Looking for ECC maintainers: evals, workflows and tool performance

[affaan-m]

I'm looking for a small group of maintainers to help improve and own ECC's tool performance and day-to-day maintenance.

Especially interested in people with hands-on experience in:

• Writing evals and regression tests that measure whether agents and tools actually work.
• Chaining skills and tools into reliable, repeatable workflows.
• Improving tool-call reasoning: tool selection, sequencing, context management and recovery when something fails.
• Tracing agent execution and observable reasoning to understand failure modes and turn them into fixes and eval cases.
• Performance engineering across quality, reliability, latency and token efficiency.

The work also includes issue triage and bug reproduction, first-pass PR reviews with testing evidence, docs and cross-harness consistency, and release preparation. The goal is to keep ECC improving while freeing me up to focus more on new features and the app.

Roles can be paid and part-time, starting with a scoped trial. We'll agree on responsibilities, compensation, hours and access before any paid work starts.

If you're interested, reply below with:

1. Your GitHub profile and a few relevant contributions or examples of your work.
2. The area you'd like to own.
3. Your realistic weekly availability.

Existing contributors are especially welcome. No long application or new unpaid assignment needed. Keep personal and payment details out of public replies; compensation and access discussions will happen privately.

[github-actions[bot]]

Discord delivery: complete (message 1566924359409606858).

[z0rd0n88]

@affaan-m I would love the opportunity to contribute to such a famous and well-established software project. I'm happy to send my full résumé privately.

Interests

• writing evals and regression tests
• skill chaining and workflows
• tool call reasoning

Availability

• 20-30 hrs/week
• in PST but flexible
• Sunday - Friday

Experience

• over six years experience with full stack at FAANG.
• over 5 years experience with cybersecurity penetration testing and research
• I'm passionate about software development and AI because it has allowed me the tools and methods to massively express my interests and curiosity.

GitHub

• z0rd0n88
• think-time is multi-agent idea generation tools with fan-out, adversarial personas, and judging to cancel bias
• idea-autopsy is evidence-gated proposal review with research-cited bias safeguards
• yt-eval is a youtube video evaluation and fact-validation workflow
• session-stickler is tools for orchestration, security, and memory management.
• friendex - creative game for user engagement in online communities
• inventory is a tool and automation tracker

[ashutoshb206]

Hi @affaan-m, I’d love to be considered for this.

About me

Ashutosh Bansal, final-year B.Tech CSE student at Manipal Institute of Technology, Bengaluru, specializing in Agentic AI Engineering
SDE & GenAI Intern at INDmoney (May to Aug 2026), working on production GenAI features

GitHub
ashutoshb206
AutoDiligence: multi-agent due diligence platform on LangGraph and LangChain, with LangSmith tracing to debug tool selection and sequencing
PR Impact Reviewer: cross-repo PR safety reviewer using a dependency graph and a LangGraph multi-agent review pipeline
Deep study of MCP, plus integrations built around it

Area I’d like to own

Evals and regression tests for tool-call reasoning
Tracing agent execution so failure modes turn into fixes and eval cases
Also happy to help with issue triage, bug reproduction and first-pass PR reviews with testing evidence

Availability

Around 25 hours per week, flexible around release windows
Happy to start with a scoped trial, and to discuss compensation and access privately

Thanks for considering it!

[ROMB]

I'd like to be considered for paid, part-time work on MCP/tool reliability and eval regression coverage.

My main GitHub is ROMB. Relevant promptfoo work:

• #7001, merged: MCP stdio lifecycle and initialization fixes, with CLI smoke coverage.
• #7016, merged: server-side evaluation-result filtering, with unit tests.
• Separately, my conditional-grading implementation was used in my own work to avoid spending judge tokens when static checks were sufficient. This PR was closed unmerged.

I'd like to own reproducible MCP/tool failure cases and turn them into regression tests that can verify fixes. I can start immediately; exact weekly hours would need to be agreed privately before a scoped paid trial.

Happy to discuss responsibilities, compensation and access privately before starting.

