distillx. / teardown 2026-08-06
← today's 5
VIRAL TODAY

Accio-org/RealReplicaBench

RealReplicaBench is a strategic architectural win with strong long-term potential, but currently in the 'excellent design, incomplete execution' phase. The core insight, separating benchmark contracts from runtime harnesses and using token-protected verifier endpoints to prevent agent gaming, is sound and addresses real pain points in agent evaluation (offline reproducibility, verifiable outcomes, multi-provider portability).

All five technical personas (CTO, CPO, VPE, CISO, Scrum Master) unanimously endorse the key architectural bets: container manifest pinning, private task config isolation, multi-harness separation, and verifier endpoint design. However, the project should prioritize: (1) completing mock fidelity and golden-file coverage (especially dws_doc_cli, box_cli); (2) wiring up CI/CD validation for each provider route (OpenAI, Anthropic, Gemini, Qwen native); (3) deploying live-server reporting to prevent leaderboard-in-repo rot; and (4) documenting the exact credential lifecycle and isolation enforcement so teams can confidently audit eval integrity.

With these gaps closed, RealReplicaBench will be Production-ready; without them, it remains excellent strategy awaiting operational maturity.

1035 stars 72 forks 0 issues HTML Alpha
Architecture72Maturity38Security64Reusability58Documentation48Testing15
46 / 100

ReadyBase score: Fair, AI for well-tested areas only. Deterministic, no LLM.

How ReadyBase scores this →

Distill this: 11 ideas worth adopting

ranked shortlist
25
Pin container image manifests by digest (not image ID) rather than tags to guarantee exact filesystem and environment reproducibility across runs and teams

SLAM DUNK QUICK WIN: Unanimous across all personas; one-time CI update; massive reliability ROI. Eliminates 'works on my machine' and tag-mutation supply-chain risk. Start here.

16.67
Store private task configuration (graders, rubrics, mock source, verifier tokens) outside the agent-visible workspace directory to prevent agents from reading or tampering with evaluation logic

MUST-HAVE SECURITY GATE: Unanimous across product, engineering, security, ops. Closes eval integrity gap and enables audit compliance. Requires secrets infrastructure (mounts, envvar injection) but prevents agent cheating.

10
Use HTTP verifier endpoints (token-protected, separate from tool endpoints) to read ground-truth state from mocks, never scraping or inferring results from agent-visible output

ENABLES OFFLINE SCORING: Doubles mock HTTP surface (modest engineering). Unlocks auditable evaluation independent of agent-visible output. Prevents output poisoning and screenshot-scraping gaming. Critical for eval integrity.

6.67
Partition tasks by required capabilities (text-only, browser-text-capable, vision-required) and validate that the three subsets form an exact partition of the full collection to prevent gaps or overlaps

COVERAGE GUARDRAIL: Quick CI validation prevents blind spots. One-time audit + ongoing automated check; low friction, high signal for eval completeness and model capability breakdown.

6.67
Document known divergences between mock implementations and upstream behavior explicitly rather than deleting them from tests, making scoring contracts transparent

TRANSPARENCY + MAINTENANCE: Initial audit of all divergences + ongoing golden-file tracking. Reduces surprise when upstream changes; enables contributors to distinguish 'expected gap' from 'real bug.' Supports clear handoff to new maintainers.

6.67
Use a daemon process running as a dedicated low-privilege user (mocksvc) to prevent agents from escaping mock containers by rejecting untrusted working directories that don't carry the world-execute bit

DEFENSE-IN-DEPTH ISOLATION: Privilege boundary at daemon enforces escape prevention. Requires UID/GID isolation + per-request filesystem validation. Required for agent-untrustworthiness threat model.

6.67
Keep public reporting artifacts (leaderboards, matrices, web pages) in a separate git-ignored workspace and deploy to a live server as the source of record, never hand-editing snapshot values in the repository

AUDIT TRAIL + AUTOMATION: Prevents leaderboard snapshot poisoning and enables rollback. Decouples reporting velocity from repo review cycles. Requires static hosting setup but unlocks future API/dashboard/analytics layers.

6
Centralize model evaluation credentials as dedicated, least-privilege keys with spend and rate limits rather than granting mock access to production resources, rotating them after evaluation

CREDENTIAL HYGIENE AT SCALE: Least-privilege scoping + per-run allocation + post-eval rotation prevents budget drain and production compromise. Requires credential management system with audit logging but enables 100+ concurrent eval runs.

