Results for “loss-function”

Sigmoid Loss for Language Image Pre-Training

This skill should be used when building automated trading systems, implementing limit orders, or adding DCA/TWAP functionality. Covers limit orders, stop losses, DCA schedules, and TWAP execution.

当感到生活失控、注意力分散，需要系统性重获专注力时

`analysis-agent`/`task-agent`/`review-agent`: use when symptoms, logs, metrics, regressions, or incidents need cause analysis; skip when no diagnosis decision exists.

Use to design overnight and follow-the-sun support — coverage models, handoff quality and measuring what is lost between shifts. Trigger for "after-hours coverage", "follow the sun", overnight support design, night shift vs defer async, handoff packet, shift handover quality, global coverage gaps, or when customers reopen tickets after every handoff.

Fix HOLD bias in RL reward function. Trigger when: (1) model learns to always HOLD, (2) trade rate is too low (<10%), (3) slippage penalty exceeds typical price moves.

当遭遇挫折、失败或结果不如预期时，用于将负面经验转化为成长燃料

Write effective fuzzing harnesses across languages to improve code coverage and find bugs in your system under test.

Diagnose OCI Functions setup, control-plane, deployment, invocation, and observability issues using a deterministic triage workflow.

Guide a team through writing a structured, blameless post-mortem after a production incident, covering timeline reconstruction, root cause analysis, impact quantification, and action item generation.

Manages automated dynamic/trailing stop losses (DSL only) for leveraged perpetual positions on Hyperliquid. Default mode: High Water (pct_of_high_water) — the trailing floor is a percentage of the peak ROE, recalculated every tick, no ceiling. Also supports fixed ROE tiers for legacy positions. Monitors price via cron, ratchets profit floors through configurable tiers, syncs the stop loss to Hyperliquid via edit_position, and auto-closes positions on breach via mcporter. Supports LONG and SHORT, strategy-scoped state isolation, and automatic cleanup.

Systematically repair a broken feature or module by scoping, tracing dependencies, diagnosing all issues, fixing them in order, and verifying end-to-end.

Scan a codebase, live app, or site and find the features that obviously should exist but don't — missing exports, bulk actions, search/filter, undo, history, notifications, dead-end workflows, data collected but never surfaced, backend endpoints with no UI, and half-built features that stopped partway. Produces an evidence-backed, ranked punch list (top 10 max) with a portable gap_packet JSON. Use whenever the user asks "what features are missing", "what's this app lacking", "find gaps", "run gap-scan", "feature audit", or wants proposal ammo from a prospect's app. Also use it to recheck an existing gap_packet — "what's left to do", "recheck the gaps", "what got built since" — re-verifying findings against current code and marking stale ones superseded. Flags ABSENCE only — broken code goes to code-audit, confusing flows to ux-audit, weak persuasion to conversion-audit, underexploited assets to a WWBD-style analysis. If something exists but is bad, it's not a gap — hand it off.

LEMON v2.0 — The Degen Fader. Identifies historically reckless traders (DEGEN activity + CHOPPY consistency) on Hyperliquid, monitors their live positions, and counter-trades them when they're bleeding at high leverage. If a cluster of degens goes max-leverage long on a coin and starts losing, LEMON shorts it — betting on their inevitable liquidation cascade. DSL exit managed by plugin runtime via runtime.yaml.

Generate one high-value follow-up workflow proposal when MoonMind execution quality issues are detected in a run.

Extract Kerberos tickets from LSASS memory, inject them into an attacker session, and perform lateral movement to access remote systems as the impersonated user.

The default, because it fails open: when the cache is unavailable, reads still

Analyzes AWS Lambda functions, including configuration, invocation metrics, cold starts, error rates, layers, and concurrency, producing structured reports.

Creates or resumes a HANDOFF.md so a fresh Claude Code session can continue work without repeating dead ends, with strict anti-hallucination verification.

Reference for MariaDB numeric and math functions, covering rounding, division, powers, logs, trigonometry, randomness, base conversion, checksums, and formatting, with guidance on exact-vs-approximate types and division-by-zero behavior.

Refactors Swift code toward declarative functional programming with immutability, value types, pure functions, transformations, and closure-based APIs.

Detect and exploit blind Server-Side Request Forgery vulnerabilities using out-of-band techniques, DNS interactions, and timing analysis to access internal services and cloud metadata endpoints.

Imported skill easing from anthropic

Executes containment strategies to stop active adversary operations and prevent lateral movement during a confirmed security breach, using network segmentation, endpoint isolation, credential revocation, and access control modifications.

Assess the test suite of a codebase — where tests are missing, where they exist but are weak, and which gaps actually matter. Use this skill whenever the user asks about test coverage, test quality, "are there enough tests", "where should we add tests", testing gaps, test debt, untested code paths, flaky tests, or generally wants to know whether a codebase is well-tested. Trigger even on casual phrasings ("is this tested?", "what's the test situation?", "audit our tests"). This skill does NOT write tests — it only assesses what exists and identifies what's missing. For test generation, this is the wrong skill.

Structured memory system with domains, decay, and links for AI agents, replacing flat memory files with a weighted, interconnected memory palace.

Analyze closed deals and lost opportunities for patterns

Measures distributional similarity between original GAN-generated images and their semantically manipulated counterparts using the Fréchet Inception Distance (FID) metric.

v4.1.0 reward function redesign to fix overtrading and DSR dominance

当需要在多个领域积累独特竞争优势，避免单一技能风险时

Invoke only when the user explicitly asks to review code through the "single level of abstraction / layered error handling" lens — a function does only its own layer's business logic while errors are handled above or below. The agent reports detections, raw-count measurements, and move directions. Apply only when the user explicitly requests this lens.

Econometrics skill for OLS regression and linear models. Activates when the user asks about: "run OLS", "linear regression", "ordinary least squares", "interpret regression results", "heteroskedasticity", "multicollinearity", "regression assumptions", "robust standard errors", "GLS", "WLS", "fit a regression model", "check regression diagnostics", "OLS假设", "最小二乘法", "线性回归", "回归系数", "残差检验", "异方差", "多重共线性", "普通最小二乘", "稳健标准误", "回归诊断"

Econometrics skill for panel data models. Activates when the user asks about: "panel data", "fixed effects", "random effects", "Hausman test", "within estimator", "between estimator", "two-way fixed effects", "clustered standard errors panel", "FE model", "RE model", "pooled OLS", "unobserved heterogeneity", "panel regression", "first difference estimator", "entity fixed effects", "time fixed effects", "面板数据", "固定效应", "随机效应", "豪斯曼检验", "双向固定效应", "面板回归", "个体效应", "时间效应", "一阶差分"