[Jeevang1-epic]

Hi Affaan, I'm Jeevan Kumar ([@Jeevang1-epic](https://github.com/Jeevang1-epic)), a CSE student and open-source contributor.

I'm interested in helping maintain ECC, particularly around agent evaluations, regression testing, and workflow reliability.

Relevant work:

• [AgentForge QA](https://github.com/Jeevang1-epic/agentforge-qa) — a TypeScript CLI I built to verify AI coding-agent claims against deterministic repository evidence, test results, and Git state.
• [Kornia contributions](https://github.com/kornia/kornia/pulls?q=author%3AJeevang1-epic) — experience with open-source contributions, regression testing, CI validation, and addressing maintainer feedback.

Area I'd like to own: Reproducible agent evaluations, regression coverage, and tool failure analysis.

Availability: Approximately 22-25 hours per week alongside university.

I'd be happy to start with a scoped trial and contribute to ECC's reliability and performance.

Thanks for considering me!

[shivansh023023]

Hi Affaan! I'd love to be considered for the ECC maintainer group.

I'm Shivansh Singh, a Computer Science undergraduate and active open-source contributor with experience across systems programming, AI/ML infrastructure, backend engineering and software reliability.

Relevant experience and contributions:

• HPX: I'm a core contributor to HPX, a C++ parallel and distributed computing framework. Apart from implementing fixes and improvements, I also review PRs from other contributors, help validate their changes, and assist the main maintainer with ongoing project maintenance. This has given me experience working with a large codebase, investigating complex issues, reviewing code and maintaining software quality. http://github.com/TheHPXProject/hpx/pull/7354
  (https://github.com/TheHPXProject/hpx)

• Google DeepMind – OpenSpiel: I've worked with OpenSpiel, gaining exposure to research-oriented software and the engineering involved in game-theoretic and reinforcement learning systems.(fix(games): correct player color indexing in kAllPlayers observations google-deepmind/open_spiel#1525)

• NVIDIA NeMo: I've worked with NVIDIA NeMo, particularly around speech-related ML infrastructure, model optimization and quantization.([ASR] Add missing Python 3 type hints to transcribe_utils.py NVIDIA-NeMo/Speech#15707)

• SBI: Contributed a fix for RestrictedPrior SIR sampling (PR feat: expand Kiro adapter to full language coverage #2101), which was merged. This involved investigating sampling behaviour and validating the correction.

• Meshery: Worked on Kubernetes connection and kubeconfig handling in the Go CLI/backend, including validation, context selection, safe file handling and edge cases.(https://github.com/meshery/meshery)

Area I'd like to own:

Agent evaluations, regression testing and workflow reliability.

I'm particularly interested in building reproducible evaluations for agent and tool execution, identifying failure modes, and converting them into regression tests that ensure fixes actually improve reliability.

I would also like to work on tool selection, sequencing and context management, especially developing ways to evaluate whether an agent has genuinely completed a task using verifiable evidence rather than simply trusting its final response.

I believe my experience in debugging complex systems, reviewing contributions, writing regression tests and maintaining production-quality software would translate well into this area.

Availability: 25–30 hours per week. I can work across all seven days of the week and have flexibility in scheduling my contributions.

I'm happy to start with a scoped trial and discuss responsibilities, expectations and compensation privately.

Thanks for opening this opportunity!
Email - singhshivansh023@gmail.com
discord - acelegend7657
github - shivansh023023

[montjeffrey]

Hi @affaan-m, I'd like to be considered. I'm an existing contributor, and I use ECC every day in production at a medium-sized HVAC and weatherization company that runs energy-efficiency programs with New Jersey's public utilities. I run Claude Code on Windows across about a dozen repos and track main weekly with a sync that validates each update and rolls it back if it fails. My aim matches this role: raise quality without giving up speed or tokens, since ECC's efficiency directly limits how many agents I can run in parallel.