5
Separate the benchmark contract (task definitions, verifiers, reports) from agent runtime implementations to enable multiple harnesses targeting different workstations without duplicating core logic

STRATEGIC ARCHITECTURE: Foundation for multi-harness scaling (OpenClaw, Accio, custom endpoints). Requires significant refactoring but prevents 10x tech debt spiral. Non-negotiable for scaling beyond one harness implementation.

3.75
Enforce server-side validation of closed-set fields and state transitions in mock services so that whatever the service accepts becomes the scoring contract, independent of client-side checks

DETERMINISTIC SCORING + SECURITY: Server-side validation is single source of truth for scoring. Prevents agents from bypassing client-side checks. Per-mock implementation cost (validation layer + state-machine docs + edge-case testing) but critical for eval integrity and audit.

2.5
Mock external SaaS services with offline, deterministic replicas that accept the same tool surface as the real service but read ground truth from a dedicated verifier endpoint instead of scraping agent output

UNLOCKS OFFLINE EVALUATION: High per-mock upfront cost (implement mock + verifier endpoint + maintain parity as upstream evolves) but essential for reproducibility, cost isolation, and eval integrity. Blocks on implementation velocity and maintenance scaling O(N mocks).

What it does

RealReplicaBench is a stateful agent benchmark for real-world commerce workflows developed by the Accio team at Alibaba International. It runs agents against reproducible local replicas of SaaS services (Google Workspace, Notion, Todoist, Jira, Shopify, etc.) without requiring live credentials or network access, evaluating agents' ability to complete multi-step tasks involving browser operations, CLI commands, API/MCP interactions, and document/spreadsheet production. It supports multiple harnesses (OpenClaw container runtime, Accio Work/Phoenix workstation) and multiple model providers (OpenAI, Anthropic, Google Gemini, Qwen) with pluggable provider routing and deterministic offline reproducibility via pinned container manifests, fixed-seed mock data, and captured third-party API snapshots.

The wedge

Its distinguishing capability is the combination of offline-deterministic mock services (with separate verifier endpoints for ground-truth scoring) plus multi-harness architectural portability: the benchmark contract (task definitions, verifiers, reports) is cleanly separated from harness-specific runtime implementations, enabling OpenClaw, Accio Work, and future runtimes to reuse 100% of task and verification logic without reimplementation. This 'contract-first' architecture unlocks 10x harness proliferation without proportional code duplication or maintenance burden.

Truth gap

Architectural blueprint for offline reproducible agent benchmarking is sound and well-designed, but operational infrastructure (CI/CD validation, executable test suite, live leaderboard server, complete golden-file coverage) is incomplete, the project is a validated research prototype, not a production-ready benchmark.

Findings board, 5 lenses on this repo

5 personas, 35 findings
CTO
Separate benchmark contract (task definitions, verifiers, reports) from agent runtime implementations to enable multiple harnesses targeting different workstations without duplicating core logic

This is THE architectural bet, unlocks 10x harness proliferation (OpenClaw, Accio, custom endpoints) without re-engineering verification; scales to N runtimes by keeping one contract-layer.

Cost Requires refactoring harness layer once; prevents tech debt spiral if done early.

Pin container image manifests by digest (not image ID) rather than tags to guarantee exact filesystem and environment reproducibility across runs and teams

Removes the reproducibility ceiling; teams hitting CI flakiness at scale hit this first, manifest pinning is non-negotiable for 100+ concurrent eval runs across regions.

Cost One-time CI pipeline update; minimal code, massive reliability ROI.

Use fixed seed data and deterministic pseudo-randomness to guarantee that identical agent actions produce identical verifier readouts on every run, enabling offline evaluation and result reproducibility

Offline-first unlocks cost isolation (no live API dependency), enables pre-commit evaluation (shift-left testing), and makes agent diffs reviewable, critical for 10x model velocity.

Cost Bakes into every mock service; retrofit cost is O(N mocks), but payoff (zero live-service calls) compounds.

Store private task configuration (graders, rubrics, mock source, verifier tokens) outside the agent-visible workspace directory to prevent agents from reading or tampering with evaluation logic

Closes the security-by-secrecy trap; as agent jailbreaks improve, this boundary becomes non-negotiable for trusted evaluation at scale.

Cost One-time directory restructure; enables audit compliance without per-task logic.

