Add review.sh pre-push review entry point

[dougqh]

What Does This Do

Adds review.sh, a pre-push review entry point, as part of APMLP-2033 (tracked under APMLP-1964). The script lives in tooling/review.sh, with a symlink at the repo root so contributors can run ./review.sh.

It does three things:

• Picks the diff: it reviews the branch changes since the merge-base with upstream master. --base <ref> overrides that, and is also normalized to a merge-base.
• Runs the harness's code review headlessly: it uses claude -p or codex exec, with a read-only tool set so the run can't stop on a permission prompt or edit files. The review applies the AGENTS.md Review Guidelines and Critical constraints.
• Chains the repo extension /perf-review: --no-extensions skips it.

Like the rest of the plan, it is advisory: findings are printed, the script exits 0 after a review, and nothing is enforced. --out <file> saves the output for the eval comparison against a human baseline.

AGENTS.md's Review Guidelines and .claude/CLAUDE.md now point at ./review.sh.

Motivation

The 9/21 "Agent Code Review in Datadog SDKs" decision asks each tracer repo for a review.sh entry point, with review expectations kept in AGENTS.md. This builds on the existing perf-review skill.

Additional Notes

This is a draft while the interface is agreed across tracer repos. Open points:

• Shared interface with other tracer repos: the flags and the output format, which matter for comparing results across languages.
• Prompt vs built-in command: the code-review step currently sends a plain-language prompt rather than calling the harness's slash command (e.g. /review), so it works the same on both harnesses. Should it call the native command instead?
• Codex: only the Claude path was run (about 45s on this branch's own diff, --no-extensions). The codex exec --sandbox read-only path is untested.
• /techdebt: it is still listed in AGENTS.md but not chained by default.
• Pre-push hook: an opt-in .githooks/pre-push is not included yet.

Follow-ups planned for the epic:

• move recurring Java review rules into AGENTS.md and the docs it links
• send checks that can be deterministic lints to static analysis (APMLP-1979)
• build an eval set against human review
• name a maintenance owner

🤖 Generated with Claude Code

[dd-octo-sts]

🟢 Java Benchmark SLOs — All performance SLOs passed

Suite	Status
Startup	🟢 pass

SLO thresholds are defined here based on automatically generated metrics. A warning is raised when results are within 5% of the threshold.

PR vs. master results

Scenario	Candidate	master	Δ (95% CI of mean)
startup:insecure-bank:iast:Agent	13.96 s	13.85 s	[+0.0%; +1.6%] (maybe worse)
startup:insecure-bank:tracing:Agent	12.87 s	12.91 s	[-1.0%; +0.3%] (no difference)
startup:petclinic:appsec:Agent	17.07 s	16.98 s	[-0.4%; +1.4%] (no difference)
startup:petclinic:iast:Agent	16.83 s	17.06 s	[-2.2%; -0.5%] (maybe better)
startup:petclinic:profiling:Agent	16.58 s	16.64 s	[-1.5%; +0.7%] (no difference)
startup:petclinic:sca:Agent	17.10 s	16.87 s	[+0.5%; +2.2%] (maybe worse)
startup:petclinic:tracing:Agent	16.05 s	16.21 s	[-1.8%; -0.1%] (maybe better)

Commit: c5d40219 · CI Pipeline · Benchmarking Platform UI

---

Load and DaCapo benchmarks can be triggered manually in the GitLab pipeline. Results will appear in the Benchmarking Platform UI after completion.