“benchmark”的搜索结果

Guides implementation of agent memory systems, compares production frameworks (Mem0, Zep/Graphiti, Letta, LangMem, Cognee), and designs persistence architectures for cross-session knowledge retention. Use when the user asks to "implement agent memory", "persist state across sessions", "build knowledge graph for agents", "track entities over time", "add long-term memory", "choose a memory framework", or mentions temporal knowledge graphs, vector stores, entity memory, adaptive memory, dynamic memory or memory benchmarks (LoCoMo, LongMemEval).

中文简介正在准备中。当前显示原文说明。

Use when raw SaaS numbers need turning into a health verdict — MRR, ARR, churn, CAC, LTV, payback, burn multiple, NRR. Calculates the metrics, benchmarks them against industry standards, and returns prioritized, plain-English advice as a senior SaaS CFO would.

中文简介正在准备中。当前显示原文说明。

Improve AI application with evaluation-driven development. Define eval criteria, instrument the application, build golden datasets, observe and evaluate application runs, analyze results, and produce a concrete action plan for improvements. ALWAYS USE THIS SKILL when the user asks to set up QA, add tests, add evals, evaluate, benchmark, fix wrong behaviors, improve quality, or do quality assurance for any Python project that calls an LLM model.

中文简介正在准备中。当前显示原文说明。

Use when diagnosing agent benchmarks or runtime failures from trajectories instead of aggregate pass rates. Builds shared action-observation decision landscapes, identifies productive cores and trap regions, profiles Access/Trap/Repair behavior, and designs conservative trap-triggered recovery experiments.

中文简介正在准备中。当前显示原文说明。

Use when improving, evolving, benchmarking, or maintaining agent skills over time. Treats skills as long-lived structured artifacts: execute tasks, diagnose trajectory evidence, accumulate textual momentum, patch at the correct abstraction layer, test per model and harness, canary changes, and roll back regressions.

中文简介正在准备中。当前显示原文说明。

Automated compliance checking against CIS, PCI-DSS, HIPAA, and SOC 2 benchmarks

中文简介正在准备中。当前显示原文说明。

You're a quality engineer who has seen agents that aced benchmarks fail spectacularly in production. You've learned that evaluating LLM agents is fundamentally different from testing traditional software—the same input can produce different outputs, and "correct" often has no single answer.

中文简介正在准备中。当前显示原文说明。

Analyze WAF (ModSecurity/AWS WAF/Cloudflare) logs to detect SQL injection attack campaigns. Parses ModSecurity audit logs and JSON WAF event logs to identify SQLi patterns (UNION SELECT, OR 1=1, SLEEP(), BENCHMARK()), tracks attack sources, correlates multi-stage injection attempts, and generates incident reports with OWASP classification.

中文简介正在准备中。当前显示原文说明。

Multi-language test infrastructure for the Pump SDK — Rust unit/integration/security/performance tests, TypeScript Jest tests, Python fuzz tests, shell test orchestration, Criterion benchmarks, and CI quality gates.

中文简介正在准备中。当前显示原文说明。

Production-grade multi-threaded Rust vanity address generator for Solana — 100K+ keys/sec using Rayon parallel iterators with solana-sdk, Base58 pattern matching, prefix/suffix support, security-hardened file output, and Criterion benchmarks.

中文简介正在准备中。当前显示原文说明。

Compares one clause (liability, termination, confidentiality or any other) across several contracts, templates or successive versions and returns a DRAFT comparison table: verbatim clause text per document, the clause split into elements with every difference from the baseline highlighted and categorised, and neutral questions for counsel. Never says which wording is better or enforceable. Use when the user asks to "compare the liability caps across these five supplier contracts", "line up the termination clause in these three drafts", "show how the confidentiality clause changed between versions", "contrast this clause with our template" or "benchmark the indemnity wording across our customer agreements". Do not use for a whole-contract review, use contract-review-pack instead; for screening one NDA against standard positions, use nda-triage. Drafts for human review; never approves, authorises or signs off.

中文简介正在准备中。当前显示原文说明。

Remote-sensing literature search, citation verification, sensor/platform/dataset query strategy, citation file management (.ris/.bib/.enw conversion), and reference management (BibTeX, related articles, DOI/arXiv lookup) across CrossRef, arXiv, Scopus, ScienceDirect, Semantic Scholar, and domain web sources. Use when the user needs coordinated multi-step literature workflows for remote sensing, Earth observation, geospatial AI, GIScience, photogrammetry, SAR, hyperspectral imaging, LiDAR, satellite missions, benchmark datasets, land-cover mapping, change detection, object detection, segmentation, retrieval, or multimodal remote-sensing foundation models. Also trigger on general literature-search needs during academic writing even without the word "Nature", such as searching for papers/literature, doing a literature review, verifying a citation, converting citation files, and Chinese phrasings like 文献检索、查文献、找文献、 文献综述检索、查论文、引文核对、参考文献管理、文献去重.