1. GitHub and relevant work: @montjeffrey

• Evals: feat(gateguard): improve fact-force signal quality and session efficiency (closes #2608) #3248 reworked GateGuard's fact-forcing against a replayable 20-scenario corpus (must-deny bypasses 8 → 0, irrelevant questions 187 → 2). feat(gateguard): measure question effectiveness with hidden-intent trials #3481 added hidden-intent trials that test whether those questions change task outcomes, and fix(gateguard): make harness graders check content and count only valid trials #3519 made the graders reject decoy answers so a wrong output can't pass.
• Tool-call reasoning and context: fix(gateguard): credit only searches that could have seen the target #3496 (credit only searches that could have seen the target) and fix(context): require a name or trigger anchor for implicit admission #3451 / feat(context): suggest skills at prompt time from a generation-bound index #3453 (skill routing at prompt time).
• Performance: perf(context): load the registry once per resolution and routing index build #3518 cut resolveTaskContext from 1332 ms to 645 ms.
• Cross-harness and release: fix(hooks): never run .sh hooks through the WSL bash launcher on Windows #3476, test: stop Windows developer tooling from leaking into setup, tar, and audit tests #3475, test: run the Hermes hook tests with a resolved Python 3 on Windows #3524 (Windows hook and test fixes) and feat(carriers): validate carrier runtime closure in CI #3457 (a carrier closure validator in CI).

2. Area I'd like to own: evals and regression gates for hooks and context routing, so every change is measured against main on replayable corpora and failures from real sessions become new cases. A concrete first trial task: open the WS1 routing-quality gate from #3458 (a 77-prompt corpus that fails CI on regression). It's built, and it's been unblocked since #3451 merged. I'm also glad to take Windows and cross-harness bug reproductions and first-pass PR reviews with test evidence.

3. Availability: 10–20 hours a week, weekday evenings into late night (US Eastern) and most weekends. Happy to start with a scoped trial.

To talk through scope, compensation and access, the email on my GitHub profile reaches me directly.

[israeldi]

Hi @affaan-m — this is an AI-agent eligibility enquiry from this account, before a maintainer application. The agent would perform implementation, testing and written communication, with no human engineer or reviewer on our side. Is that arrangement eligible for the paid part-time work?

1. GitHub and work example: israeldi; original synthetic HTML/JavaScript form-repair fixture. The AI-authored example records 14 scoped Chrome interaction checks, including reproduction of its original zero-value and navigation bugs. Its README records the verification limits. It is an independent synthetic example, rather than client work or an ECC contribution.
2. Preferred area: reproducible bug reports and regression-case design with test evidence, within the maintenance responsibilities you described.
3. Weekly availability: not established for this preliminary enquiry. I would confirm a realistic allocation after the actual trial scope and existing shared commitments are checked.

If eligible, what private route should I use to confirm whether the scoped trial is paid and agree its complete deliverables, acceptance, compensation, hours and access before any work or commitment?

[Kernel-Hunter]

Hi Affaan, I'm Karim (Kernel-Hunter). I use ECC every day in Claude Code.

This week I picked up #3846. I tested #3849 locally and left a review: its JWT regex goes quadratic on long hyphenated input (a 300 KB command takes about 27 s to sanitize), and a lookbehind in place of \b fixes it with all its tests still passing. Then I opened #3858 for the observe.sh half. Observations now redact bare sk- keys and JWTs, including ones cut at the capture limit, and it closes two older leaks for long labeled values and labeled PEM blocks. It has 25 regression cases that fail on main, plus checks that git SHAs, UUIDs, hashes and base64 image data stay readable.

I'd like to own evals and regression tests for hooks and redaction, issue triage with reproductions, and first-pass PR reviews with testing evidence like the one on #3849.

Outside ECC I maintain open source agent tooling. Mudroom runs coding agents in a Linux VM sandbox with a network allowlist, and you review their changes like a PR. Kandle ships an MCP server with 28 tools. Recent merged fixes elsewhere: icloud-photos-downloader #1380, #1381 and #1382.

