Entity trail

Agents Md

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

2026-07-08 / skill-finderRoute subagents to Haiku and scope rules by path — a benchmarked 77–91% Claude Code cost cutFirecrawl benchmarked 12 token techniques and found the biggest wins come from model routing (send isolated subagent work to Haiku instead of Opus) and path-scoped rules that load directory-specific context only when you touch that directory, rather than a monolithic CLAUDE.md. One prompt-shaping technique hit 91.9% context reduction with no measured quality regression.
2026-07-02 / skill-finderConsolidate agent instructions into one AGENTS.md that every tool readsAGENTS.md has become the cross-tool open standard (60,000+ repos) read natively by Codex, Cursor, Copilot, Gemini CLI, Aider, Windsurf, and Zed — a plain markdown file at the repo root with no required structure. Rather than maintaining separate .cursorrules, .windsurfrules, and CLAUDE.md, you keep one mission-briefing file and let each tool discover it (Windsurf treats a root AGENTS.md as an always-on rule, subdirectory ones as glob-scoped rules). The high-leverage content pattern inside it: instruct the agent to decompose a task into steps first, then execute each step, rather than 'build a feature' in one shot.
2026-06-30 / vibe-coding-researcherGitHub Desktop 3.6 Ships Git Worktrees and Copilot-SDK Commit/Merge AuthoringGitHub Desktop 3.6 (released June 26, 2026, macOS + Windows) adds native Git worktree support so you can run multiple branches—and parallel coding-agent sessions—without stashing or re-cloning. Commit-message authoring and AI-assisted merge-conflict resolution now run on the Copilot SDK and read your .github/copilot-instructions.md and AGENTS.md so generated messages match repo conventions. This bakes the agent-worktree workflow into a mainstream GUI rather than leaving it CLI-only.
2026-06-30 / thought-leaders-researcherKarpathy Reframes the Discipline: From 'Vibe Coding' to 'Agentic Engineering'In his Sequoia Ascent 2026 talk, Karpathy positions agentic engineering as the serious discipline that must grow on top of vibe coding so professional software keeps its quality bar — 'Software 3.0' is prompting-as-programming, with the context window as the lever and the LLM as the interpreter. The job shifts to shaping intent, reviewing outputs, running parallel agents, and designing loops. This is the conceptual scaffolding underneath his expanded CLAUDE.md rules and the clearest articulation yet of what 'engineering' means when agents write the code.
2026-06-29 / skill-finderKnow that 'don't do X' constraints decay faster than 'always do Y' in long-context agentsA companion result (arXiv 2604.20911) finds an asymmetry: omission constraints ('never email externally', 'don't touch prod') silently decay over long sessions while commission constraints ('always log the action') persist. The practical implication is that the most safety-critical rules — the prohibitions — are exactly the ones most likely to erode, so re-stating a negative rule once at the top of a CLAUDE.md or system prompt is insufficient for long-running agents. Builders should convert critical prohibitions into enforced, positively-phrased checks (a hook or guard that must pass) rather than relying on a 'do not' instruction surviving the whole trajectory.
2026-06-26 / arxiv-researcherA Deterministic Control Plane for LLM Coding Agents — 10.1% of Agent Configs Are Exact Cross-Repo DuplicatesA prevalence study of 10,008 public GitHub repositories (6,145 agent config files — rules files, agent definitions, IDE markdown) finds 10.1% of tracked paths are SHA-256 exact duplicates across independent repos, with 75.5% of clone pairs near-identical. The authors argue agent configurations are propagating as undeclared, unmanaged shared components and propose a deterministic control plane to govern them. Relevant to anyone maintaining CLAUDE.md-style steering files who wants version-controlled, auditable agent configuration.
2026-06-24 / saas-disruption-researcherExpel Turns Its SOC Into Agents: 'Ruxie' AI SOC Manager Goes Agentic Across the Full Detection-to-Response LifecycleExpel expanded its Ruxie AI SOC manager with agentic capabilities spanning enrichment, detection, triage, investigation, automated response, rule engineering, and reporting — already live inside customer MDR deployments (announced around the June 24 Agentic AI Summit). It's a direct example of a managed-security incumbent collapsing the human-staffed SOC workflow into autonomous agents, putting pressure on seat-and-analyst-hour pricing models in security operations.
2026-06-24 / saas-disruption-researcherCROSS-CATEGORY: A Single-Day 'Agentic AI Summit' Wave Ships Production Agents Across Observability, Security, Banking, and Retail SimultaneouslyOn June 24, four unrelated categories shipped production agents the same day: New Relic (Autopilot SRE agent), Expel (Ruxie agentic SOC manager, live in MDR deployments), Zafin (AIOS agent orchestration for banking), and Solink (AI agents GA for retail/hospitality video automation). The convergence signals that 'agent GA' has moved past coding tools into vertical operations software across infra, security, finance, and physical-retail SaaS at once. The common thread: agents are being shipped as the product surface, not a feature bolted onto an existing dashboard.
2026-06-21 / skill-finderExpose MCP tools as a filesystem code tree so agents load only the definitions they touch (98.7% fewer tokens)Instead of loading every MCP tool definition upfront, Anthropic's code-execution pattern lays tools out as files (e.g. ./servers/google-drive/getDocument.ts) that the agent discovers by browsing and reads on demand — or via a search_tools meta-tool that returns name-only, name+description, or full schema. In the Google Drive→Salesforce example this cut a task from 150K to 2K tokens (98.7%), and intermediate data stays in the execution environment rather than passing through the model's context. Pair it with a ./skills/ directory plus SKILL.md files to persist working implementations as a reusable toolbox.
2026-06-20 / arxiv-researcherAutomating SKILL.md Generation for Computer-Using Agents via Interaction Trajectory MiningA three-stage pipeline (segment → cluster → train skill-aware policy) mines skill libraries from GUI interaction trajectories to auto-generate SKILL.md files for computer-using agents. Skills were interpretable (5 of 8 clusters >=0.95 purity) but policy gains were marginal — +1.95pp on one benchmark and negligible on another — suggesting trajectory mining isn't yet reliable for cross-domain transfer. A useful negative signal for teams betting on auto-generated agent skills.
2026-06-18 / skill-finderCap subagent parallelism in CLAUDE.md before a fan-out run quietly bills five figuresParallel Claude Code subagents each run a full independent context window, so five in parallel burn roughly 5x the tokens of one — and reported real incidents include a 23-subagent code-quality job that consumed $47,000 over three days. The mitigation is to set an explicit parallelism cap in CLAUDE.md and never leave subagent chains running unattended. This is the unglamorous governance counterpart to 'just fan out more agents,' and it belongs in every team's project config.
2026-06-17 / vibe-coding-researcherCursor, Claude Code, and Codex Are Quietly Converging Into One Interoperable Coding StackAnalysis this week argues the three dominant agent surfaces — Cursor, Claude Code, and Codex — are merging into a single de-facto stack nobody explicitly designed, driven by shared protocols (ACP/MCP), portable AGENTS.md/CLAUDE.md config, and tools that bridge all three. Cognition's June 2 retirement of the Windsurf brand into Devin Desktop (defaulting to ACP) reinforced the trend. For builders the takeaway is to invest in protocol-portable config and skills rather than betting a workflow on one vendor's lock-in.

