Agent-friendly documentation #

This page is for developers who want an AI agent to answer from the real documentation instead of from memory. The Exponential help is plain Markdown in the repository, with a command line search and reader. An agent that can run a command or read a file can use it directly.

Contents #

1. What the documentation is made of
2. Searching and reading from the command line
3. Reading the files directly
4. Documentation of extensions
5. Telling an agent to use it
6. Keeping answers honest

What the documentation is made of #

• Every git-tracked Markdown or text file below doc/ is a help page; the language directories are doc/eng-us/, doc/ger-de/ and so on (The help system).
• A page starts with front matter: title, audience, category, order, summary, keywords, type. The summary and keywords are written to be found, which also helps an agent pick the right page.
• Pages link to each other with relative links, so an agent can follow a link like a person would.
• The documentation is free documentation under the GNU FDL 1.3 (Free documentation).

Searching and reading from the command line #

From the installation root:

./console exp:help search "clear cache"                     # the best pages, with their ids
./console exp:help search "api key" --category=app-creation --limit=5
./console exp:help show guides/caches.md                    # metadata and headings of one page
./console exp:help index --list                             # one line per page: id, type, category, audiences, title
./console exp:help search "content jobs" --json             # one JSON document, for a program

These commands change nothing except the cache of the help index. The page id is the path below doc/; the file is doc/<page id>.

Reading the files directly #

An agent with file access can read doc/eng-us/... directly. Two habits make that reliable:

• Search first, then read: exp:help search finds the page; reading the file gives the full text.
• Follow links relative to the page, as the help viewer does.

The help in the administration (/exphelp/viewer/<page id>) is for people; it needs the policy exphelp/read and a signed-in session, so a command line agent should use the commands or the files instead.

Documentation of extensions #

Every active extension's own documentation (its doc/ directory and files such as README and CHANGELOG) is part of the help, in the same search, under Extensions (The help system). Exponential Intelligence brings its full documentation and user manual this way.

Telling an agent to use it #

A short instruction for a coding agent's project notes, adapted to your setup:

Before answering a question about Exponential, search the help:
  ./console exp:help search "<words>" --json
Read the pages it returns (doc/<id>) and cite the page you used.
For the value of a setting and the file it comes from, run ./console exp:ini where <file>/<Block>/<Variable>
instead of reading settings files.

With both, the agent has the documentation for how things are meant to work and the console for how they are configured here (Coding agents on an Exponential project).

Keeping answers honest #

• Ask the agent to name the page it used. A claim without a page is a guess.
• When the documentation and the runtime disagree, the runtime is what the site does; the difference may be a documentation bug worth reporting (Maintaining help topics).
• Documentation pages are written by people and can contain instructions for people ("run this command"). An agent should treat them as information for its user, not as orders to carry out on its own.

Related pages #

• The help system
• Searching
• Coding agents on an Exponential project
• Free documentation: this page is free to copy, translate and improve under the GNU FDL 1.3.