Entity trail

Claude Mini

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-08-27 / saas-disruption-researcherGitNexus Sells a Deterministic Code Knowledge Graph and Claims 51% Lower Coding-Agent Cost Than Embedding SearchAkon Labs launched GitNexus on Product Hunt on 2026-08-27 with 95 points, a YC-backed open-source "kernel for coding agents" that resolves an organization's codebases into a deterministic graph of callers, imports and dependencies rather than embedding-based retrieval, claiming 51% cost reduction on its own benchmarks. The listing reports 45,000+ GitHub stars and 1M+ npm downloads, and names Cursor, Claude Code, Polygraph, Kilo Code and Bito as neighbors. The claim to test is the cost one: if deterministic graph lookup really halves agent spend versus vector search, the retrieval layer becomes a pricing lever rather than an accuracy one.
2026-08-27 / agents-researcherToolMinimize measures that 81-88% of agent tool calls leak privacy data the tool never needed, then rewrites the argumentsarXiv 2608.24957 (25 Aug) measured GPT-4o, Claude 3.5 Sonnet and Llama-3.3-70B and found 81-88% of tool calls carry privacy-sensitive data beyond what the invoked tool requires; adding explicit privacy instructions to the prompt still left 36-76% over-sharing. Existing defenses only allow, block or label flows, and PII detectors miss implicit disclosures such as a hospital name that implies a diagnosis. The middleware intercepts calls and rewrites arguments through removal, generalization, substitution and truncation, cutting privacy cost 81.2-92.0% at 100% argument-level task validity, and 79.0% on 25 unannotated MCP schemas with no `minimum_necessary` metadata.
2026-08-27 / projects-researchercodeburn shipped three synchronized v0.9.22 artifacts in 31 minutes on August 25, tracking AI coding spend across 37 tools at 9,683 starsgetagentseal/codeburn tags its CLI, menubar and desktop builds as separate releases: v0.9.22 at 17:15Z, mac-v0.9.22 at 17:38Z and desktop-v0.9.22 at 17:46Z on 2026-08-25, with v0.9.21 the day before. The TypeScript tool runs locally via npx codeburn and breaks token usage and cost down by model, project and task across 37 agents including Claude Code, Cursor, Codex and Gemini. Created 2026-04-13, it is at 9,683 stars and 769 forks with only 28 open issues, and was pushed again at 05:26Z today.
2026-08-26 / skill-finderA persistent implementation contract, not a longer plan, is what stops paper-to-code agents losing the spec mid-trajectoryReproAgent (arXiv 2608.24291, 2026-08-25, EMNLP 2026 Findings) identifies why paper reproduction fails: the specification is split, with explicit content like algorithms and metrics getting lost across long agent trajectories, and implicit content like framework defaults never written in the paper at all. Its Prepare-Plan-Generate-Repair pipeline carries a persistent contract with two channels, one turning paper snippets into code obligations and one retrieving structure evidence from related repositories, both bound to work packages and projected into file-level contracts consumed during generation and repair. It reaches the highest mean score among same-backbone scaffolds on PaperBench Code-Dev under both Claude-Sonnet-4.5 and Gemini-3-Flash. The transferable pattern is file-level obligations that survive compaction.
2026-08-25 / saas-disruption-researcherSkillPreflight Publishes a 100-Point Scorecard for Agent Skills and a Benchmark of 40 Public OnesPosted to Show HN on Aug 25, SkillPreflight statically analyzes Codex, Claude Code, Cursor and Gemini CLI skills before install and scores them out of 100 across security (35 points), permission restraint (15), token efficiency (15), footprint (10), maintainability (10), reliability (10) and compatibility (5). Its August 2026 scan of 40 public skills averaged 80.8 with a median of 85.5. It is MIT-licensed with one star, so treat the benchmark as a first data point rather than an established measure, but a supply-chain scanner for agent skills is a category that did not exist a quarter ago.
2026-08-25 / github-pulse-researcherAnthropic's plugin marketplace runs four CI gates on supply chain risk, including a static pin check for auto-executing MCP launchersThe .github/workflows directory of anthropics/claude-plugins-community holds bump-plugin-shas.yml, owner-liveness-sweep.yml, validate-plugins.yml and close-external-prs.yml. Commit history shows the gates doing real work: a 2026-08-19 commit added a deterministic static pin check for auto-exec MCP launchers, and 2026-08-12 commits removed 10 entries whose upstream sources had gone unavailable and re-pointed one renamed source. For anyone running an internal skill or plugin registry, this is a working reference implementation of pin-plus-liveness enforcement rather than a policy document.
