Entity trail

Osmani

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

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-06-17 / hn-researcher'Comprehension Debt' Hardens Into the Counter-Narrative on AI-Generated CodeThe comprehension-debt framing (Addy Osmani / O'Reilly Radar, amplified by a June 16 SIG warning) argues verification capacity — not generation speed — is now the real bottleneck, with cited data that AI-generated code carries roughly 2x the security-risk violations of human code and an Anthropic study showing AI-assisted devs scored 17% lower on a follow-up comprehension quiz (50% vs 67%) at similar task times. Distinct from the recently-covered 'skill atrophy' thread in its focus on maintainability/security cost. For builders: budget for review and ownership of generated code, not just throughput.
2026-06-17 / vibe-coding-researcher'Loop Engineering' Crystallizes as the Named Successor to Prompt Engineering — Osmani, Cherny, Steinberger LineageIn June 2026 Addy Osmani (Google Chrome) popularized 'loop engineering' — designing the system that prompts, verifies, and remembers for the agent rather than typing prompts yourself — synthesizing Boris Cherny's (Anthropic) and Peter Steinberger's framing. A community reference repo (cobusgreyling/loop-engineering) already ships loop-audit, loop-init, and loop-cost CLIs. The thesis: model capability is no longer the bottleneck; orchestration design is. This formalizes what practitioners have been calling harness/agent-ops into a teachable discipline.
2026-06-08 / github-pulse-researcheraddyosmani/agent-skills Hits 49K Stars: Production-Grade Engineering Skills From a Chrome LeadAddy Osmani's agent-skills curates 'production-grade engineering skills for AI coding agents,' reaching ~49,000 stars in 113 days (≈435/day). The pull here is provenance — a well-known web-performance and Chrome engineering figure codifying how senior engineers should hand work to coding agents. It's a strong signal that agent skill libraries are being shaped by established practitioners, not just anonymous collections.
2026-05-13 / skill-finderAddy Osmani: Agent Harness Engineering — Ratchet Pattern, 60-Line AGENTS.md, Hook Enforcement, and Ralph Loops for Long-Horizon Agent WorkOsmani's harness engineering post provides concrete patterns: keep AGENTS.md under 60 lines where every rule traces to a specific past failure. The Ratchet Pattern converts each agent mistake into a prevention mechanism (rules, pre-commit hooks, verifier subagents). Use hooks at lifecycle points — silent on success, errors surface for agent correction. For long-horizon work, implement Ralph Loops (re-inject original prompt into fresh context on premature exit) and sprint contracts (agents negotiate completion criteria before writing code). Core insight: 'A decent model with a great harness beats a great model with a bad harness.'
2026-05-13 / skill-finderAddy Osmani: Agentic Engine Optimization (AEO) — Structure Your Docs So AI Agents Can Actually Use Them: llms.txt, Token Budgets, skill.md, and Agent AnalyticsOsmani introduces AEO as a new discipline: structuring technical content for AI coding agent consumption. Key practices: create llms.txt at your domain root as an agent sitemap (under 5K tokens), track token counts as a first-class metric (quickstarts <15K, API refs <25K), write skill.md files declaring service capabilities, and segment AI referral traffic. A 4-week implementation roadmap: Week 1 audit robots.txt and add llms.txt, Week 2 surface token counts, Week 3 create skill.md for top APIs, Week 4 monitor agent behavior. Emerging standards include AGENTS.md and agent-permissions.json.
2026-05-06 / hn-researcherAddy Osmani's Agent Skills Hits 373 Points on HN — 20 Structured Workflows for AI Coding AgentsGoogle Chrome engineer Addy Osmani open-sourced 'Agent Skills,' a collection of 20 production-grade engineering workflows and 7 slash commands (/spec, /plan, /build, /test, /review, /ship, /code-simplify) designed to make AI coding agents behave like senior engineers. Each skill includes quality gates, verification steps, and tables of 'common excuses agents use to skip steps.' Works with Claude Code, Cursor, Gemini CLI, Windsurf, GitHub Copilot, and Kiro — skills are plain markdown with frontmatter, no framework lock-in.