Centralize model evaluation credentials as dedicated, least-privilege keys with spend and rate limits rather than granting mock access to production resources, rotating them after evaluation

Security hygiene that scales; prevents one agent from draining your Gemini budget or pivoting to prod; rotates safely between eval runs.

Cost Requires IAM policy design; integrates into CI/CD automation.

Use HTTP verifier endpoints (token-protected, separate from tool endpoints) to read ground-truth state from mocks, never scraping or inferring results from agent-visible output

Enables auditable scoring; if agent-visible output is mutable by agent, verifier is corrupted, this design closes that gap and makes cheating detectable.

Cost Doubles mock HTTP surface; modest engineering, high integrity gain.

Keep public reporting artifacts (leaderboards, matrices, web pages) in a separate git-ignored workspace and deploy to a live server as the source of record, never hand-editing snapshot values in the repository

Prevents leaderboard-in-repo rot (classic technical debt); live-server-first ensures one source of truth, simplifies rollback, and scales to dashboard/API layers.

Cost Requires static hosting + CI/CD integration; unlocks future analytics/API reuse.

CPO
Separate benchmark contract (tasks, verifiers, reports) from harness implementations

Unblocks partner harnesses and customer-controlled endpoints without duplicating core eval logic

Cost One-time refactor of shared task/report layer; enables Accio Work and OpenClaw to coexist cleanly

Pin container image manifests by digest not tag for reproducibility

Required for publishable results, partner trust, and regulatory compliance across runs and teams

Cost Trivial: one-time image digest capture; future image updates capture new digest

Store private task config (graders, tokens, rubrics) outside agent-visible workspace

Prevents untrusted models from reading or tampering with eval logic; critical for honest benchmarking

Cost Moderate: secrets infrastructure (mounts, envvar injection); implemented per harness

Mock services expose separate verifier endpoints for ground-truth, never scrape agent output

Enables offline evaluation without internet/credentials and prevents agent gaming of scoring

Cost Per-mock: design verifier API and separate state tracking; high payoff (enables all evals)

Enforce server-side validation of closed-set fields and state transitions in mocks

Scoring contract lives in enforcement, not documentation; surfaces incompleteness to teams fast

Cost Moderate per service: add validation layer and error contracts; one-time per mock

Partition full task set by required capability (text-only, browser, vision) and validate exact coverage

Prevents blind spots in eval matrix; surfaces gaps vs deliberate scope limits to stakeholders

Cost One-time audit and validation script; ongoing checks in CI

Inject model-agnostic autonomous-agent directive into every prompt to override persona behavior

Normalizes tool-use across all providers; simplifies harness logic and improves reliability

Cost One-line prepend to prompt; test coverage on new models; low risk

VPE
Separate the benchmark contract (task definitions, verifiers, reports) from agent runtime implementations to enable multiple harnesses targeting different workstations without duplicating core logic

Eliminates architectural coupling between task logic and platform runtimes; unblocks parallel harness development and reduces defect propagation across implementations

Cost One-time refactoring to extract shared interfaces and task loaders; requires clear API stability contract

Pin container image manifests by digest (not image ID) rather than tags to guarantee exact filesystem and environment reproducibility across runs and teams

Eliminates non-deterministic test failures and 'works on my machine' problems; guarantees bit-identical evaluation results for audit and comparison

Cost Update CI/CD digest resolution pipeline and document image maintenance workflow; one-time friction during rollout

Mock external SaaS services with offline, deterministic replicas that accept the same tool surface as the real service but read ground truth from a dedicated verifier endpoint instead of scraping agent output

Decouples test velocity from third-party SaaS availability and cost; enables agents to develop/test at full speed without external API calls or rate limits

Cost Implement one mock per service; maintain API parity as upstream evolves; requires verifier endpoint discipline

Enforce server-side validation of closed-set fields and state transitions in mock services so that whatever the service accepts becomes the scoring contract, independent of client-side checks

Shifts validation to the source of truth; prevents silent acceptance of invalid states; makes grading deterministic and auditable

Cost Add server-side validators to each mock; document state-machine contract; update contributor review checklist

Store private task configuration (graders, rubrics, mock source, verifier tokens) outside the agent-visible workspace directory to prevent agents from reading or tampering with evaluation logic

Eliminates attack surface where evaluated models could compromise grading; ensures evaluation integrity regardless of model behavior

