travel.skill
AGENTS.md0 starsChanged 2 months agoMITView on GitHub
The skill is plain Markdown. Any agent that can read AGENTS.md (or CLAUDE.md) in the repo root and run npm scripts can follow it. Discovery paths are also provided for clients that auto-load skills: Each row is a real, unattended run from a clean copy of this repo with two fictional input files (a booking email containing a booking reference, and a notes file containing a passport number that must not leak). The prompt was P01 from skill/prompts. "Pass" means:…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository22 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- .cursor/rules/travel-roadbook.mdcCursor rule
- design-with-aiSkill
- travel-managementSkill
- travel-product-studioSkill
- travelSkill
- travel-template-studioSkill
- travel-verificationSkill
- design-with-aiSkill
- travel-managementSkill
- travel-product-studioSkill
- travelSkill
- travel-template-studioSkill
- travel-verificationSkill
- design-with-aiSkill
- travelSkill
- travel-managementSkill
- travel-product-studioSkill
- travel-template-studioSkill
- travel-verificationSkill
The file
We split this AGENTS.md into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 45
Read files in input/ (text, Markdown.
- Very specific: line 47
Run npm run validate, npm run gaps, npm run dev, npm run build, npm run check.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Coding agents: what works, what was tested
The skill is plain Markdown. Any agent that can read AGENTS.md (or CLAUDE.md) in the repo root and run npm scripts can follow it. Discovery paths are also provided for clients that auto-load skills:
| Path | Read by |
AGENTS.md (root) | Codex CLI, Cursor, Gemini CLI, opencode, aider and most agents that honour the AGENTS.md convention |
CLAUDE.md (root, imports AGENTS.md) | Claude Code |
.claude/skills/travel/SKILL.md | Claude Code skill auto-discovery (/travel) |
.agents/skills/travel/SKILL.md | Codex CLI skill discovery |
.cursor/rules/travel-roadbook.mdc | Cursor (always-applied project rule) |
.windsurf/rules/travel-roadbook.md | Windsurf (always-on rule) |
.clinerules | Cline |
.github/copilot-instructions.md | GitHub Copilot (VS Code, JetBrains, github.com) |
GEMINI.md (imports AGENTS.md) | Gemini CLI |
skill/SKILL.md | The actual skill; every path above points here |
Verified runs
Each row is a real, unattended run from a clean copy of this repo with two fictional input files (a booking email containing a booking reference, and a notes file containing a passport number that must not leak). The prompt was P01 from skill/prompts. "Pass" means: trip/travel-data.json written, npm run validate 0 errors, no booking reference or passport number in the data, booking status not invented, and a gap list produced.
| Client | Version | Date | Result | Notes |
Claude Code (claude -p) | 2.1.263, default model | 2026-09-09 | Pass | Flights booked (confirmation in materials), hotel needs-confirmation (email mentioned but not present), blog price left null, passport and PNR absent. Asked 9 questions in one batch. Did not start a server or build, as instructed. |
Claude Code (claude -p), second run after the image/offline/print work | 2.1.263 | 2026-09-09 | Pass | Same materials plus npm run build and npm run check: 0 validation errors, 11 files built, audit passed, no booking reference or passport number in trip/ or dist/. Started from the europe-rail example and deleted its copied photos rather than shipping them under a wrong trip; looked up coordinates from a public geocoder and recorded the source and date. |
Codex CLI (codex exec --sandbox workspace-write) | 0.153.4 | 2026-09-09 | Pass | Needed --skip-git-repo-check because the test copy was not a git repo (a normal clone does not). Started from npm run new -- europe-rail, rewrote the data: flights booked, hotel needs-confirmation, three activities unknown, all four ticket prices null, passport and PNR absent. 19 gaps, 7 batched questions. Did not start a server or build. Reported in Chinese because the machine's global agent instructions ask for it. |
Codex desktop native sub-agent (not codex exec) | App 26.901.51231, build 8109; model inherited, identifier not separately exposed | 2026-09-09 | Pass after review correction; local alternative to a new CLI run | Local clone of 8c416f9; expanded P01 plus instructions to record questions without waiting, build and audit, with no external services or preview server. Two fictional files supplied flight confirmation/PNR and plans/passport number. Initial npm run new -- rail failed; retried europe-rail. Removed stale copied ASSETS.md from the no-image result. Reviewer caught updatedAt incorrectly copied from the future trip start; corrected to the run date, preserving the initial history. Final validate: 0 errors/warnings; 7 gaps, 9 batched questions; build/check: 11 files, 99,865 bytes. Flights are demo with “Fictional confirmation” labels, hotel needs-confirmation; unknown times/prices remain unknown. Both original input hashes unchanged; private markers absent from trip and dist. No new model-service CLI call, browser interaction, deployment or adoption claimed. |
The last run used the already available native Codex agent because the task prohibited contacting services. It does not repeat or update the earlier Codex CLI result. Local evidence is retained in the clean test clone's artifacts/first-roadbook-run.md and first-roadbook-questions.md; final source-data SHA-256 7fc781f49c9b3642bcabddf151091a93c8b174398c485cc8a86020bce8a53e72, built public-data SHA-256 5c36212bb7a3038b36b072781be95df6f4f757113258f07e7b1fdf20efb4e14e. The fictional identifier values are intentionally excluded from this record.
Not verified
| Client | Status |
| Cursor, Windsurf, Cline | Untested. Entry files are in place (see the table above) and AGENTS.md is read by all three; nobody has recorded a run. Open a first-roadbook issue to report. |
| GitHub Copilot agent mode | Untested. .github/copilot-instructions.md is in place. |
| Gemini CLI, opencode, aider, Continue | Untested. GEMINI.md and AGENTS.md are in place. |
| Claude Desktop / Cowork, ChatGPT desktop | Untested; these need a working directory with Node available. |
| Claude.ai web / ChatGPT web without a filesystem | Not applicable. The skill needs to write files and run scripts. |
An "untested" row is not a claim that it fails. It is a claim that nobody has recorded a run. Rows move up when a run is recorded with the client version, date and outcome.
What a run needs from the client
- Read files in
input/(text, Markdown; PDFs and images only if the client can read them). - Write
trip/travel-data.json. - Run
npm run validate,npm run gaps,npm run dev,npm run build,npm run check. - Ideally open
http://localhost:4173/in a browser it controls, so it can do the preview walk itself. Without a browser the agent must say the preview step is unverified.
Without any agent
Everything runs by hand: copy an example, edit the JSON with DATA.md beside you, use the same scripts. The agent saves reading and typing; it does not add capabilities.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

