# Map > Every Agent Experience guide, ax-check rule and evidence record on one map, grouped round its topic. Canonical: https://agentexperience.tech/map/ The map draws every guide, ax-check rule and evidence record as a point round its topic, with a line from each page to the pages to read next. It needs JavaScript; in words, it is the list below. [Explore](https://agentexperience.tech/explore/) lists every page of the site, and the map's data is [explore-graph.json](https://agentexperience.tech/explore-graph.json). ## [Start here: what AX is](https://agentexperience.tech/insights/topics/foundations/) For anyone new to AX: what the term means, the framework this site uses, and what the people who do the work actually do. Guides: - [What is Agent Experience?](https://agentexperience.tech/insights/what-is-agent-experience/) - [An Agent Experience (AX) framework](https://agentexperience.tech/insights/agent-experience-framework/) - [Five meanings of agent experience](https://agentexperience.tech/insights/meanings-of-agent-experience/) - [What does an AX engineer do?](https://agentexperience.tech/insights/what-does-an-ax-engineer-do/) ## [Help agents find and understand your product](https://agentexperience.tech/insights/topics/discovery/) For teams whose website, docs or sign-up an AI agent has to find and read before it can help anyone. Guides: - [Build websites agents can understand](https://agentexperience.tech/insights/agent-friendly-websites/) - [Write agent-readable documentation](https://agentexperience.tech/insights/make-documentation-legible/) - [AX, UX, DX, and GEO: where each fits](https://agentexperience.tech/insights/ax-vs-ux-dx-geo/) - [Does llms.txt actually work?](https://agentexperience.tech/insights/state-of-llms-txt/) - [The state of agent-web standards](https://agentexperience.tech/insights/agent-web-standards/) - [Agentic Resource Discovery, implemented](https://agentexperience.tech/insights/agentic-resource-discovery/) - [Let agents start before anyone signs up](https://agentexperience.tech/insights/let-agents-start-before-sign-up/) Rules: - [SKILL.md frontmatter missing or invalid](https://agentexperience.tech/checks/axc-d022/) - [Skill name breaks the naming rules](https://agentexperience.tech/checks/axc-d023/) - [Bundled reference missing](https://agentexperience.tech/checks/axc-d025/) - [Help is not reachable](https://agentexperience.tech/checks/axc-c009/) - [Claimed URL is broken](https://agentexperience.tech/checks/axc-f001/) - [Claimed URL could not be verified](https://agentexperience.tech/checks/axc-f002/) - [Claimed npm package does not exist](https://agentexperience.tech/checks/axc-f003/) - [Claimed PyPI package does not exist](https://agentexperience.tech/checks/axc-f004/) - [Claimed package version does not exist](https://agentexperience.tech/checks/axc-f005/) - [Claimed MCP tool is not served](https://agentexperience.tech/checks/axc-f006/) - [Served MCP tool is not mentioned](https://agentexperience.tech/checks/axc-f007/) - [Claimed CLI command does not exist](https://agentexperience.tech/checks/axc-f008/) - [Claimed CLI flag does not exist](https://agentexperience.tech/checks/axc-f009/) - [Linked local file is missing](https://agentexperience.tech/checks/axc-f010/) - [Claims extracted but not verified](https://agentexperience.tech/checks/axc-f011/) Evidence: - [Coding agents mostly read instruction files, not docs sites](https://agentexperience.tech/evidence/ev-0011/) - [Compact documentation did not help when the source was present](https://agentexperience.tech/evidence/ev-0012/) - [FAQ blocks and structured data showed no citation effect within a domain](https://agentexperience.tech/evidence/ev-0013/) - [A capable agent skipped the index and guessed the page](https://agentexperience.tech/evidence/ev-0027/) - [A warning that names the failing plan redirected agents; a tip did not](https://agentexperience.tech/evidence/ev-0029/) - [Adding the context7 MCP server gave no lift; its tools went unused](https://agentexperience.tech/evidence/ev-0030/) - [Vendor claim: agents never invoked a docs skill in 56% of eval cases](https://agentexperience.tech/evidence/ev-0032/) - [Agent skills recommended a package that does not exist](https://agentexperience.tech/evidence/ev-0040/) ## [Make tools easier to choose and use](https://agentexperience.tech/insights/topics/tools/) For teams that give agents tools, an API or written instructions, and want the agent to pick the right one and use it well. Guides: - [Write tool descriptions an agent can act on](https://agentexperience.tech/insights/tool-descriptions/) - [Agent tool catalogues need a pruning strategy](https://agentexperience.tech/insights/agent-tool-catalogs/) - [An agent dashboard should show the next decision](https://agentexperience.tech/insights/show-the-next-decision/) - [How to expose a 3,000-operation API to agents](https://agentexperience.tech/insights/large-api-surfaces-for-agents/) - [Keep agent instructions true](https://agentexperience.tech/insights/keep-agent-instructions-true/) Rules: - [Missing description](https://agentexperience.tech/checks/axc-d001/) - [Placeholder description](https://agentexperience.tech/checks/axc-d002/) - [Description only restates the name](https://agentexperience.tech/checks/axc-d003/) - [Does not say when to use it](https://agentexperience.tech/checks/axc-d004/) - [Does not say when not to use it](https://agentexperience.tech/checks/axc-d005/) - [No neighbouring tool named](https://agentexperience.tech/checks/axc-d006/) - [Closed vocabulary hinted in prose, not declared as an enum](https://agentexperience.tech/checks/axc-d007/) - [Parameter has no description](https://agentexperience.tech/checks/axc-d008/) - [Near-duplicate descriptions](https://agentexperience.tech/checks/axc-d013/) - [Duplicate name](https://agentexperience.tech/checks/axc-d014/) - [Description too short](https://agentexperience.tech/checks/axc-d015/) - [Description length outlier](https://agentexperience.tech/checks/axc-d016/) - [Context footprint estimate](https://agentexperience.tech/checks/axc-d017/) - [Required property not defined](https://agentexperience.tech/checks/axc-d018/) - [Example does not match the schema](https://agentexperience.tech/checks/axc-d019/) - [Promotional language in a description](https://agentexperience.tech/checks/axc-d020/) - [Many visible parameters](https://agentexperience.tech/checks/axc-d021/) - [Skill description over 1,024 characters](https://agentexperience.tech/checks/axc-d024/) - [SKILL.md over 500 lines](https://agentexperience.tech/checks/axc-d026/) Evidence: - [An enum in the schema ends silent failures from example-only vocabularies](https://agentexperience.tech/evidence/ev-0001/) - [Tools as code match or beat JSON tool calls for most models](https://agentexperience.tech/evidence/ev-0009/) - [Injected skills lowered pass rates and raised token cost on average](https://agentexperience.tech/evidence/ev-0017/) - [Skill rules that name a command or path change what agents do](https://agentexperience.tech/evidence/ev-0018/) - [Skill selection precision collapses as the skill pool grows](https://agentexperience.tech/evidence/ev-0019/) - [Praise and list order move tool selection](https://agentexperience.tech/evidence/ev-0020/) - [Search-and-execute meta-tools cut catalogue tokens by 99% in production](https://agentexperience.tech/evidence/ev-0022/) - [Hiding tools is not enforcing permissions](https://agentexperience.tech/evidence/ev-0023/) - [Launch post and documentation disagree on agent output format](https://agentexperience.tech/evidence/ev-0039/) ## [Handle approvals, failures, and handoffs](https://agentexperience.tech/insights/topics/workflows/) For teams whose agents take real actions that a person has to approve, recover from, or pick up halfway. Guides: - [Agent workflow design: scoping the first one](https://agentexperience.tech/insights/start-with-workflows/) - [Agent context handoff: carry the reason for the work](https://agentexperience.tech/insights/preserve-context-through-handoffs/) - [Human approval workflows for AI agents](https://agentexperience.tech/insights/approval-is-a-workflow/) - [Agent error recovery: design the path back](https://agentexperience.tech/insights/design-for-recovery/) - [Record retries as product history](https://agentexperience.tech/insights/retries-are-product-history/) - [Approval, handoff and recovery patterns for AI agents](https://agentexperience.tech/insights/approval-handoff-recovery-patterns/) Rules: - [Side-effect annotation contradicts the description](https://agentexperience.tech/checks/axc-d009/) - [No side-effect annotations](https://agentexperience.tech/checks/axc-d010/) - [openWorldHint false for a tool that reaches outside](https://agentexperience.tech/checks/axc-d011/) - [Object schema does not set additionalProperties to false](https://agentexperience.tech/checks/axc-d012/) - [Command hangs without a terminal](https://agentexperience.tech/checks/axc-c001/) - [Error