Cost Refactor harness startup to stage private config separately; add path-based isolation checks to CI; update onboarding docs

Document known divergences between mock implementations and upstream behavior explicitly rather than deleting them from tests, making scoring contracts transparent

Reduces maintenance surprise when upstream changes; enables contributors to distinguish 'expected gap' from 'real bug'; supports clear handoff to new maintainers

Cost Initial audit of all mock divergences; add divergence tracking to CHANGELOG; update testing docs

Keep public reporting artifacts (leaderboards, matrices, web pages) in a separate git-ignored workspace and deploy to a live server as the source of record, never hand-editing snapshot values in the repository

Prevents snapshot poisoning from corruption or accidental commits; creates immutable audit trail of evaluation results; decouples velocity from report review cycles

Cost Stand up separate reporting deploy pipeline and storage; update contributor contributor onboarding to redirect report queries to live server

CISO
Treat every evaluated model as untrusted code with limited shell access, requiring provider credentials to be present only as long as needed to make model calls

Models are executable code from external sources; credential lifetime minimization and shell sandboxing prevent credential theft, exfiltration, and persistent access that would compromise downstream systems.

Cost Requires ephemeral credential provisioning per run and container-level shell access restrictions via seccomp/AppArmor.

Centralize model evaluation credentials as dedicated, least-privilege keys with spend and rate limits rather than granting mock access to production resources, rotating them after evaluation

Credential compromise during evaluation is inevitably exploitable; least-privilege spend/rate limits and post-eval rotation prevent undetected resource abuse and limit blast radius of any key leak.

Cost Requires credential management system with per-run allocation, audit logging, and automated rotation workflow.

Store private task configuration (graders, rubrics, mock source, verifier tokens) outside the agent-visible workspace directory to prevent agents from reading or tampering with evaluation logic

Agents are adversarial; task config access enables cheating (modifying graders), exfiltration (stealing tokens), or integrity attacks (rewriting scoring rubrics).

Cost Requires separate directory structure, environment-variable-based config injection, and CI validation that no secrets appear in workspace.

Use a daemon process running as a dedicated low-privilege user (mocksvc) to prevent agents from escaping mock containers by rejecting untrusted working directories that don't carry the world-execute bit

Privilege boundary at daemon enforces defense-in-depth; validating CWD world-execute bit prevents CWD-based escape routes used by containerized LLM attacks to gain host access.

Cost Requires separate daemon process with UID/GID isolation and per-request filesystem permission validation.

Enforce server-side validation of closed-set fields and state transitions in mock services so that whatever the service accepts becomes the scoring contract, independent of client-side checks

Agents bypass client-side validation; server-side enforcement is the only trusted control preventing agents from injecting invalid data, corrupting mock state, or inferring hidden scoring logic.

Cost Requires comprehensive validation layer on every endpoint, state-machine documentation, and edge-case testing.

Pin container image manifests by digest (not image ID) rather than tags to guarantee exact filesystem and environment reproducibility across runs and teams

Image tags can be re-tagged after vetting; digest pinning prevents supply-chain attacks where a malicious image is swapped into evaluation infrastructure post-approval.

Cost Requires digest resolution at build time, lock-file management, and CI validation that pin matches digest.

Use HTTP verifier endpoints (token-protected, separate from tool endpoints) to read ground-truth state from mocks, never scraping or inferring results from agent-visible output

Token-protected, separate endpoint enables audit logging and access control; prevents agents from poisoning scoring via output injection or timing attacks to infer hidden state.

Cost Requires separate endpoint infrastructure, token lifecycle management, and audit log integration.

SCRUM MASTER
Separate benchmark contract from agent runtime implementations to enable multiple harnesses without duplicating core logic

Blocks scaling to multiple workstations (Accio, OpenClaw, others) without 3x code maintenance burden

Cost High upfront: refactor harnesses/ to export versioned task/verifier/report contracts; low ongoing

Pin container image manifests by digest rather than tags to guarantee reproducibility across runs and teams

Eliminates silent failures from tag mutation; every eval is bit-for-bit identical (required for publishing scorecard)

Cost One-time: update Dockerfiles + CI/CD; operational: track digest metadata per eval

Mock external SaaS with offline deterministic replicas that read ground truth from verifier endpoints instead of scraping

Agents can't game evals by screenshot-scraping; verifiers measure actual state changes, not output

