Navigara

github.com-openai-openai-agents-js

all · 5 devs · built 2026-07-10

Repository snapshot

Performance · 90d

38.6ETV

+942.0% since Q2 2025

ETV / dev / mo

14.7ETV

1 dev · 30-day window

Last 7 days · per dev

4.3ETV

per developer per week

Work mix

Growth27%· 32.0Maintenance49%· 58.7Fixes24%· 28.9
119.6 ETV all-time597 commits5 all-time devsJun 2025 – Jul 2026

Monthly reports

  • Highlights

    • Introduced new capabilities for *SDK-only custom data* within *tool outputs* via [pr/1360], enhancing flexibility for internal metadata.
    • Implemented *opt-in pre-approval input guardrails* for tool execution via [pr/1358], improving agent interaction safety.
    • Enhanced *sandbox error handling* by exposing retryability metadata for `SandboxError` instances, particularly for *Cloudflare sandbox errors* ([edd0a07b · Kazuhiro Sera]).
    • Improved *Multi-Capability Protocol (MCP)* servers to expose *serializable full tool results* via [pr/1380], enabling richer structured data returns.
    • Addressed critical bugs including restoring *system messages in conversation history* ([75bf1dfc · Jianke LIN]), handling *malformed realtime messages* ([pr/1414]), preventing *ambiguous tool dispatch* ([pr/1417]), and recovering *legacy Vercel sandbox authentication* ([6d9152dd · Kazuhiro Sera], [pr/1405]).

    Observations

    • Total output (Grow + Maintenance + Waste) decreased by 41% (10 vs 17) compared to the 5-month average, reflecting a period of reduced overall activity.
    • Grow score saw a significant 65% decrease (1 vs 4) compared to the 5-month average, indicating a strong shift away from new feature development.
    • Commit volume was 36, a 65% decrease from the 5-month average of 104 commits, aligning with the overall reduction in development pace.
    • Maintenance score decreased by 32% (6 vs 8) and Waste score decreased by 35% (3 vs 5) compared to the 5-month average, consistent with the lower activity levels.
    • A high proportion of commits (11 out of 36) were dedicated to `fix` types, addressing issues in *conversation memory* ([75bf1dfc · Jianke LIN]), *realtime message handling* ([90d46ee4 · Minh Vu]), *tool dispatch* ([1a5e2c0d · Kazuhiro Sera]), and *Vercel sandbox authentication* ([6d9152dd · Kazuhiro Sera], [9c6db7ce · Scott Trinh]), indicating a focus on stabilization.
    • Multiple commits ([82e583cc · Kazuhiro Sera], [85670fa3 · Kazuhiro Sera], [4a9d7518 · Kazuhiro Sera], [bf1f1ef0 · Kazuhiro Sera]) focused on refining the *`maintainer-review` skill* and its evaluation framework, suggesting an internal effort to streamline review processes.
    • A substantial number of commits (9 out of 36) were dedicated to *documentation updates*, including adding *SDK maintainer knowledge assets* ([pr/1416]), improving *tracing guides* ([d44d7ba4 · Traccia-Official], [de78a68b · William]), and fixing *GFM table rendering* after an *Astro 6.4 upgrade* ([a9ebc50e · Kazuhiro Sera]).
    • Routine dependency and version updates were performed across various *agents* packages ([pr/1413], [pr/1423], [8f6a84b9 · dependabot[bot]], [pr/1411], [pr/1389]), ensuring the project leverages the latest features and bug fixes.

Performance over time

ETV stacked by Growth, Maintenance and Fixes — 90-day moving average, normalized to ETV / month.

Average performance per developer

ETV per active developer per month — 30-day moving average.

Active developers over time

Unique developers committing each day — 90-day moving average.

Knowledge concentration

How dependent is this repo on a small number of contributors? Higher top-1 share = higher key-person risk.

Top 1
92.6 %
Top 3
99.5 %
Top 5
100.0 %

Kazuhiro Sera owns 92.6 % of commits.

Top contributors

Most impactful commits

Top 20 by ETV in the all-time window.