Claude Skills by Undermybelt

Author in-repo SKILL.md: frontmatter, validator, structure.

Hermes配置优化 - 精简路由文件、添加中文触发词、压缩memory，减少token消耗

Use when Hermes MCP servers show failed or disconnected behavior and need structured self-heal: test each server, verify the backing listener or command, repair launchd persistence, and retest from Hermes.

Extend Hermes MCP server (mcp_serve.py) to expose internal capabilities to external agents. Use when adding tools to Hermes's MCP interface so Codex, Windsurf, Claude Code, and other MCP clients can access skills, routing, or other Hermes features.

Use when Hermes skills are accumulating, getting stale, or producing mixed results. Activate when you need to classify skills by observed quality, decide whether to auto-fix, iteratively improve, or freeze a skill, and keep the skill set reliable without letting weak skills silently rot.

Use when the local machine is out of disk space or the user asks to delete unused, untagged, old, cache, log, snapshot, Codex, or Hermes local state without touching important personal files.

Use when turning Hypatia into a local external memory layer for Hermes or another agent, especially when you need a thin CLI bridge and must verify that FTS search actually indexes newly inserted knowledge. Also use when Hypatia search returns no results after successful writes.

Use when Hermes needs a local external memory/knowledge-graph backend without remote services. Trigger when storing durable notes outside built-in memory, querying local knowledge shelves, syncing selected Hermes memory into a searchable graph, or building a fallback local memory layer.

Class-level umbrella for ict-engine factor-research, mutation scoring, parameter sweeps, autoresearch scripting, and structural interpretation of optimization bottlenecks. Use when working on factor-research experiments, mutation evaluation, scoring anatomy, cluster jumps, state isolation, experiment scripting, or turning reusable factor-training lessons into durable Hermes skills in ict-engine. Also use for profitability-factor transaction-cost, commission, fee-model, and cost-survival verif...

Use when committing ict-engine work, repairing provider/runtime readiness, changing shell/env setup, or turning repeated repo maintenance lessons into durable skills and routing entries.

Use when publishing ict-engine to the private release mirror, when GitHub Actions fails after publication, when a published tag needs correction, or when source repo and release mirror targets may be confused.

Class-level umbrella for ict-engine feature/surface integration work: extending FrameFeatures, wiring timed PDA and BBN evidence, safely integrating new output surfaces into main.rs, and staged extraction/triage of monolithic integration paths.

Absorb agent-facing patterns from trading/financial CLIs such as kraken-cli without blindly integrating live execution. Use when evaluating exchange CLIs, paper-trading shells, MCP-enabled finance tools, or when porting their JSON-first subprocess contracts, stable error taxonomies, paper-first workflow patterns, and machine-readable tool catalogs into Hermes or trading research systems like ict-engine.

Compact coding-discipline overlay for Hermes based on Karpathy-inspired guidelines: think before coding, simplicity first, surgical changes, and goal-driven verification. Use when doing non-trivial code edits, reviews, refactors, or debugging where overengineering, hidden assumptions, or diff sprawl are risks.

Use when Codex feels slow or bloated, when local sessions/logs/worktrees/config have grown over time, or when a user wants safe maintenance for Codex Desktop/CLI state. Provides a read-only report by default, backs up before applying changes, archives instead of deleting, normalizes Windows extended paths, prunes dead config projects, rotates large logs, and moves stale worktrees.

Wrap an existing standalone script or replay flow into a typed app adapter/orchestrator with minimal drift. Use when a repo already has working legacy automation that must be integrated into a new architecture without rewriting the old script first.

Use Google Magika for fast ML-based file-type detection from file bytes, not just extensions or MIME. Use when the user needs reliable file identification, suspicious upload triage, content-type validation, malware-prep filtering, or bulk file classification.

Add and validate market-specific frame-feature label forks in ict-engine when raw pre-Bayes labels are wrong even though factor signal quality is strong.

Use when evaluating or absorbing ideas from math-formalization agents such as MathCode into Hermes workflows. Activate when a task involves theorem proving, reusable theorem/axiom stores, subgoal decomposition, proof planning, Lean feedback loops, or tool/skill/plugin split for math work.

Route newly learned information into the right Hermes layer: repo doc, skill, memory, or session_search/L4. Use when deciding where a fact, workflow, correction, or task artifact should live after execution.

全面清理 Hermes 记忆层 + skills 目录 + 路由引用。 当用户说"记忆太臃肿"、"清理记忆"、"优化记忆"、"skills太多"、"磁盘满了"时触发。

Use when Hermes needs a local external memory layer via MemPalace, or when a user asks to install, isolate, validate, or MCP-connect MemPalace. Activate when you need durable verbatim searchable memory outside Hermes' small built-in memory budget.