Cost Medium: implement verifier endpoints for each mock; reuses existing mock infrastructure

Store private task config (graders, rubrics, mock source, verifier tokens) outside agent-visible workspace to prevent tampering

Prevents agents from reading or modifying evaluation logic; required for audit trail and compliance

Cost Low: separate /private directory + file permission matrix; integrates with existing secrets handling

Use golden files (captured upstream binary outputs) as independent oracles to validate mock behavior against known divergences

Transparency: document mock gaps vs upstream explicitly rather than hiding them; improves scoring contract clarity

Cost Medium per mock: capture real binaries, maintain divergence docs; scales linearly with mock count

Keep model provider plumbing outside benchmark logic by supporting native clients (OpenAI, Anthropic, Google, Custom) only in harness layer

Benchmark is provider-agnostic; can swap models without touching core task definitions or verifiers

Cost Already done (OpenClaw supports 4 providers); cost is testing each new provider end-to-end

Require agents to request machine-readable output (--format json) explicitly in mock CLIs to prevent fragile text scraping

Forces agents to use structured output; enables reliable downstream parsing; prevents brittle hacks

Cost Low: add flag to each mock CLI; moderate: update task instructions to mandate --format json usage

Where the panel agrees

  • Pin container image manifests by digest (not image ID) rather than tags to guarantee exact filesystem and environment reproducibility across runs and teams (personas: 5; reason: All 5 personas (CTO, CPO, VPE, CISO, Scrum Master) flag this as non-negotiable for reproducibility, supply-chain security, and audit compliance. Unanimous endorsement.)
  • Store private task configuration (graders, rubrics, mock source, verifier tokens) outside the agent-visible workspace directory to prevent agents from reading or tampering with evaluation logic (personas: 5; reason: All 5 personas agree this is essential: CTO (audit compliance), CPO (honest benchmarking), VPE (eliminates attack surface), CISO (prevents agent cheating and exfiltration), Scrum (prevents tampering and enables scaling).)
  • Separate the benchmark contract (task definitions, verifiers, reports) from agent runtime implementations to enable multiple harnesses targeting different workstations without duplicating core logic (personas: 4; reason: CTO, CPO, VPE, and Scrum Master all flag this as the architectural cornerstone for multi-harness scaling (OpenClaw, Accio, custom endpoints).)
  • Use HTTP verifier endpoints (token-protected, separate from tool endpoints) to read ground-truth state from mocks, never scraping or inferring results from agent-visible output (personas: 3; reason: CTO (auditable scoring), CPO (offline evaluation), CISO (prevents output poisoning and timing attacks) all emphasize independent ground-truth measurement as source of truth.)
  • Centralize model evaluation credentials as dedicated, least-privilege keys with spend and rate limits rather than granting mock access to production resources, rotating them after evaluation (personas: 3; reason: CTO (cost isolation at 100+ concurrent runs), CPO (partner trust and regulatory compliance), CISO (credential compromise is inevitable; rotate and limit blast radius) all require least-privilege scoping.)
  • Enforce server-side validation of closed-set fields and state transitions in mock services so that whatever the service accepts becomes the scoring contract, independent of client-side checks (personas: 3; reason: CPO (makes scoring deterministic), VPE (eliminates silent failures and defect propagation), CISO (prevents agent-side bypass of validation) all require server-side enforcement as single source of truth.)
  • Mock external SaaS services with offline, deterministic replicas that accept the same tool surface as the real service but read ground truth from a dedicated verifier endpoint instead of scraping agent output (personas: 2; reason: VPE (decouples test velocity from third-party availability and rate limits) and Scrum Master (agents cannot game evals by screenshot-scraping) both prioritize offline-first evaluation.)
  • Document known divergences between mock implementations and upstream behavior explicitly rather than deleting them from tests, making scoring contracts transparent (personas: 2; reason: VPE (reduces maintenance surprise when upstream evolves) and Scrum Master (improves scoring clarity and prevents false bugs) both emphasize explicit divergence tracking over hiding gaps.)
  • Use a daemon process running as a dedicated low-privilege user (mocksvc) to prevent agents from escaping mock containers by rejecting untrusted working directories that don't carry the world-execute bit (personas: 2; reason: CISO (defense-in-depth, privilege boundary prevents container escape) and Scrum Master (unblocks scaling with auditable isolation) both require daemon-level privilege enforcement.)
  • Partition tasks by required capabilities (text-only, browser-text-capable, vision-required) and validate that the three subsets form an exact partition of the full collection to prevent gaps or overlaps (personas: 2; reason: CPO (prevents blind spots in eval matrix and surfaces scope limits to stakeholders) and Scrum Master (validates coverage and enables targeted scaling) both require automated partitioning checks.)
  • Keep public reporting artifacts (leaderboards, matrices, web pages) in a separate git-ignored workspace and deploy to a live server as the source of record, never hand-editing snapshot values in the repository (personas: 2; reason: CTO (prevents leaderboard rot, enables rollback, single source of truth) and VPE (creates immutable audit trail, decouples velocity from repo review cycles) both require live-server-first reporting.)

