the same versioned instructions through the read-only skill_get tool. Remote execution remains opt-in and requires approval before an exact committed repository revision is uploaded.

GitHub Marketplace

Install the JAIPilot GitHub App to repair failed Dependabot and Renovate pull requests for Maven and Gradle repositories.

See what JAIPilot changed

JAIPilot does not hide its value behind skill routing or build logs. Every meaningful skill milestone uses the same restrained, one-line grammar:

**JAIPilot · <capability>** — <completed outcome>; <strongest fresh proof>.

For example: JAIPilot · Test generation — 7/7 focused tests passed; fresh line and branch coverage reached 100%.

The update appears only after completed work, never for intent or routing. When completion coincides with the final response, the same line becomes its outcome lead and is immediately followed by one consolidated impact section. Nested skills contribute to that section instead of printing a stack of banners.

For example, a measured Petclinic test-generation run ends with evidence in this shape:

JAIPilot · Test generation — 10 useful tests added; changed-method line and branch coverage reached 100%.

JAIPilot impact

• Test generation: 75 → 85 tests (+10); changed-method line and branch coverage 0% → 100% (+100 percentage points)
• Evidence: ./mvnw -q clean verify — 85/85 tests passed

Impact uses fresh, same-scope before/after evidence already useful to the task. When a comparison is not available, JAIPilot says not measured; when the value is a maintainer decision or rejected unsafe change, it describes that outcome without inventing hours saved, risk percentages, or a counterfactual “without JAIPilot” score.

Example prompts

Open a Java repository and ask JAIPilot the same way you would ask a teammate. Each example shows the skills JAIPilot selects and why that workflow helps.

Check whether an issue is worth picking up

Prompt

Can you look into issue #184 and tell me whether I should work on it, or if someone is already
handling the same problem?

Skills and flow: jaipilot-maintainer-intent researches the issue, related work, contribution rules, and code history before recommending whether to proceed, join existing work, ask, wait, or stop.

Why it helps: You avoid duplicating another contributor's work or building a fix maintainers are unlikely to accept.

Get a branch ready to merge

Prompt

I've finished the changes on this branch. Can you make sure they're ready to merge?

Skills and flow: jaipilot-optimize-java coordinates bounded cleanup and improvement, jaipilot-generate-tests fills meaningful test gaps, and jaipilot-review-diff reviews the complete result. jaipilot-fast-execution and jaipilot-remote-java accelerate substantial verification when safe and useful.

Why it helps: You get a smaller, reviewed change with missing tests added and the repository's real checks run before handoff.

Upgrade a framework and its dependencies

Prompt

Can you upgrade this Spring Boot service and its outdated dependencies while keeping Java 17
support?

Skills and flow: jaipilot-maintainer-intent first checks repository history, active or prior upgrade work, maintainer direction, and established compatibility constraints. The supported upgrade then uses jaipilot-clean-java to inventory and modernize the requested dependency and build paths. jaipilot-openrewrite is selected only when a repeated type-aware migration justifies it, followed by jaipilot-generate-tests, jaipilot-review-diff, jaipilot-fast-execution, and jaipilot-remote-java as applicable.

Why it helps: The upgrade follows the repository's intended direction and avoids duplicating or conflicting with existing work. Accepted changes remain compatible, reversible, and verified instead of becoming a broad version bump with hidden runtime or transitive-dependency breakage.

Fix a failing dependency-bot update

Prompt

Dependabot's Jackson update is failing CI. Can you figure out why and fix it?

Skills and flow: jaipilot-maintainer-intent checks the bot PR, maintainer direction, related attempts, and the correct delivery path. The supported fix then uses jaipilot-clean-java or jaipilot-openrewrite, followed by jaipilot-generate-tests and jaipilot-review-diff. Substantial commands use jaipilot-fast-execution and jaipilot-remote-java when appropriate.

Why it helps: The agent fixes the actual compatibility problem on the right branch instead of merely forcing the bot's version change through CI.

Add tests for a service change

Prompt

Please add unit tests for my changes to OrderService, especially the validation and error cases.

Skills and flow: jaipilot-generate-tests maps the requested classes and useful cases, creates and executes safely independent test-class work in parallel, and reports fresh configured coverage or mutation evidence. jaipilot-fast-execution and jaipilot-remote-java accelerate the test and verification work when applicable.

Why it helps: You get behavior-focused tests across the complete requested scope, not hollow tests written only to inflate a coverage number.

Simplify a difficult module

Prompt

This payments module has become hard to follow. Can you remove dead code and simplify it without
changing its behavior?