On time: I can give about 1.5 to 2 hours a day, so roughly 10 to 14 hours a week.

[hemv-857]

Hi @affaan-m, I'd like to be considered for the ECC maintainer group.

1. GitHub and relevant work — @hemv-857 (40 repos, 7 live):

• This repo (affaan-m/ECC): fix(install): tolerate JSONC comments in the OpenCode config membership check #3590 / fix(install): integrate reviewed OpenCode JSONC fix with malformed-comma guard #3732 — fixed the JSONC unterminated-comment crash that corrupted hook-consent files (issue OpenCode installer crashes on JSONC comments in user's opencode.json (strict JSON.parse in hook-consent.js disableOpenCodeHookPluginRegistration) #3546); 80/80 hook-consent tests pass, all four review threads answered, and the malformed-comma guard is in the merged integration.
• swiftlang/swift (69k★): #93473 — per-repository exponential-backoff retry in update-checkout; 88/88 unit tests, CI green on Linux/Windows/macOS.
• ArturSepp/QuantInvestStrats: docs: fix README reference to non-existent opencode-ecc npm package #192, Merge agent/0-07-bv-blackbox into main #225, 🐛 Bug: observe.sh environment variable syntax error in continuous-learning-v2 #227 — merged; full suite 6,701 passed across the CI matrix.
• bloomberg/causal-ts: feat: add comprehensive Go development support with agents, commands, and skills #83, docs: add missing MIT LICENSE file #85, Codex/create pg branch for postgresql development ktwinl #86 — merged batched-precision + fallback fixes; suite 794 green.
• MCP Vetter Agent (repo): an agent-auditor that scans MCP servers via static analysis + Docker-sandboxed dynamic probes mapped to the OWASP Agentic Top 10, with a human-in-the-loop approval gate on local Ollama — built to catch bad tool behaviour before tools get installed.
• Failing-first workflow across the board: sympy #30901 (47/47 CI green), timesfm docs: add Antigravity setup and usage guide #552 (65 passed) — repro first, fix second, then full matrix.

2. Area I'd like to own — evals and regression coverage from real failures: turning every reproduced bug into a failing-first regression case (hooks, skills, and connectors are where my merged work lives); first-pass PR reviews with testing evidence; issue triage with reproducible root causes, in the same vein as the merged ECC JSONC fix. I'm comfortable with Windows/Linux userspace and shell behaviour as needed for git hooks and tests.

3. Availability — about 10–14 hours per week, consistent, with room to flex if a release or review window demands it. Happy to start with a scoped trial and work out compensation and access with you privately.

[shoyann]

Hey @affaan-m,

I'd like to throw my hat in the ring!

I'm [@shoyann](https://github.com/shoyann). I've been contributing to ECC for a while now, with 14 merged PRs out of 16 submitted. Most of my work has been around fixing reliability issues, improving existing workflows, and making sure things behave as expected across different environments.

Some of my ECC contributions:

• [#3252](fix(state-store): preserve concurrent writes and cleanup failures #3252) and [#3283](fix(sessions): serialize alias read-modify-write operations #3283) — Fixed concurrency issues in the state store and session alias operations that could lead to lost updates.
• [#3282](fix(context): respect indirect skill citations in auto routing #3282) — Fixed unintended skill loading from indirect references and negated instructions.
• [#3130](fix: preserve edited Codex user configuration during reinstall and repair #3130) — Prevented ECC from overwriting user-edited Codex configurations during reinstall and repair.
• [#3499](fix(llm): honor selected models in provider resolution #3499) — Fixed LLM providers ignoring the selected model.
• [#3587](fix(install): preserve excluded MCP servers in install state #3587) — Fixed excluded MCP servers being restored during repair.
• [#3445](fix(hooks): run Windows formatter shims in quality gate #3445) — Fixed formatter execution on Windows.

I also contributed the [OSINT investigation skill (#3198)](#3198), adapted from my own open-source project, [RZK-The-Hunter](https://github.com/shoyann/RZK-The-Hunter).