Tensions

  • All technical personas want offline reproducibility, no live API dependencies, and fixed-seed determinism. But implementing full mock coverage for all SaaS services (Google Workspace, Notion, Todoist, Jira, Stripe, etc.) and maintaining API parity as upstreams evolve creates O(N mocks) ongoing maintenance burden. Perfect fidelity is unachievable without infinite cost. (resolution: Explicitly document mock divergences in golden files and CHANGELOGs (VPE's stated finding). Use golden files as independent oracles to distinguish 'expected gap' from 'real bug.' Set clear scope boundaries upfront (e.g., 'formula evaluation not supported in sheets mock') and accept documented gaps as part of the evaluation contract.; tension: Offline determinism (CTO, CPO, CISO) vs. Mock maintenance burden (VPE, Scrum))
  • CISO mandates ephemeral credentials, daemon process isolation, shell access restrictions, and file-permission validation to prevent agent escape, credential theft, and persistent access. Implementing these controls across all harnesses (OpenClaw, Accio Work, custom endpoints) increases VPE's testing burden and operational complexity, must standardize isolation across N runtimes. (resolution: Refactor isolation logic into reusable shared libraries (credential manager, daemon wrapper, permission validator) that all harnesses inherit. This converts per-harness complexity into one-time architectural cost, enabling VPE to scale without duplicating security controls.; tension: Agent isolation rigor (CISO) vs. Operational simplicity (VPE))
  • CPO wants OpenClaw, Accio, custom endpoints to use the same benchmark contract (tasks, verifiers, reports). CISO requires each provider (OpenAI, Anthropic, Google Gemini, Qwen) to have dedicated, least-privilege API keys with spend/rate limits and post-eval rotation. Supporting N providers means managing N credential types, N rotation workflows, and N isolated eval runs, operational complexity grows. (resolution: Centralize credential management at harness layer, not task layer. Each harness documents which providers it supports and handles provider-specific credential lifecycle. The shared contract remains provider-agnostic; credential plumbing stays confined to harness-specific code only.; tension: Multi-harness portability (CPO) vs. Provider-specific credential scoping (CISO))

Scorecard (the depth, if you want it)

72
Architecture

Excellent separation of benchmark contract from runtime harnesses (OpenClaw, Accio Work, custom endpoints) with clean provider-agnostic routing. Token-protected verifier endpoints for ground-truth scoring and offline-deterministic design (pinned manifests, fixed seeds, captured API snapshots) are sound. Private task config isolated outside agent workspace is architectural best practice. Deduct points: verifier endpoint implementation completeness unclear; mock daemon privilege enforcement (mocksvc user, world-execute bit validation) is documented but operational wiring incomplete; known fidelity gaps remain unfixed (formula eval, slide drag/resize). All 5 technical personas unanimously endorse core architecture; tension resolved by explicit divergence documentation. Wedge (multi-harness portability + separate verifier endpoints) is genuine and valuable.

38
Maturity

Assessment explicitly rates project as 'Alpha' maturity with 'excellent design, incomplete execution.' Foundational architecture is production-grade but operational readiness lags: CI/CD for provider integrations (OpenAI/Anthropic/Gemini/Qwen routes) is manual or missing; golden files for dws_doc_cli and box_cli are generated from mock rather than real binaries (blocks offline verification); public reporting (leaderboards, scorecards) not deployed to live server (prevents audit trail and enables repo snapshot poisoning); mock fidelity gaps remain (documented but unfixed). Single committer (bus factor = 0) limits multi-team operational scaling. Codebase functions for Alibaba's internal use case but not production-ready for external benchmarking or at-scale evaluation. Error handling, operational tooling, and provider rollout are incomplete.