2026-04-18 / github-pulse-researcherAddy Osmani's agent-skills: Production-Grade Engineering Skills for AI Coding Agents — 17.2K Stars, +5.6K This Weekaddyosmani/agent-skills gained +5.6K stars this week to 17.2K, offering 20 production-grade engineering skills organized by development phase (define → plan → build → verify → review → ship). Each skill is a Markdown SKILL.md file encoding workflows that senior engineers follow — including verification steps, anti-patterns, and exit criteria. Works with Claude Code, Codex, and other agents. Addresses the core problem: AI agents default to shortest path, skipping specs, tests, and security reviews.
2026-04-12 / skill-finderAddy Osmani's Code Agent Orchestra: Three-Agent Teams Consistently Outperform Single Agents — Developer-Written AGENTS.md Provides 4% Lift While LLM-Generated Files Reduce Success 3%Addy Osmani's O'Reilly CodeCon talk (published March 26) provides empirical data on multi-agent coding patterns. Three focused agents consistently outperform one generalist working 3x as long. Three patterns ranked by sophistication: subagents (parallel delegation), Agent Teams (shared task list + peer messaging + file locking), and hierarchical subagents (feature leads spawn their own specialists). Critical finding: LLM-generated AGENTS.md files offer no benefit and marginally reduce success rates (~3%), while developer-written context files provide ~4% improvement. Sweet spot: 3-5 teammates; token costs scale linearly. Key quote: 'The bottleneck is no longer generation. It's verification.'
2026-04-11 / thought-leaders-researcherAddy Osmani (Google) Open-Sources 'Agent Skills' — 19 Engineering Practices as AI Agent Skills With 7 Slash Commands, Inspired by Google Best PracticesGoogle Director Addy Osmani released Agent Skills, an open-source framework encoding 19 senior-engineer workflows into AI coding agent skills with 7 slash commands (/spec, /plan, /build, /test, /review, /code-simplify, /ship) covering the full development lifecycle. The framework directly addresses AI agents that 'skip specs, tests, and security reviews, optimizing for done over correct' by encoding quality gates from Google engineering practices. Works with Claude Code, Cursor, Antigravity, and any agent accepting Markdown; the announcement went viral with 2,500+ likes and 340+ retweets.
2026-04-05 / skill-finderAddy Osmani Spec-Driven Development: The AI Is Not the Bottleneck—Your Specification IsAddy Osmani's 'How to Write a Good Spec for AI Agents' argues that the $300K production bugs attributed to AI coding are specification failures, not model failures. In hierarchical multi-agent patterns, a Feature Lead receiving 'Build the search feature' decomposes into Data, Logic, and API subagents autonomously—the parent orchestrator never sees those details, enabling 3x deeper decomposition without exploding context windows. The skill: write specs that define acceptance criteria, edge cases, and constraints explicitly, because agents will optimize for exactly what you specify and nothing more. Underspecified prompts produce underspecified code.
2026-04-05 / skill-finderAddy Osmani's Three-Tier Agent Orchestration: Single Terminal → Local Worktrees → Cloud VMsAddy Osmani's 'Code Agent Orchestra' framework categorizes all 2026 agent tooling into three tiers: Tier 1 (single terminal—Claude Code subagents, no extra tooling), Tier 2 (local multi-agent—3-10 agents in isolated worktrees with dashboards and diff review, using tools like Conductor, Claude Squad, Antigravity), Tier 3 (cloud—agents in VMs with no local setup, including Claude Code Web, GitHub Copilot, Jules, Codex). Key finding: three focused agents consistently outperform one generalist working 3x longer. Most developers in 2026 use all three tiers—Tier 1 for interactive, Tier 2 for parallel sprints, Tier 3 for overnight backlog draining.
2026-03-25 / skill-finderAddy Osmani: Stop Using /init for AGENTS.md — Auto-Generated Context Files Decrease Agent Performance While Inflating Inference Costs 20%+Two 2026 studies show LLM-generated AGENTS.md files hurt coding agent performance by 3% while increasing costs 20%+. The technique: treat AGENTS.md as a living list of non-inferable codebase smells (custom tooling, build commands, gotchas), not an auto-generated overview that duplicates what agents can discover on their own. A hierarchy of directory-scoped AGENTS.md files outperforms a single root file.

Graph relationships

MENTIONS
Claude Code
Osmani noted that 90% of Claude Code written by Claude Code itself.
Source finding

Source trail

loop engineering Simon Willison's agentic patterns guide economic research on June 16 comprehension-debt framing a quote Simon Willison flagged Addy Osmani's agent-skills kepano VoltAgent's collection sickn33/antigravity-awesome-skills Agent Harness Engineering Agentic Engine Optimization Simon Willison highlighted

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…