Skills and flow: jaipilot-clean-java proves what can be removed or consolidated and keeps uncertain candidates. jaipilot-generate-tests covers concrete regression gaps, and jaipilot-review-diff checks the final patch. Verification routes through jaipilot-fast-execution and jaipilot-remote-java when useful.

Why it helps: The module becomes easier to maintain without speculative deletion, silent API breakage, or a cleanup that only looks smaller.

Review work before opening a pull request

Prompt

Can you review my current diff before I open a pull request?

Skills and flow: jaipilot-review-diff reads the complete Java and build change, checks behavior, compatibility, unnecessary code, and missing proof, then runs the applicable repository checks via jaipilot-fast-execution and jaipilot-remote-java when appropriate.

Why it helps: You get ranked, evidence-backed findings across the whole change rather than a shallow review of the most obvious file.

Speed up a slow endpoint

Prompt

The /orders endpoint gets slow for large accounts. Can you find the bottleneck and improve it?

Skills and flow: jaipilot-optimize-java bounds the change, jaipilot-clean-java profiles and tests measured hypotheses, jaipilot-generate-tests protects behavior, and jaipilot-review-diff checks the candidate. jaipilot-fast-execution and jaipilot-remote-java provide controlled execution and comparable measurements.

Why it helps: The retained optimization is tied to a real bottleneck and repeatable evidence, not a plausible-looking micro-optimization or one noisy timing.

Run a slow build faster

Prompt

Our full Gradle build takes about 20 minutes. Can you run it faster and tell me where the time goes?

Skills and flow: jaipilot-fast-execution sizes safe batching and native build parallelism for the available resources. jaipilot-remote-java offloads the work when the repository does not need laptop-only services or state.

Why it helps: You reduce wall time without skipping tests, weakening gates, or creating misleading results through unsafe concurrency.

The Codex plugin installs all eight skills locally and configures the optional hosted MCP. A direct MCP connection can instead discover the same versioned skills through skills/list, skills/get, digest-verified resources/read, and standard mcp/skill resources. Clients that do not yet promote server skills can use the read-only skill_get fallback. Signing in creates no upload or compute; remote execution starts only after the separate exact-repository upload consent described below. Claude Code retains its plugin path.

When remote execution is useful, your agent asks before uploading one exact committed Git revision. Approve the upload; if authentication is not already active, sign in when prompted. The agent uses git archive for that commit and handles integrity checking, private upload, execution, logs, and cleanup. You do not configure a VM, provide an AWS key, or copy files manually. Staged, unstaged, untracked, ignored, and .git content never enters the build, and remote files never return automatically.

If packaging or upload cannot be verified, JAIPilot does not start the build. Your agent must show the failing step instead of silently uploading a different source tree.

JAIPilot vs no JAIPilot

The original Petclinic PR already had a green build. JAIPilot reviewed that exact head and produced this companion change:

The important result is not simply “more tests.” Without JAIPilot, the build passed while the new behavior had zero coverage and unused code remained. With JAIPilot, the same behavior stayed green, the edge cases became executable tests, and production code became smaller.

The comparison uses the original PR head and JAIPilot's direct child commit, clean worktrees, the same ./mvnw -q clean verify command, and fresh JaCoCo 0.8.14 reports.

Measured performance: Apache Calcite

On skrcode/calcite at exact commit d3a5d8d, JAIPilot profiled Calcite's existing DefaultDirectedGraphBenchmark.removeAllVertices{10,50,90}Benchmark. The 50% workload attributed 31.1% of runnable samples to Collection.removeIf: the implementation scanned the complete global edge set once for every removed vertex.

The candidate changed two files (+28/-5), removed the repeated scans, and added behavior tests for the majority-removal and self-loop paths. Lower JMH scores are better:

Baseline and candidate ran on the same 4 CPU/8 GiB remote workspace with the same Temurin JDK 17, built JMH jar, command, and workload. Each row contains 21 measured observations: seven forks with three measured iterations per fork after warm-up. The identical focused command passed 15/15 tests before and after the production edit. A fresh exact-SHA :core:clean :core:check then completed 16,644 tests with 0 failures and 155 skips, and the tested remote diff matched the local candidate digest.

This is a controlled result for Calcite's existing graph-removal workloads, not a claim that every Java workload becomes faster.

Measured performance: OpenTelemetry Java

On skrcode/opentelemetry-java at exact commit 35636ae, JAIPilot found that immutable attribute sets were sorted by key name during construction but still used a full linear scan for every lookup. This matters at the default span limit of 128 attributes.

