Nemoclaw Maintainer Analyze Pr Value Stream

Analyze a Pull Request Value Stream

Run the deterministic read-only analyzer once for one pull request. Every invocation collects the complete observable retained lifetime. It writes a Perfetto-compatible trace, summary, and manifest under .nemoclaw-maintainer/pr-value-stream/pr-<number>/ and emits the summary as bounded JSON on stdout.

Do not run a reduced or latest-revision-only analysis. The latest-revision waterfall remains one derived part of the complete lifetime result.

Prerequisites

• Run from a NemoClaw checkout with authenticated gh access.
• Use Node.js 22.19 or later with type stripping enabled.
• Retained Vitest artifact timing is conditional: install this checkout's dependencies so node_modules/vitest/vitest.mjs is available, and install zipinfo and unzip. Optional artifact failures appear in caveats and do not become test-timing evidence.

Run the analysis

node --experimental-strip-types --no-warnings \
  .agents/skills/nemoclaw-maintainer-analyze-pr-value-stream/scripts/analyze-pr-value-stream.mts \
  --workdir "$PWD" \
  --number <pull-request-number>

The default repository is NVIDIA/NemoClaw. Use --repository OWNER/REPO only when the user requests another repository. The ten-minute target defaults to 10; use --target-minutes only when the user requests another target.

The analyzer owns completeness and safety bounds. Do not use legacy pagination or truncation options to reduce the standard analysis. If required retained history exceeds a bound, the command fails rather than calling a partial trace complete.

Outputs

A successful invocation writes:

• summary.json: the complete bounded value-stream report printed on stdout.
• trace.json: Chrome Trace Event JSON for Perfetto or chrome://tracing.
• manifest.json: exact pull request identity, head commit, output sizes, and completeness declarations. This file is written last.

The trace contains lifecycle events, all retained PR revisions, workflows, job queues, job executions, steps, external checks, reviews, comments, inline feedback, commit publication spans, and temporal feedback-to-next-change spans.

Open trace.json in Perfetto to zoom, filter categories, inspect event arguments, and measure arbitrary intervals.

Interpret the result

Report these fields first:

1. target.status and target.theoreticalFastestSeconds.
2. The largest entries in bottlenecks.
3. elapsed.approvalDelaySeconds and elapsed.mergeLagAfterReadySeconds.
4. The lifetime artifact paths and counts.
5. All applicable caveats.

Do not describe a fallback branch timestamp as exact. Do not claim causal attribution from temporal feedback-to-change spans or counterfactual approval calculations. Treat null timing fields as queued or otherwise unobserved. Treat artifact failure caveats as status evidence, not test timing evidence.

Trust boundaries

The script performs no GitHub writes. It invokes executables with argument arrays, bounds GitHub output, pagination, revisions, workflow runs, jobs, trace events, artifact sizes, and file sizes, and rejects malformed or incomplete required history. Lifetime artifact publication uses a per-PR lock. An active publication waits for up to 30 seconds; a lock left untouched for more than five minutes is reclaimed when its recorded owner identity no longer matches or owner metadata is unreadable. When either process-start identity is unavailable, the lock is retained unless the recorded process is confirmed absent. A timeout reports the exact active lock path for diagnosis.

GitHub does not expose canonical branch-creation timestamps or pull request description edit history. The trace uses the first retained exact-commit workflow as an observable push signal and records that limitation. Draft transitions appear only when GitHub's timeline returns them.

When retained CLI shard artifacts are available, the script accepts only an exact run and commit match, one regular blob-*.json ZIP entry, and bounded compressed and expanded sizes. It uses a private temporary directory and removes it after the attempt.