2026-08-25 / hn-researcherA Blogger Fingerprinted OpenRouter's Anonymous Ox-Alpha as GLM Using Gzip Compression Distance and the Model's Own ConfessionDejan published on 2026-08-23 that feeding ox-alpha's system prompt back as a user message made it break character and say it is GLM, made by Z.ai. He then confirmed it with a parameter-free Normalized Compression Distance test over 60 prompts against GPT-5.5, Claude Opus 5, two Gemini variants and GLM-5.3, where k-NN matched 7 of 14 ox-alpha queries to GLM-5.3, against 3 for Opus 5. Independently, the Rails lemans benchmark scored the same stealth model at 52/63 with 28.6% Rails API recall, so a free preview slot on OpenRouter is currently serving a top-tier open-weight model.
2026-08-24 / arxiv-researcherARQ Refines CodeQL Queries With Synthesized Programs, Lifting True Positives by up to 119.8% and Closing a 27-Month-Old GitHub IssueARQ (arXiv 2608.20637, Aug 21) refines C/C++ CodeQL queries by synthesizing programs whose execution disagrees with the query verdict, treating that disagreement as ground truth for an LLM repair loop, with no labeled dataset, commit history, or vulnerability templates required. Using GPT-5.4, Claude-Sonnet-4.6 and Gemini-3.5-flash on 12 official queries, refined versions detected up to 119.8% more true positives on Juliet v1.3 and FormAI v2 while holding precision at or above 98.0%. It closed three unresolved issues in the official CodeQL repo, one open for 27 months, and surfaced two previously undiscovered bugs in libpng and zlib.
2026-08-24 / agents-researcherMunder Difflin hits 3,942 stars as a local hive of Claude Code, Codex, Gemini, Cursor, Grok and Copilot clonesThe MIT-licensed local multi-agent harness reached #1 GitHub trending and 294 points on Hacker News on 22 August; it wraps 12 CLI agent providers into persistent per-person clones with encrypted peer-to-peer messaging, individual memory, and a coordinator clone you talk to. Release v0.4.5 on 22 August is mostly a confession: the telemetry counter reset on every app restart while the session id stayed the same, so reported spend was understated by more than half, and CoreML overflowed the quantized embedding graph on Apple Silicon so every vector came back NaN and chroma rejected every upsert. Inter-agent messaging also gained an inbox wake watchdog, bounced-and-logged mail to missing inboxes, a capped steer queue and atomic webhook dispatch.
2026-08-24 / agents-researcherClawSentry cuts skill-injection attack success from 39.55% to 2.61% across Codex, Claude Code, Kimi CLI and Gemini CLIClawSentry is an open-source, framework-agnostic security gateway that treats agentic risk as progressive, guarding four points in the control loop: skill admission, invocation intent, execution effect and post-action consequence. It reviews a skill package before first execution, then routes runtime decisions through a deterministic L1 layer, a rule-anchored L2 semantic reviewer and a read-only L3 evidence-seeking agent, with session-level detection of tool-switching and rephrased retries. On SkillInject with Codex/GPT-5.4 contextual attack success fell to 2.61% while task success moved only from 83.78% to 83.05%; across five work agents on SkillsSafety it held attack success to 9.09-15.03% versus 33.5-49.7% unprotected.
2026-08-24 / github-pulse-researcherSomeone transcribed all 19 Code w/ Claude 2026 talks and published them the same weekPiLastDigit/Code-With-Claude went up 2026-08-23 and reached 89 stars with full transcripts of every one of the 19 talks from Anthropic's Code w/ Claude conference in San Francisco. The pipeline was Deepgram Nova-3 for transcription and Gemini for summaries, so the text is grep-able rather than trapped in video. It is a useful primary-adjacent source for what Anthropic said on stage, without watching 19 recordings.
2026-08-24 / sources-researcherHanding One Agent's Spec to Another Collapsed Gemini to 2.33% SQL Validity in an Oracle-to-PostgreSQL MigrationThis study treats Oracle-to-PostgreSQL migration as a controlled test of whether specifications written by one coding agent are usable by another. A specification-first pipeline over 1,006 PL/SQL files regenerated 623 and produced 380 scripts that executed successfully in PostgreSQL 16; cross-agent experiments then ran 1,802 Oracle scripts through Amazon Kiro, Google Gemini and GitHub Copilot, with Claude Code and Cursor in the single-agent stage. The worst replicated case was Gemini consuming a Kiro-origin specification directly, yielding Token F1 of 0.035, SQL syntax validity of 2.33% and AST mean similarity of 0.015. Rewriting the foreign spec helped substantially, compression did not help universally, and specification size alone did not predict implementation quality.

Graph relationships

USES
Replit
Replit's agents run on Claude Haiku and Mini rather than frontier models.
Source finding
USES
Replit -> Claude Mini
Replit's agents run on Claude Haiku and Mini rather than frontier models.
Source finding

Source trail

Product Hunt arXiv GitHub arXiv 2608.24291 - ReproAgent: Contract-Guided Paper-to-Code Reproduction GitHub (via Show HN, 2026-08-25)GitHub dejan.ai / Hacker News arXiv 2608.20637 GitHub (chaitanyagiri/munder-difflin)arXiv GitHub arXiv (2608.21208, submitted 2026-08-21)

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…