The three-file draft change preserves the small-set and first-four-entry fast path, then uses binary search for the rest. It also adds a large-set behavior test and a repository-native JMH benchmark. Lower values are better:

Baseline and candidate ran in the same 4 CPU/8 GiB remote workspace with Temurin JDK 21, the same JMH jar, command, warm-up, and workload. Each row has 21 observations. The new focused behavior test passed before and after the production edit; a clean :api:all:check passed all 147 tasks including Animal Sniffer, Checkstyle, Spotless, tests, and japicmp. The tested remote Git delta matched the local candidate digest. The Value workload still allocates about 16 B/op; JAIPilot reports the lookup-time win without claiming that allocation disappeared.

Measured performance: Micrometer

On skrcode/micrometer at exact commit 22207bf, JAIPilot found that adding or replacing one Tag or KeyValue went through temporary varargs and iterable merge machinery even though the backing arrays were already sorted.

The six-file draft change adds a bounded binary-search merge for the single-value overloads, behavior tests, and four workloads in Micrometer's existing JMH module:

Single-value insertion reduced median allocation by 17.6% and p95 allocation by 46.2%. Its median latency improved by only 6.7–8.9%, below JAIPilot's 10% shared-hardware threshold, so it is not presented as a speed win. The same-workspace experiment used 21 observations per workload; 92 focused tests and the final clean scoped build passed, with 1,132 tests, zero failures, and the remote production diff matching the local digest.

Measured database work: Petclinic

On the current Petclinic JDBC vet listing, JAIPilot found a real N+1 query path: six vets required eight SQL statements (2 + N). It kept the ordered vet query and replaced the per-vet specialty lookups with one joined association query.

The new tests lock vet ordering, specialty ordering, vets without specialties, duplicate links, shared specialty identity, empty data, and the exact two-statement ceiling. The final clean Maven build and JaCoCo report passed on a 4 CPU/8 GiB remote workspace, and the local and remote binary diff digests matched. This is deterministic query-count evidence; JAIPilot does not turn it into an invented latency claim.

Why teams use JAIPilot

• Less agent drift — changes stay bounded, lean, and aligned with the repository.
• Better Java code — remove proven waste, reduce complexity, improve tests, review risky diffs, and optimize measured bottlenecks.
• Verified modernization — upgrade the JDK, Maven or Gradle wrapper, build plugins, frameworks, BOMs, and direct dependencies to stable compatible versions in isolated, reversible batches.
• Real verification — use the repository's Maven or Gradle build, tests, coverage, architecture rules, and performance measurements.
• Remote-first execution — use disposable hardware with JDK 17, 21, and 25, Maven, and Gradle ready unless private networks, local services, secrets, hardware, or state require the laptop.
• Remote performance lab — profile and compare a bounded optimization on one 4 CPU/8 GiB workspace, with matching patch identity, raw observations, median, p95, and correctness evidence.
• Keeps dirty work private — remote execution accepts only an explicitly approved exact commit; staged, unstaged, untracked, ignored, and .git content stays on the laptop.
• Your agent stays in control — Codex or Claude Code chooses every edit and command and reports exactly what JAIPilot achieved.

More proven results

Additional acceptance runs used repository-native verification:

These are reproducible acceptance results, not claims that every repository will see the same coverage, code reduction, or speed.

MCP Registry

Clients that consume the official MCP Registry can discover io.github.JAIPilot/jaipilot. The one hosted endpoint serves the eight Java engineering skills directly and forwards only tools/list/tools/call to the existing bounded OAuth remote-execution service. Skill discovery and reads do not contact remote execution; tool calls retain its authorization and safety boundary.

Both skill catalogs are paginated five skills and then three so direct clients can load all eight. The direct MCP connection remains available alongside the combined Codex plugin submitted to the official OpenAI plugin directory.

Included skills

JAIPilot can work with repository-configured tools such as JaCoCo, PIT, ArchUnit, OpenRewrite, Checkstyle, PMD, SpotBugs, SonarQube, JMH, and JFR. It never weakens a quality gate merely to get a green result.

Remote build beta

The beta permits one active normal AWS CodeBuild attempt. Each attempt starts from a private exact commit archive, has no persistent cache or workspace, and terminates when its command exits, is cancelled, or reaches its hard timeout. The default profile is 2 CPU/4 GiB; 4 CPU/8 GiB is available for memory-heavy builds and profiling. Builds have public outbound internet access and never commit, push, or receive GitHub write credentials.

JAIPilot stays local when a corporate VPN, VPC-only dependency, private artifact service, internal database, unavailable secret, raw profiler artifact, machine-specific state, or other laptop-only resource is required. CodeBuild startup is variable and can take several minutes on a cold compute class, so this is batch-style remote execution rather than an interactive shell.

See Security, Privacy, Support, Terms, and the Changelog.

Licensed under the MIT License.