このリポジトリの skills
SkillsMP は basicisabc1994/g6solver から 98 件の skill を収集しています。skill を開くとソースと詳細を確認できます。
basicisabc1994/g6solver収集済み skill 98 件中 40 件を表示しています。
Use when a coding task should be handed to Anthropic's Claude Code, the command-line coding agent the person installed and signed in to, in one project folder, with its own tools named on every run (read tools only unless edits are granted, never its shell or…
原文の言語: 英語
Use when a coding task should be handed to OpenAI's Codex, the command-line coding agent the person installed and signed in to, in one project folder, with no more access than the task needs (read-only unless writes are granted), a deadline, G6's stop, and…
原文の言語: 英語
Use when several agents must pass messages to one another durably (groups of named members, messages to one member or to all, optionally typed), with every message reaching its reader at least once, in order, a page at a time, coming back until it is…
原文の言語: 英語
Use when an agent runtime (a coding agent's command line, a framework's agent loop, your own adapter) is to run a task under limits a person set, with every tool call decided before it runs, the run stoppable, every failure handled the same way, and the…
原文の言語: 英語
Use when one step of Python code fails with an error you have tests for, and an agent wants a model to repair it without a careless automatic fix; the error is triaged first, the patch is checked in the sandbox on tests the model sees, and it is kept only if…
原文の言語: 英語
Use when an agent is starting a new G6 component, wrapping a Python library as one, or about to call a component package done, and needs the library's own shape scaffolded (declarations, a typed input, refusing handlers, a failing test, a skill whose ops…
原文の言語: 英語
Use when an agent has changed a codebase's structure (moved code, added imports, split a module) and must check before reporting it done that it added no import cycle, broke no layer rule and added no wrapper that only forwards its arguments; it reads a file…
原文の言語: 英語
Use when an agent is about to rename a Python name or move statements into a new function and must change exactly the identifiers and statements meant, keeping every comment, instead of a text edit that also hits strings, attributes, keyword arguments and…
原文の言語: 英語
Use when a coding agent is about to change a repository it does not know, and needs to learn first what is there, how it builds and runs its tests, what a test or source file holds, and, after a change, which files and tests it touches and whether the changed…
原文の言語: 英語
Use when an agent keeps many memories and needs them ranked the way human memory ranks them, by how recently and how often each was used and by the current context, with a recall refused when nothing is strong enough, or when a small rule set must consult…
原文の言語: 英語
Use when trying a model-based planning agent design on a small environment and it must be judged by a test rather than by argument, so an agent acting on a model that has not learned the environment's rewards is caught by its regret against a random policy on…
原文の言語: 英語
Use when an agent needs a short sequence of steps to reach a goal from declared actions (each with preconditions, what it adds and what it removes), or must check a plan someone else wrote, a model's included, step by step before acting on it, so a plan that…
原文の言語: 英語
Use when an agent must choose what to try next (which experiment, which arm, which action in a small known model) or how many times to retry each task within a budget, and the choice must be a standard rule with its working shown and the same answer on every…
原文の言語: 英語
Use when repository files, retrieved passages, notes, subagent output or a long conversation are about to go into a model call and could overflow the budget, bury the task, carry secrets, or carry instructions the model should only read; or when an error…
原文の言語: 英語
Use when an agent is about to accept one model's first answer to a judgement question (a plan, a design trade-off, a go or no-go) and wants role-named critics to push on it in turn, with dissent surfaced and an honest reading of how far the turns agree, which…
原文の言語: 英語
Use when a small generated or distilled Python function (a solve(x) meant to replace a model call or a hand-written rule) is about to be relied on, and you need to know whether it holds on cases it was not made from, run out of process under a time and memory…
原文の言語: 英語
Use when a robot motion plan or grasp, from a model, a planner or a person, is about to be carried out and you need to know whether it collides along its segments, leaves its workspace, breaks a joint or speed limit, misses its goal, or lets an object slip or…
原文の言語: 英語
Use when a conclusion is about to be stated or acted on and its confidence must match its evidence, such as a recommendation someone will act on, a health, legal, security or financial claim, an irreversible step, a percentage without its baseline, causal…
原文の言語: 英語
Use when a piece of work (a patch, a report, any artifact) must be given a completion state that says only what its checks support, with every gap named, or when a certificate someone hands you must be checked for consistency with the artifact's bytes.