Contributions outside ECC:

I've also had 9 PRs merged in [REA](https://github.com/morluto/rea) and 1 in [Zed](https://github.com/zed-industries/zed). A few examples:

• [REA #1821](fix(setup): retain symlinked data roots during uninstall morluto/rea#1821) — Fixed an uninstall bug that could accidentally delete data outside REA's managed directory.
• [REA #2104](fix(javascript): resolve module-relative Worker URL references morluto/rea#2104) — Improved JavaScript analysis to correctly recognize module-relative Web Worker references.
• [REA #1834](test(web): cover nullable HTTP captures through CDP and CLI morluto/rea#1834) — Added regression tests covering HTTP captures through Chrome DevTools Protocol and the public CLI.
• [Zed #65684](Fix Rust task variables for files outside Cargo projects zed-industries/zed#65684) — Fixed Rust task variable resolution for files outside Cargo projects.

Working across these projects has given me experience with different codebases, testing approaches, and review processes.

Area I'd like to own:

I'd be most interested in workflow reliability and cross-harness compatibility, particularly installation, configuration handling, context routing, and regression testing.

I'm also happy to help reproduce bugs, review smaller PRs, and work through the issue backlog.

Availability: 20–30 hours/week.

I've enjoyed working on ECC and would be happy to get more involved in maintaining it. Open to starting with a scoped trial and discussing the details.

Thanks!

[lomsadzelomsa775-cmyk]

How to connect it on cloud I'm with phone

[ub3dqy]

Hi Affaan, I’d be interested in a paid trial on tool reliability and regression tests. I ran ECC’s current resolver checks on Windows, including the simulated WSL path cases. A couple of examples from my GitHub are this site-release workflow and Filament regression tests.

I can start with one case, capped at four hours in the first week. Is there a startup or workflow recovery issue you’d like me to take first? I’m happy to agree scope, hours, and compensation in a private GitHub thread.

Dima Mak / ub3dqy

[AnuAnsh17]

Hey! I'd be very interested in contributing to ECC and helping with its ongoing maintenance.

GitHub: https://github.com/AnuAnsh17

A few relevant projects I've been working on:

• WyvernOS — An independent Rust-based agent runtime I've been building. It explores hierarchical runtime orchestration, skill/tool execution, provider-neutral model routing, persistent context, memory, capability management, and scheduling. This has given me hands-on experience with the engineering challenges involved in making agent systems predictable and reliable.

• ECC — I've been exploring the ECC codebase and its architecture as well, particularly the agent harness, skills, tooling, and performance side. I'd like to contribute more directly to its testing, evaluation, and maintenance.

• Smart Utility / OCR-Trisynex — Built a document-processing pipeline involving OCR, structured extraction, API integration, and performance optimisation. I've worked on ingestion bottlenecks, execution latency, and making extraction results more reliable.

• RatLab — An independent Android security research project involving event processing, persistent queues, and system-level observability.

Area I'd like to own: Agent/tool reliability and evaluation.

I'm particularly interested in writing regression tests and evals, reproducing tool-execution failures, tracing agent behaviour, and improving tool selection, sequencing, and recovery. I'm also comfortable working across backend infrastructure and performance optimisation, so I can contribute wherever there's a concrete maintenance need.

Availability: Around 15–20 hours per week, with flexibility depending on academic commitments.

I'm happy to begin with a scoped trial, work through existing issues, and build trust through measurable contributions. I'd prefer to start by understanding the current testing and evaluation gaps and taking ownership of a well-defined area.

Looking forward to contributing!

[charlesreffett]

GitHub: https://github.com/charlesreffett/
LinkedIn: https://www.linkedin.com/in/charlesreffett/
Example of work: https://github.com/charlesreffett/agentkeeper

Preferred area: Anything with workflow optimizations, cross-harness consistency and handoffs, performance optimizations. Really a broad range, but also interested in other work as necessary.

Availability: 20-30hr/wk (part-time), 30-40hr/wk (full-time if available), or as needed