Guides

llms.txt, llms-full.txt, and skill.md

Machine-readable docs surfaces designed for AI agents. Drop into a system prompt, a project rules file, or fetch programmatically.

HoneyCoin docs publish three top-level files plus a directory of manifests built for AI consumption. Use them when you want an agent to plan or build an integration without scraping HTML.

The three top-level files

FilePurposeWhen to use
/llms.txtConcise sitemap of the docs. Lists every page with its title and one-line description, plus links to manifests.As a search index. Drop into a system prompt so the agent knows what pages exist.
/llms-full.txtFull prose dump of every documentation page concatenated.When the agent needs detail and can't fetch URLs. Paste once at the start of a session.
/skill.mdPortable "skill" — the rules, the manifests, and the recipes summarized in one file optimized for paste-into-prompt use.Use as a Cursor .cursorrules, a Claude Project knowledge file, or a system prompt. ~10–15kb.
📘

Publishing

These three files are emitted automatically by the docs build. The published URLs will be https://docs.honeycoin.app/llms.txt, /llms-full.txt, and /skill.md once v1.0.4 deploys.

/llms.txt

A short text file that names every documentation page with its slug and a one-line description. Agents use it to plan which pages to fetch.

Example:

# HoneyCoin Docs

> Build money movement across Africa and emerging markets.

## Start
- /getting-started: Get started with HoneyCoin in five minutes
- /authentication: Public Key + API Key → bearer token

## Reference
- /reference/charge-mobile-money: POST /api/b2b/initiate-mobile-money-charge
- /reference/bank-payouts: POST /api/b2b/payouts
...

/llms-full.txt

The same sitemap, but each entry is followed by the full Markdown of the page. Useful for offline reasoning or when the agent's tools cannot fetch URLs.

/skill.md

A standalone skill file. The format is opinionated — designed to be pasted into a single prompt or stored as project rules.

Contents:

  1. Identity (what HoneyCoin does).
  2. Hard rules (from agent-instructions).
  3. Coverage summary (compact view of the corridor matrix).
  4. Wallet support summary.
  5. Error response shape + how to branch on errorCode.
  6. Webhook envelope + processing pattern.
  7. Five worked example recipes.
  8. Links to manifests for further detail.

Manifests

The /manifests/ directory contains the canonical machine-readable data. Pages on this site render from these files; agents should treat them as the single source of truth.

ManifestPurpose
manifests/coverage.jsonEvery supported country, currency, collection method, payout destination, operator, and per-corridor min/max amount.
manifests/wallets.jsonEvery supported chain with address type, regex, token support, onramp/offramp eligibility, and the ownership model.
manifests/errors.jsonCanonical response shape plus the full catalogue of API errors, transaction errors, and webhook errors with retryable flags and developer actions.
manifests/webhooks.jsonWebhook envelope, signature rules, delivery and retry policy, and every event with dataFields, statusValues, and recipe cross-links.
manifests/wallet-webhooks.jsonThe same for wallet webhooks plus the resend endpoint.
manifests/agent-policy.yamlThe list of rules every AI agent must follow when generating HoneyCoin integrations.
manifests/flows.jsonFlow definitions (collect-mobile-money, send-payout, onramp, offramp, wallet-receive) with step-by-step decision points.
manifests/test-data.jsonSandbox fixtures for success, failure, OTP, and redirect paths per recipe.
manifests/changelog.jsonStructured changelog with version, date, breaking, added, changed, deprecated, fixed arrays.

Each manifest carries a version (ISO date). Pin to a version if you want reproducible behavior. (JSON Schemas for these manifests are planned for a future release.)

See Machine-readable manifests for the full inventory with download links.

Using manifests in code

The manifests are public — fetch directly:

curl https://docs.honeycoin.app/manifests/coverage.json
curl https://docs.honeycoin.app/manifests/wallets.json
curl https://docs.honeycoin.app/manifests/errors.json

A typical agent pattern:

import json
import httpx

# Load once at agent startup.
coverage = httpx.get("https://docs.honeycoin.app/manifests/coverage.json").json()
wallets = httpx.get("https://docs.honeycoin.app/manifests/wallets.json").json()

# Lookup before recommending.
def supports(country_code: str, method: str) -> bool:
    for c in coverage["countries"]:
        if c["countryCode"] == country_code:
            return any(m["method"] == method for m in c["collections"])
    return False

Refresh cadence

SurfaceRefresh
/llms.txt, /llms-full.txt, /skill.mdRebuilt on every docs deploy.
manifests/*.jsonUpdated when HoneyCoin coverage or behavior changes; version date in the file.

For programmatic use in production, cache for at most 24 hours and surface the version date in your logs.

Related