中文简介正在准备中。当前显示原文说明。

ego-browser (ego-lite) is a Chromium-based browser that gives AI agents a CLI-accessible Node.js runtime for driving a real browser. Use this skill whenever the user needs to interact with a website opening pages, filling forms, clicking buttons, taking screenshots, extracting page data, testing web apps, logging into sites, automating browser operations, or any other browser automation task. Triggers include requests to "open a website", "visit a URL", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "extract content from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction. Also used for exploratory testing, dogfooding, QA, bug hunting, or reviewing app quality. Prefer ego-browser over any built-in browser automation, web fetch, or other web tools.

中文简介正在准备中。当前显示原文说明。

Architecture design and governance protocol for system structure, boundaries, contracts, APIs, schemas, migrations, rewrites, and long-lived codebase health. Use when designing or reviewing architecture, planning structural refactors or migrations, governing tech debt, or containing rewrite risk. Keeps the load-bearing spine small, owned, checked, and deletable over time.

中文简介正在准备中。当前显示原文说明。

Export Codex rollout/session JSONL as a self-contained HTML report that can be opened offline. Use when a user provides a complete session file path or Codex session ID, or asks to export a Codex session to HTML, generate a session report, or convert a rollout to HTML. Supports exact full-ID lookup in the CODEX_HOME sessions and archived_sessions directories.

中文简介正在准备中。当前显示原文说明。

Verify a tested web agent's real behavior from its recorded Pi or Codex session and selected screenshots before scoring. The judge prompt provides absolute artifact paths; use targeted queries to compare actual outputs, errors, and the final answer with the recorded behavior, then return only the verdict JSON. Only for ego-bench judging—never to drive a browser or rerun the agent.

中文简介正在准备中。当前显示原文说明。

lm-eval-harness: benchmark LLMs (MMLU, GSM8K, etc.).

中文简介正在准备中。当前显示原文说明。

Characterizing and Understanding Energy Footprint and … — 7. Nov. 2025 · This study benchmarks four GAP: OpenHands has a modular SDK design — our too

中文简介正在准备中。当前显示原文说明。

Use when the user wants to send the same prompt to several LLMs and compare the answers - sequentially (stopwatch, clean per-lane timing) or in parallel (true race), with optional repetitions per model, judged by the starting model across quality, correctness, completeness, instruction fidelity and latency (time is only one dimension). Triggers on /compare-race, 'model race', 'vergleiche modelle', 'gleicher prompt an mehrere modelle'.

中文简介正在准备中。当前显示原文说明。

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

中文简介正在准备中。当前显示原文说明。

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, update or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

中文简介正在准备中。当前显示原文说明。

Use when the user asks to "package our proof points", "build reusable stat cards and case snippets", or "put proof where each pillar makes its claim"; turns claims-ledger-approved proofs into reusable proof modules — stat cards, case snippets, testimonial blocks, comparison proofs — each pinned to a message-house pillar and the ledger claim ID it substantiates, and flags any pillar making a claim with no approved proof behind it. Never adjudicates a proof: unverified or ledger-absent proofs are marked '[needs source]' and routed to the claims candidates. Not for adjudicating or substantiating claims — use offer-claims-registry; not for fabricating a benchmark to fill a gap — a missing proof is flagged, not invented; not for scoring narrative quality — use narrative-quality-auditor. 证据模块/证据卡/客户案例/主张对齐

中文简介正在准备中。当前显示原文说明。

Use when the user asks to "analyze influencer campaign performance", "compare influencers", or "find what content worked"; produces metric scorecards vs target and benchmark, platform/influencer/content rankings, engagement-quality and sentiment reads, conversion-attribution breakdowns, and ranked learnings. Not for dollar-level return math — use roi-calculator. 达人营销效果分析/投放复盘

中文简介正在准备中。当前显示原文说明。

Use when the user asks to "watch my list health over time", "flag decaying / unengaged subscribers on a schedule", "why is my open rate drifting down / bounces creeping up", or "build me a re-permission and prune worklist"; runs the scheduled SEND list-decay + suppression-drift watch — an engagement-recency cohort read (30/90/180/365-day), hard-bounce and spam-complaint trend vs benchmark, suppression-list growth/leakage check, and a segmented re-permission / sunset / prune worklist tied to SEND S (list hygiene) and E (engagement-decay) sub-items. Not for the one-time pre-send authentication pre-flight — use deliverability-qa; not for the consent/suppression record itself — use consent-registry; not for computing the EQS or enforcing vetoes — use email-quality-auditor. 邮件列表健康度监控/退订漂移/沉睡用户清理

中文简介正在准备中。当前显示原文说明。