# Agent Harness Resources All links were checked live on 2026-09-30. Claude Code was v2.1.284 locally. ## Knowledge ### Claude Code (official docs, code.claude.com) - [Claude Code 101 (Anthropic Academy)](https://academy.claude.com/courses/claude-code-101) Four modules: the agentic loop, first prompt, daily workflows, customising. Use for: the pre-work before lesson 1. - [How Claude Code works](https://code.claude.com/docs/en/how-claude-code-works) The loop (gather context, act, verify), tools, the context window, and how subagents start fresh. Use for: lesson 1. - [Context window](https://code.claude.com/docs/en/context-window) The "what survives compaction" table. Use for: lesson 1's case that prose rules drift. - [Permission modes](https://code.claude.com/docs/en/permission-modes) The default (Manual), acceptEdits, plan, auto, dontAsk and bypassPermissions modes, and plan mode. Use for: lessons 1 and 4. - [Best practices](https://code.claude.com/docs/en/best-practices) Explore, then plan, then code, then commit; keeping CLAUDE.md short; `/clear`. Use for: lesson 4. - [Memory: CLAUDE.md, AGENTS.md, imports, rules](https://code.claude.com/docs/en/memory) Native AGENTS.md (v2.1.277+), load order, 4-hop imports, `paths:` rules, the under-200-lines target. Use for: lesson 2. - [Settings](https://code.claude.com/docs/en/settings) and the [settings reference](https://code.claude.com/docs/en/settings-reference) Precedence, and the `attribution` key. Use for: lesson 7. - [Permissions](https://code.claude.com/docs/en/permissions) The allow/ask/deny syntax (`Bash(git commit *)`), evaluated deny, then ask, then allow. Use for: lesson 7. - [Skills](https://code.claude.com/docs/en/skills) The SKILL.md frontmatter and `disable-model-invocation`; commands merged into skills. Use for: lesson 5. - [Subagents](https://code.claude.com/docs/en/sub-agents) `.claude/agents/*.md`, the `tools` allowlist, fresh context, explicit invocation. Use for: lesson 6. - [Hooks guide](https://code.claude.com/docs/en/hooks-guide) and the [hooks reference](https://code.claude.com/docs/en/hooks) PreToolUse deny JSON, exit code 2, `${CLAUDE_PROJECT_DIR}`. Use for: lesson 7. - [MCP](https://code.claude.com/docs/en/mcp) `claude mcp add`, the local/project/user scopes, `.mcp.json`, the approval prompt. Use for: lesson 8. - [Output styles](https://code.claude.com/docs/en/output-styles) The built-in Learning style (`TODO(human)`), and output style vs CLAUDE.md, skill or hook. Use for: lessons 1 and 2. - [Common workflows](https://code.claude.com/docs/en/common-workflows) Plan mode, resuming sessions, worktrees. Use for: lessons 4 and 10. - [Changelog](https://code.claude.com/docs/en/changelog) Use for: re-checking every `version-specific` badge. ### Tool-neutral and other agents - [AGENTS.md standard](https://agents.md/) "A README for agents", stewarded by the Agentic AI Foundation (Linux Foundation); the nearest file wins. Use for: lesson 2. - [Codex: AGENTS.md discovery](https://learn.chatgpt.com/docs/agent-configuration/agents-md) The global and override files, root-to-cwd concatenation, the 32 KiB default. Use for: lesson 2's Codex panel. - [Codex: config basics](https://learn.chatgpt.com/docs/config-file/config-basic), [approvals and security](https://learn.chatgpt.com/docs/agent-approvals-security), [rules](https://learn.chatgpt.com/docs/agent-configuration/rules) `approval_policy` (`untrusted` retired), `sandbox_mode`, `prefix_rule` allow/prompt/forbidden. Use for: lessons 1 and 7. - [Codex: skills](https://learn.chatgpt.com/docs/build-skills), [hooks](https://learn.chatgpt.com/docs/hooks), [subagents](https://learn.chatgpt.com/docs/agent-configuration/subagents), [MCP](https://learn.chatgpt.com/docs/extend/mcp?surface=cli) `.agents/skills`, `.codex/hooks.json`, `.codex/agents/*.toml`, `[mcp_servers.x]`. Use for: the Codex panels in lessons 5 to 8. - [Antigravity: rules](https://antigravity.google/docs/rules), [skills](https://antigravity.google/docs/skills), [workflows](https://antigravity.google/docs/ide/workflows), [permissions](https://antigravity.google/docs/permissions), [hooks](https://antigravity.google/docs/hooks), [MCP](https://antigravity.google/docs/mcp), [subagents](https://antigravity.google/docs/subagents) AGENTS.md is read as always_on; `.agents/rules`, `.agents/skills`, `.agents/hooks.json`, `.agents/mcp_config.json`; workflows are deprecated by 2026-11-01. Use for: the Antigravity panels. ### .NET and guard tooling - [EF Core InMemory provider](https://learn.microsoft.com/en-us/ef/core/providers/in-memory/) and [Testing without the database](https://learn.microsoft.com/en-us/ef/core/testing/testing-without-the-database) InMemory is "strongly discouraged" for testing and has no ExecuteUpdate. Use for: lesson 3's InMemory trap. - [EF Core SQLite limitations](https://learn.microsoft.com/en-us/ef/core/providers/sqlite/limitations) DateTimeOffset ordering and comparison are evaluated on the client. Use for: lesson 8. - [SQLite transactions](https://www.sqlite.org/lang_transaction.html) One writer at a time, and BEGIN IMMEDIATE. Use for: lessons 6 and 9 (the stock invariant ADR). - [dotnet format](https://learn.microsoft.com/en-us/dotnet/core/tools/dotnet-format) `--verify-no-changes` and `--include`. Use for: lesson 5's quality gate and lesson 7's format hook. - [MSBuild command line (-warnAsError)](https://learn.microsoft.com/en-us/visualstudio/msbuild/msbuild-command-line-reference) Use for: lesson 5. - [gitleaks](https://github.com/gitleaks/gitleaks) `gitleaks git --pre-commit --staged` (v8.30.1; `protect` is deprecated). Use for: lesson 7. - [githooks and core.hooksPath](https://git-scm.com/docs/githooks) Hooks must be executable, and `core.hooksPath` has to be set in every clone. Use for: lesson 7. - [Conventional Commits 1.0.0](https://www.conventionalcommits.org/en/v1.0.0/) Use for: lesson 5's commit skill and lesson 7's commit-msg hook. - [context7](https://github.com/upstash/context7) Live library docs over MCP (`@upstash/context7-mcp` 4.1.1, or the remote `https://mcp.context7.com/mcp`). Use for: lesson 8. ## Wisdom (communities) - [anthropics/claude-code on GitHub](https://github.com/anthropics/claude-code) Issues and changelog, maintained by Anthropic. Use for: checking whether a behaviour is a known bug before building around it. - [OpenAI Developer Community: Codex](https://community.openai.com/) Use for: Codex config and AGENTS.md behaviour. - [Google AI Developers Forum](https://discuss.ai.google.dev/) Use for: Antigravity rules, skills and permissions. - Internal: a weekly 30-minute harness review at Osmosys, where trainees swap repos and run each other's swap test. Use for: real feedback on whether a harness steers someone else's agent. ## Gaps - No official source for Codex's commit-attribution default, or for Codex's session file path. Test on the installed version. - No maintained, official read-only SQLite MCP server; the reference server is archived. Lesson 8 uses a SQLite connection opened with `mode=ro` instead.