exits with status 0](https://agentexperience.tech/checks/axc-c002/) - [Error text not on stderr](https://agentexperience.tech/checks/axc-c003/) - [Error does not name a next step](https://agentexperience.tech/checks/axc-c004/) - [Unknown command gives no suggestion](https://agentexperience.tech/checks/axc-c005/) - [No machine-readable output flag](https://agentexperience.tech/checks/axc-c006/) - [Documented JSON output does not parse](https://agentexperience.tech/checks/axc-c007/) - [No dry-run flag documented](https://agentexperience.tech/checks/axc-c008/) - [ANSI colour codes without a terminal](https://agentexperience.tech/checks/axc-c010/) - [NO_COLOR not honoured](https://agentexperience.tech/checks/axc-c011/) - [Version output is not plain](https://agentexperience.tech/checks/axc-c012/) - [Advertised JSON output not verified](https://agentexperience.tech/checks/axc-c013/) Evidence: - [Constraints stated only in prose produce silent failures on live APIs](https://agentexperience.tech/evidence/ev-0002/) - [Error text that names the next tool lifts recovery](https://agentexperience.tech/evidence/ev-0003/) - [Naming recovery tools is the active ingredient in failure receipts](https://agentexperience.tech/evidence/ev-0004/) - [Idempotency keys cut duplicate writes from 28% to 4%](https://agentexperience.tech/evidence/ev-0005/) - [An evidence contract cuts false-success reports after tool failures](https://agentexperience.tech/evidence/ev-0006/) - [Browser agents declare success on most of their failures](https://agentexperience.tech/evidence/ev-0007/) - [Allow/ask/never policies blocked less overreach than per-action approval](https://agentexperience.tech/evidence/ev-0014/) - [Approvals that outlive their task raise attack success](https://agentexperience.tech/evidence/ev-0015/) - [Approval records omit the effects a command goes on to trigger](https://agentexperience.tech/evidence/ev-0016/) - [Prompt injection split across tool channels evades defences](https://agentexperience.tech/evidence/ev-0026/) - [Vendor claim: users approved about 93% of permission prompts](https://agentexperience.tech/evidence/ev-0034/) - [A CLI exits 0 when a destructive command is refused](https://agentexperience.tech/evidence/ev-0036/) - [Destructive and bulk commands that report success but do nothing](https://agentexperience.tech/evidence/ev-0037/) - [Dry-run output printed secrets until a fix redacted them](https://agentexperience.tech/evidence/ev-0038/) ## [Find evidence for what to improve next](https://agentexperience.tech/insights/topics/evaluation/) For teams that need to check how ready one job is, measure a change, and decide what to fix first. Guides: - [The Open Agent-Readiness Rubric](https://agentexperience.tech/insights/agent-readiness-rubric/) - [Evaluate the decisions in an agent workflow](https://agentexperience.tech/insights/evaluating-agent-workflows/) - [How do you measure agent experience?](https://agentexperience.tech/insights/measure-agent-experience/) - [Run an Agent Experience review](https://agentexperience.tech/insights/run-an-ax-review/) - [Agent evaluation loop: run, judge, triage, fix, repeat](https://agentexperience.tech/insights/build-a-reliable-agent-evaluation-loop/) - [Agent readiness scores compared](https://agentexperience.tech/insights/agent-readiness-scores-compared/) - [Agent experience examples you can check](https://agentexperience.tech/insights/agent-experience-examples/) Evidence: - [Declared completion and logged completion differ by up to 41 points](https://agentexperience.tech/evidence/ev-0008/) - [Agent scaffolding, not MCP versus CLI, drove cost](https://agentexperience.tech/evidence/ev-0010/) - [Agents leave available tools unused: the adoption gap](https://agentexperience.tech/evidence/ev-0021/) - [Harnesses alter shell calls and the wrong action runs silently](https://agentexperience.tech/evidence/ev-0024/) - [Run-to-run variance dominates, and a verification tool beats a verification prompt](https://agentexperience.tech/evidence/ev-0025/) - [A JSON input mode for a CLI raised agent cost 4x to 11x](https://agentexperience.tech/evidence/ev-0028/) - [A model with cheaper tokens cost 3.7x more per run](https://agentexperience.tech/evidence/ev-0031/) - [Vendor claim: agents were 48% of Wrangler CLI use](https://agentexperience.tech/evidence/ev-0033/) - [Vendor claim: 70% of new Supabase databases are created by agents or AI tools](https://agentexperience.tech/evidence/ev-0035/)