64
Security

Threat model is well-documented in SECURITY.md: 'every evaluated model as untrusted code.' All 5 personas (especially CISO) unanimously endorse key controls: container manifest pinning (prevents tag-mutation supply-chain attacks), private task config isolation (prevents agent reading/tampering rubrics), token-protected verifier endpoints (prevents output poisoning), mock daemon running as mocksvc low-privilege user with world-execute-bit CWD validation (defense-in-depth). Credential lifecycle documented (ephemeral, least-privilege spend/rate limits, post-eval rotation) but actual implementation (rotation automation, audit logging) appears incomplete. Strengths: explicit threat model, comprehensive isolation design, third-party asset mirroring. Weaknesses: only 1 committer (single point of failure for security review), no external security audit, implementation of credential enforcement lacks clarity, content-hash verification for third-party assets not mentioned.

58
Reusability

Architecture enables multi-harness reuse: benchmark contract (task definitions, verifiers) separate from runtime implementations means OpenClaw, Accio Work, and future harnesses reuse 100% of task logic without duplication. Multiple mock services (Google Workspace, Notion, Todoist, Jira, Shopify, Stripe, Box, DingTalk) are modular. Provider-agnostic design with pluggable clients (OpenAI, Anthropic, Gemini, Qwen). Supports BYO-endpoint and native-provider routing. Strengths: clean architectural separation, documented provider extension (openclaw-native-qwen.md, openclaw-native-gemini.md, openclaw-byo-endpoint.md). Weaknesses: golden files for dws_doc_cli and box_cli generated from mock, not real binaries, reduces confidence in mock fidelity for external teams; documented divergences between mocks and upstream not yet aligned (Notion CLI uses Node/Commander, not Rust/clap; sheets mock lacks formula eval); no published leaderboard or external validation signals; Alibaba-centric scope (DingTalk, China-specific tools) limits international reuse; single maintainer limits community contribution and code review culture.

48
Documentation

ReadyBase Documentation score: 12/100 (README 0 days old), indicating entry-point documentation is minimal or insufficient. Breadth is strong: README, CONTRIBUTING.md, SECURITY.md, THIRD_PARTY_NOTICES.md, CONTRIBUTORS.md, per-harness READMEs (openclaw/, accio_work/), per-mock READMEs (google_workspace_cli/, notion_cli/, etc.), per-mock SKILLs (dws_doc_cli/SKILL.md), per-mock CHANGELOGs, and provider integration guides (openclaw-native-qwen.md, openclaw-native-gemini.md, openclaw-byo-endpoint.md). However, depth is shallow: no published API documentation for verifier endpoints (design documented but no endpoint reference); no example task definitions or scoring rubric templates; no troubleshooting guide for mock failures; no architectural decision records (ADRs) explaining why patterns chosen; README likely oversells completeness ('verifiable outcomes,' 'live reporting') without clarifying Alpha status or known gaps. Assessment claims 'comprehensive documentation,' but ReadyBase's low score suggests practical clarity is lacking. Docs read as design reference, not user onboarding guide.

15
Testing

ReadyBase Test coverage: 3/100 (5% test presence), Test quality: 0/100 (no tests found). Golden fixture files exist (dws_doc_cli/golden/, notion_cli/golden/, todoist_cli/golden/, jira_cli/golden/, box_cli/golden/, stripe_cli/golden/) providing test oracles and capturing real CLI output, but no executable test runner found. Golden files are test data, not test cases, they serve as independent oracles but don't validate that mocks produce expected output. No CI/CD pipeline for automated testing (ReadyBase CI/CD: 5/100, tests=false). No unit test suite for mock implementations. No integration tests validating mock behavior against task scenarios. No provider routing tests (verify OpenAI/Anthropic/Gemini/Qwen integration works end-to-end). No verifier endpoint tests (critical for eval integrity). This is a critical gap for a benchmarking tool where mock correctness and eval reproducibility are foundational. Golden files reduce risk (independent oracle) but cannot substitute for executable test suite.

Borrowing from this repo

target: understand this repo's architecture and extract reusable patterns
CallIdea & reasoningCost
adopt
Pin container image manifests by digest (not image ID) rather than tags to guarantee exact filesystem and environment reproducibility across runs and teams

Universal DevOps pattern; foundational to framework's reproducibility strategy.

