Results for “max-iterations”

FREE — God-tier long-context memory for AI agents. Injects 500K-1M clean tokens, auto-summarizes with tone/intent preservation, compresses 14-turn history into 800 tokens.

FREE — God-tier long-context memory for AI agents. Injects 500K-1M clean tokens, auto-summarizes with tone/intent preservation, compresses 14-turn history into 800 tokens.

Enhances AI agent memory by injecting large external context, auto-summarizing conversations with tone and intent preservation, compressing multi-turn histories, and verifying facts.

Push the LLM to reconsider, refine, and improve its recent output. Use when user asks for deeper critique or mentions a known deeper critique method, e.g. socratic, first principles, pre-mortem, red team.

FREE — God-tier long-context memory for AI agents. Injects 500K-1M clean tokens, auto-summarizes with tone/intent preservation, compresses 14-turn history into 800 tokens.

Stateful single-mission improvement loop with strict evaluator contract, markdown decision logs, and max-runtime stop behavior

FREE — God-tier long-context memory for AI agents. Injects 500K-1M clean tokens, auto-summarizes with tone/intent preservation, compresses 14-turn history into 800 tokens.

当面对新任务或目标，因过度思考“如何开始”或“需要多久”而陷入拖延、焦虑，迟迟无法行动时

Implement iterative evaluation and refinement loops for AI agent outputs, using self-critique, evaluator-optimizer patterns, and rubric-based scoring to improve quality.

Creates, optimizes, and iteratively refines agent prompts, system prompts, developer prompts, and reusable prompt templates with evals.

Optimizes hot-path scalar loops in .NET 8+ with cross-platform Vector128/Vector256/Vector512 SIMD intrinsics, or replaces manual math loops with single TensorPrimitives API calls.

Configure fault tolerance, straggler detection, preemption, in-process restart, and re-run state machine for Megatron Bridge training jobs.

Cleans up and improves existing code without changing behavior.

Push the LLM to reconsider, refine, and improve its recent output. Use when user asks for deeper critique or mentions a known deeper critique method, e.g. socratic, first principles, pre-mortem, red team.

Push the LLM to reconsider, refine, and improve its recent output. Use when user asks for deeper critique or mentions a known deeper critique method, e.g. socratic, first principles, pre-mortem, red team.

Injects up to 1 million tokens of external context into AI agent memory, auto-summarizes conversations with tone and intent preservation, and compresses multi-turn history into dense token sequences.

Autonomously runs iterative experiment loops to optimize code against a measurable metric. Use when the user wants to improve execution time, memory usage, test pass rate, or any numeric performance goal across repeated experiments — NOT for one-shot bug fixes or simple code review.

Write idiomatic Elixir code with OTP patterns, supervision trees, and Phoenix LiveView. Masters concurrency, fault tolerance, and distributed systems. Use PROACTIVELY for Elixir refactoring, OTP design, or complex BEAM optimizations.

Strip AI-generated tells from a draft and rewrite it in Heath's personal voice (GTM Juice). Fire whenever the user asks to "humanize", "de-AI", "clean up", "rewrite in my voice", "make this sound like me", "kill the AI tells", or "polish this draft". Also fires as the mandatory final pass inside juice-content-dispatch — every LinkedIn post, newsletter, and video script drafted there must run through this skill before the Notion write. Applies the 29-pattern Wikipedia "Signs of AI writing" catalog (em dashes, significance inflation, promotional language, rule of three, synonym cycling, inline-header lists, sycophantic tone, hyphenated word pairs, persuasive-authority tropes, signposting, etc.), then applies a hardcoded Heath-voice calibration: self-implicating first, failure-derived, plain declaratives over bumper-sticker aphorisms, no bow-tied endings, never Mixmax corporate brand voice. Forked from blader/humanizer (MIT).

Guides cost-effective image generation by iterating at low resolution before producing final high-resolution output.

Run metric-driven iterative optimization loops -- define a measurable goal, run parallel experiments, measure each against hard gates or LLM-as-judge scores, keep improvements, and converge on the best solution. Use when optimizing clustering quality, search relevance, build performance, prompt quality, or any measurable outcome that benefits from systematic experimentation.

明楼（谍战虚构）认知与表达框架（压缩蒸馏）：三面间谍、辞令层叠、亲情作人质 触发：伪装者 等。禁止间谍违法教程

Generate and rank research ideas given a broad direction. Use when user says "找idea", "brainstorm ideas", "generate research ideas", "what can we work on", or wants to explore a research area for publishable directions.

Write idiomatic Elixir code with OTP patterns, supervision trees, and Phoenix LiveView. Masters concurrency, fault tolerance, and distributed systems.

Multi-tiered memory management (HOT/WARM/COLD) for context compaction. Invoke ONLY for explicit compaction events: post-`/compact` cleanup, MEMORY.md tier promotion, archive batch, or "trim my context". NOT for general recall (use deep-recall) or routine memory writes (use storage-router). Triggers: "compact memory", "promote to durable", "archive old context", "tier this".

Sharpens and stress-tests an existing plan file through multiple codebase-grounded review passes, editing it in place and recording every finding and iteration in cross-referenced companion files. Use this skill whenever the user wants to iterate on, refine, tighten, or improve a plan. Also use it when the user asks to verify, validate, or confirm feasibility of an approach. Does not implement plan steps, write test plans, review code, or investigate bugs, and does not generate new plans from scratch — use plan-a-feature for a new plan. Runs its review rounds to completion without pausing between them; to review each round as it lands, use pairing.

Elite brainstorming specialist for facilitated ideation sessions. Use when the user asks to talk to Carson or requests the Brainstorming Specialist.

Mines a codebase for evidence-backed improvement opportunities, forcing two escalation gates to surface breakthrough ideas, and outputs a ranked portfolio with a plan seed without implementing.

Expand an annotated outline into full article prose using brand voice anchored in example articles, hitting the outline's per-section depth targets. Triggered after /product-mentions.

Language-specific super-code guidelines for elixir.

Debug and optimize AI agents by analyzing reasoning traces, context degradation, tool confusion, instruction drift, repeated task failures, and performance regressions.

Forces a second, harder escalation pass using the original glazer prompt words verbatim, pushing for a sharper replacement thesis rather than more polish.

Runs NCBI BLAST+ 2.17.0 sequence searches from the command line: makeblastdb (with -parse_seqids), blastn/blastp/blastx/tblastn with tabular -outfmt 6/7 for parsing, correct -task choice (megablast vs blastn vs blastn-short), -taxids/-negative_taxids taxonomic scoping, and -mt_mode multithreading; plus a DIAMOND blastp --ultra-sensitive path for large protein searches. Warns that -max_target_seqs is a heuristic keep-count, not a top-N best-hits filter. Use when the user wants command-line BLAST, makeblastdb, a local BLAST database, blastn/blastp/blastx/tblastn searches, or DIAMOND protein search. For the Bio.Blast web NCBIWWW API prefer alterlab-biopython; for quick one-liner database lookups prefer alterlab-gget. Part of the AlterLab Academic Skills suite.

Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies. Use when training models >1B parameters, need maximum GPU efficiency (47% MFU on H100), or require tensor/pipeline/sequence/context/expert parallelism. Production-ready framework used for Nemotron, LLaMA, DeepSeek.

Run a single experiment iteration. Edit the target file, evaluate, keep or discard. Use when the user runs /ar:run or asks for one manual autoresearch iteration.