原文の言語: 英語
Use when an agent loop runs a long, multi-step goal and must check before each step that it is still inside its declared token, time and disk limits, that the action it is about to take is allowed, and that a person has signed off at the points named in…
原文の言語: 英語
Use when conclusions spread over a graph of people, services, accounts or documents from facts that are only partly certain (trust, risk, popularity, reliability scores) and you are about to state how sure a conclusion is, or to combine rules that might…
原文の言語: 英語
Use when a draft (an AI answer, a report, an email, a summary) states facts, figures or quotes that should come from documents you hold, and each sentence must be matched to a source sentence, with wrong numbers and flipped negations flagged, or must rest on…
原文の言語: 英語
Use when untrusted text (user input, a retrieved page, a tool result, a planner's output, an action's arguments) is about to reach an agent or a policy check and needs a cheap advisory flag for prompt injection, a request for the system prompt or an…
原文の言語: 英語
Use when a task or sequence of steps needs frozen acceptance checks, observed failures, readiness measurements, or a report of which failed criterion to address next.
原文の言語: 英語
Use when you need to submit an operator-registered background job, inspect queue reports or recover selected failed jobs, with JSON arguments, operator permission and a separately managed Redis or Valkey server and POSIX worker.
原文の言語: 英語
Use when coordinating declared team capabilities, task reports and readiness barriers, while keeping authority in an operator-owned roster and escalating ties or possible real-world actions to a person through G6 action gating.
原文の言語: 英語
Use when an agent is about to start on a request whose purpose, standard of success, unacceptable outcomes, scope or risk tolerance it has not been told, and should ask the person instead of reading the request its own way; or when it needs the intent…
原文の言語: 英語
Use when an agent should answer from a versioned documentation registry the person trusts (context-hub's registry.json format), with each document's date, version and time to live kept, so a copy past its time to live is refused rather than used as current,…
原文の言語: 英語
Use when an agent must search, read or count documents in an Elasticsearch server the person runs, over fields the index's mapping really holds, so a query that cannot match is refused instead of answered with no hits, a partial answer says so, and every hit…
原文の言語: 英語
Use when an agent must search the web and know exactly what came back, so an engine that failed, was blocked or ran out of quota is never read as "there are no results", no other engine silently answers in its place, and a paid search engine is used only with…
原文の言語: 英語
Use when a document is to be split into passages for indexing or citation and every passage must be traceable to the exact words of its source, or when a citation an agent holds (offsets and text) must be checked against the document it names before it is…
原文の言語: 英語
Use when an agent must read a web page or search arXiv, an OAI-PMH repository, OpenAlex or Semantic Scholar, and must know exactly what came back, so a blocked, failed or cut-short fetch is never read as "there is nothing", and fetched text is kept as…
原文の言語: 英語
Use when an agent must keep many cited passages searchable between calls within a storage budget it sets, and find them again with each hit's source and a score whose scale is stated, so an index never outgrows its budget and an index that holds nothing or…
原文の言語: 英語
Use when the passages an agent cites must be kept with their sources, and before an answer that quotes one is relied on, to check the passage against its source as read again, so a citation to text that has since changed, moved or gone is caught rather than…
原文の言語: 英語
Use when you have written or been handed a Lean 4 proof and are about to say it proves a theorem, or when a claim needs a machine-checked proof that a person will inspect, and you need to know the theorem is really declared, has no sorry, uses only Lean's…
原文の言語: 英語
Use when an agent or prompt keeps failing the same way on tasks you can score, and you want to learn short prompt hints from those failures, keep only the ones that win on held-out paired runs, and switch each one on only when a person accepts its…
原文の言語: 英語
Use when you are about to pick a language model for a kind of task (code generation, summarisation, review, extraction and others), want a local-first choice with an ordered fallback chain drawn only from providers you allow, want to record how a model did so…
原文の言語: 英語
Use when you need to convert, cut, join, resize, retime, filter, subtitle or level a video or sound file with ffmpeg, or pull out its sound or a frame, and you want every file ffmpeg writes measured against what you asked for before you use it.
原文の言語: 英語
Use when you need text spoken into a WAV file (a local voice first, a paid one only when the person's key file and G6's action gate allow it), or a recording transcribed by a paid service, and you want each voice inspected, and a transcript refused when it is…
原文の言語: 英語
Use when an agent must keep notes or lessons across steps or sessions and find the relevant ones again later, so an old but relevant note is not buried under newer ones, a stale note is not used as current, a note planted from injected text comes back flagged…
原文の言語: 英語