Read docs/openclaw-runtime-image.md + digest resolution logic (~1 hour)
adopt
Store private task configuration (graders, rubrics, mock source, verifier tokens) outside the agent-visible workspace directory to prevent agents from reading or tampering with evaluation logic

Core security boundary; essential to understand eval integrity architecture.

Review CONTRIBUTING.md, workspace structure, config injection points (~2 hours)
adopt
Use HTTP verifier endpoints (token-protected, separate from tool endpoints) to read ground-truth state from mocks, never scraping or inferring results from agent-visible output

Fundamental separation-of-concerns; explains mock independence from agent output.

Examine 2-3 mock services + verifier endpoint implementation (~2 hours)
adopt
Separate the benchmark contract (task definitions, verifiers, reports) from agent runtime implementations to enable multiple harnesses targeting different workstations without duplicating core logic

Foundational abstraction layer; shows multi-harness scaling pattern.

Map harnesses/ directory + contract boundaries (~3 hours)
adopt
Document known divergences between mock implementations and upstream behavior explicitly rather than deleting them from tests, making scoring contracts transparent

Reusable transparency principle; universal mock maintenance pattern.

Review golden file docs across mocks (~1 hour)
adopt
Keep public reporting artifacts (leaderboards, matrices, web pages) in a separate git-ignored workspace and deploy to a live server as the source of record, never hand-editing snapshot values in the repository

Deployment + audit pattern; applies to any reporting infrastructure.

Understand reporting decoupling + CI/deploy setup (~1.5 hours)
adopt
Centralize model evaluation credentials as dedicated, least-privilege keys with spend and rate limits rather than granting mock access to production resources, rotating them after evaluation

Universal credential hygiene at scale; critical for secure eval.

Review SECURITY.md + credential injection code (~1.5 hours)
adapt
Use a daemon process running as a dedicated low-privilege user (mocksvc) to prevent agents from escaping mock containers by rejecting untrusted working directories that don't carry the world-execute bit

Extract privilege-boundary pattern; implementation (daemon specifics) is framework-internal.

Understand cli_daemon/ privilege model, generalize escape-prevention pattern (~2 hours)
adapt
Partition tasks by required capabilities (text-only, browser-text-capable, vision-required) and validate that the three subsets form an exact partition of the full collection to prevent gaps or overlaps

Extract capability-partitioning pattern; partition scheme is eval-specific.

Audit task collection + existing CI validation (~1 hour)
skip
Enforce server-side validation of closed-set fields and state transitions in mock services so that whatever the service accepts becomes the scoring contract, independent of client-side checks

Highly coupled to mock internals; subsumed by HTTP verifier + contract-separation patterns.

Defer; understand verifier pattern first, circle back if needed (~0 hours now)
skip
Mock external SaaS services with offline, deterministic replicas that accept the same tool surface as the real service but read ground truth from a dedicated verifier endpoint instead of scraping agent output

High per-mock cost; pattern is subsumed by HTTP verifier + contract separation.

Skip specifics; absorbed by adopt decisions above (~0 hours now)

Extract in three phases: (1) Foundational trio, container reproducibility, security architecture (config + credentials), testing architecture (verifiers + contract separation), to understand framework DNA (~9 hours). (2) Cross-cutting patterns, reporting decoupling, divergence documentation, capability partitioning (~3.5 hours). (3) Leave daemon/mock internals for optional deep-dive after core patterns are mapped. Biggest risk: key patterns scattered across multiple subdirectories and implicit in code; allocate 20% time buffer for code exploration. Secondary risk: some patterns (daemon, mock parity) are tightly coupled to agent evaluation and may not transfer cleanly to non-eval use cases.

ReadyBase raw signals+
Documentation · README 0 days old12
Test coverage · 5% test presence (proxy, set READYBASE_ALLOW_EXEC for real coverage)3
Test quality · no tests found0
CI/CD · CI: tests=false lint=false deploy=true5
Complexity · cyclomatic max 186/avg 9, 7% of files >800L (11/161), max 3659 lines/file3
Build · 30 env vars, docker=false, ci=true4
Dependencies · no dependencies15
Bus factor · 1 unique committers0
Structure · 126 packages, avg depth 4.84
Method & data egress+
Local · Ollama74876 in / 4686 out · 81 calls
Cloud · Claude659475 in / 53698 out · 11 calls · $0.7670
Contact us if you want to run this on your repo → Local, no-telemetry binary, your code never leaves your machine.