Apply the user's preferred model/API fallback strategy when task execution may require provider rotation, retries, failover, or token-efficient routing.

Debug Node.js via --inspect + Chrome DevTools Protocol CLI.

古法编程模式。在 AI 辅助开发中强制执行问题建模、执行流模拟、复杂度分析与架构控制， 防止认知空心化，提升系统级正确性、性能与可维护性。适用于高并发、底层系统、逆向工程与复杂问题排查。

Create self-contained plan prompts that index papers and code repos without implementing anything

Map an academic paper or formalization repo to concrete, implementable engineering changes in a trading/research engine. Avoid math gimmicks; focus on regime fusion, scoring, gating, and feature construction.

Plan mode: write markdown plan to .hermes/plans/, no exec.

Write implementation plans: bite-sized tasks, paths, code.

对真实 Chrome + CDP 的支付/回调编排，先固化 contract 与边界，再用 probes 采证；未知站点规则不得直接升入主线。

Fix HTTP 403 blocks from custom API providers (WAF/Cloudflare) that reject SDK User-Agent headers. Fixed UA is set in run_agent.py; no more rotation pool.

Debug Python: pdb REPL + debugpy remote (DAP). 触发方式：Hermes 路由名 `software-development/py-debugpy`、通用 slash command `/skill software-development/py-debugpy`

Evaluate the QMD repo/tool before adoption, install it safely on the host, and integrate it into an agent workflow without violating the repo's own automation constraints.

Use when the user says they already cloned several repos locally and wants Hermes to inspect the newest repos, read their README/docs, then decide whether to learn from them, install them, or absorb reusable ideas. Best for local repo triage under ~ where recency matters more than a named target.

Absorb high-value patterns from OpenEvolve, RTK, Meta-Harness, and bgclick reverse-engineering into Hermes-native execution: slash-routed background loops, token-thrifty output surfaces, harness-search framing, and evidence-first reverse workflows.

Staged workflow for clearing Rust clippy debt in large repos without thrashing — compile/API migration first, then low-risk mechanical lints in batches, then signature/type refactors, with cargo check between phases.

Extract structs and functions from a large Rust main.rs into dedicated module files without breaking compilation. Covers the analyze.rs vs analyze/mod.rs ambiguity pitfall and bottom-up dependency ordering.

Use when a project's agent docs, AGENTS.md, CLAUDE.md, rules, or skills are getting bloated, duplicated, or hard to route. Activate when you need to restructure agent-facing documentation into a lean skill architecture, add triggerable skill descriptions, separate rules from workflows, or capture high-value gotchas without burying them.

Use when Hermes skills have grown beyond a handful and need explicit routing, trigger governance, and maintenance. Activate when you want a central skill router, AGENTS.md routing table, auto-update discipline for newly created skills, and a repeatable way to decide which skill to load for which user request. 触发方式：Hermes 路由名 `software-development/skill-routing-gov`、通用 slash command `/skill software-development/skill-routing-gov`

Use when auditing Hermes SKILL.md files for YAML/frontmatter breakage and repairing invalid frontmatter safely at scale.

Route user intent to Hermes slash commands before normal tool or skill execution. Use when the request is really a session-control, backgrounding, planning, queueing, resume, rollback, config, or gateway command trigger.

Use Slopless to review English Markdown for deterministic AI and human slop signals, including vague phrasing, formulaic prose, weak rhythm, filler, cliches, and readability issues.

Throwaway experiments to validate an idea before build.

Execute plans via delegate_task subagents (2-stage review).

Use when a user rejects an agent output format, system prompt rule, memory habit, or host-agent instruction and wants Hermes, Codex, Claude Code, and OpenClaw behavior made consistent.

Use when encountering any bug, test failure, or unexpected behavior. 4-phase root cause investigation — NO fixes without understanding the problem first.

Use when implementing any feature or bugfix, before writing implementation code. Enforces RED-GREEN-REFACTOR cycle with test-first approach.

用于吸收 Vibe-Skills / VCO 一类大规模 skill OS 仓的治理精华，而不直接照搬其运行时。 当用户提到 Vibe-Skills、VCO、skill sprawl、promotion gate、replay ledger、M/L/XL 路由、proof bundle、workspace memory plane 时触发。

Patch the VS Code OpenAI Codex extension so Hermes/Codex reply completion triggers an audible macOS beep and optional VS Code toast when the shipped extension ignores desktop-notification-show.