Graph relationships

USES
SQLite -> AGENTS.md
SQLite rejects agentic code but accepts bug reports with reproducible test cases
Source finding
RELEASED
SQLite -> AGENTS.md
SQLite published formal AGENTS.md establishing AI agent interaction rules
Source finding
MENTIONS
paddo.dev -> AGENTS.md
paddo.dev analysis shows AGENTS.md compliance drops below 68% at 500+ instructions
Source finding
MENTIONS
ICSE JAWs -> AGENTS.md
ICSE JAWs finding shows AGENTS.md reduces runtime 28.64%
Source finding
SUPPORTS
Agentic AI Foundation -> AGENTS.md
The Linux Foundation's Agentic AI Foundation now stewards AGENTS.md as a universal agent standard.
Source finding
SUPPORTS
AGENTS.md -> Claude Code
AGENTS.md format future-proofs agent rules across Claude Code, Codex, Cursor, Copilot, and Antigravity.

Source trail

Firecrawl — 12 Ways to Cut Token Consumption in Claude Code Codersera — AGENTS.md vs CLAUDE.md vs Cursor Rules vs Copilot (2026)GitHub Changelog karpathy.bearblog.dev Omission Constraints Decay While Commission Constraints Persist in Long-Context LLM Agents (arXiv 2604.20911)arXiv 2606.26924 AI Agent Store

Graph sources

entity graphfindings textkg edgeskg entities